kev-0.6b: decision-v7 recipe (transfer 0.620 dev / 0.642 locked)
Browse files- README.md +15 -15
- adapter_model.safetensors +1 -1
- head.pt +2 -2
- provenance.json +25 -18
- result.json +1746 -695
- train.log +316 -274
- training_config.json +16 -5
- training_metrics.json +7 -7
README.md
CHANGED
|
@@ -33,22 +33,22 @@ model-index:
|
|
| 33 |
- task: { type: text-classification, name: typed decision (choice / noul / score) }
|
| 34 |
dataset: { type: mixed, name: "decision-v4 development (1,204 records; ten trained public sources + programmatic policy pairs)" }
|
| 35 |
metrics:
|
| 36 |
-
- { type: accuracy, value: 0.
|
| 37 |
-
- { type: expected_calibration_error, value: 0.
|
| 38 |
- task: { type: text-classification, name: typed decision, out-of-domain }
|
| 39 |
dataset: { type: mixed, name: "transfer-v4 development (764 records; six never-trained sources + held-out policy structures)" }
|
| 40 |
metrics:
|
| 41 |
-
- { type: accuracy, value: 0.
|
| 42 |
-
- { type: brier_score, value: 0.
|
| 43 |
---
|
| 44 |
|
| 45 |
# kev-0.6b
|
| 46 |
|
| 47 |
`kev-0.6b` is a **decision model**: one document (the *state*) and a set of typed questions in, a probability distribution per question out, in one forward pass. No text generation. It is a LoRA adapter (r=16) plus a pointer head on `Qwen/Qwen3-0.6B-Base`, and it serves TypeSafe's public `/v1/systemone` contract.
|
| 48 |
|
| 49 |
-
**The small member of the kev family.** It is the best 0.6B checkpoint under a frozen, checksummed evaluation protocol (three seeds, eight one-knob mutations
|
| 50 |
|
| 51 |
-
- Hub: `jaredpalmer/kev-0.6b` (this repo; trial `
|
| 52 |
- Code, suites, results, and the full research log: [github.com/jaredpalmer/kev](https://github.com/jaredpalmer/kev) — see `PLAN.md`, `runs/leaderboard.md`, and `evals/v4/*/manifest.json`
|
| 53 |
|
| 54 |
## What changed since kev-0.5b
|
|
@@ -56,22 +56,22 @@ model-index:
|
|
| 56 |
| | kev-0.5b | kev-0.6b (this) |
|
| 57 |
|---|---|---|
|
| 58 |
| backbone | Qwen2.5-0.5B | Qwen3-0.6B-Base |
|
| 59 |
-
| training records | 9,000 (six sources) |
|
| 60 |
| none-of-the-above | augmentation fix only | + minimal pairs: same state rendered with the true option present and removed |
|
| 61 |
-
| in-distribution accuracy (decision-v4 dev) | 0.712 | **0.
|
| 62 |
-
| out-of-domain accuracy (transfer-v4 dev) | 0.575 | 0.
|
| 63 |
-
| none-option present, accuracy | 0.25 (transfer-v1) | 0.
|
| 64 |
-
| seeds behind the number | 1 | 3 (transfer 0.
|
| 65 |
|
| 66 |
-
Jev (`typesafe-ai/jev` via Vercel AI Gateway) on the same frozen development sets: **0.845** in-distribution, **0.857** out-of-domain. Per-source transfer accuracy for this checkpoint: QNLI 0.85, SciQ 0.
|
| 67 |
|
| 68 |
## Known limits
|
| 69 |
|
| 70 |
- **Out of domain it is a 0.6B model.** Transfer accuracy is flat at ~0.60 across every hyperparameter we tried (eight one-knob mutations, three seeds). The same recipe at 4B reaches 0.72–0.75 and at 8B 0.74–0.77; capacity, not data, is the bottleneck at this size.
|
| 71 |
- **Held-out policy reasoning fails**: on programmatic policy pairs whose rule structure was never trained, both-siblings-correct is 6–11% (kev-4b 0.73, Jev 0.86).
|
| 72 |
- **Ordinal hedging**: on 3-level Score questions with date arithmetic it collapses to the middle level.
|
| 73 |
-
- Confident-error rate out of domain is
|
| 74 |
-
- **Locked test, read once** (`runs/locked/kev-06b-
|
| 75 |
|
| 76 |
## Architecture
|
| 77 |
|
|
@@ -79,7 +79,7 @@ Prefill-only causal LM with a block-causal attention mask: a shared state prefix
|
|
| 79 |
|
| 80 |
## Training
|
| 81 |
|
| 82 |
-
Frozen suite `evals/
|
| 83 |
|
| 84 |
## Evaluation protocol
|
| 85 |
|
|
|
|
| 33 |
- task: { type: text-classification, name: typed decision (choice / noul / score) }
|
| 34 |
dataset: { type: mixed, name: "decision-v4 development (1,204 records; ten trained public sources + programmatic policy pairs)" }
|
| 35 |
metrics:
|
| 36 |
+
- { type: accuracy, value: 0.801 }
|
| 37 |
+
- { type: expected_calibration_error, value: 0.086, name: "ECE, raw probabilities" }
|
| 38 |
- task: { type: text-classification, name: typed decision, out-of-domain }
|
| 39 |
dataset: { type: mixed, name: "transfer-v4 development (764 records; six never-trained sources + held-out policy structures)" }
|
| 40 |
metrics:
|
| 41 |
+
- { type: accuracy, value: 0.620 }
|
| 42 |
+
- { type: brier_score, value: 0.536 }
|
| 43 |
---
|
| 44 |
|
| 45 |
# kev-0.6b
|
| 46 |
|
| 47 |
`kev-0.6b` is a **decision model**: one document (the *state*) and a set of typed questions in, a probability distribution per question out, in one forward pass. No text generation. It is a LoRA adapter (r=16) plus a pointer head on `Qwen/Qwen3-0.6B-Base`, and it serves TypeSafe's public `/v1/systemone` contract.
|
| 48 |
|
| 49 |
+
**The small member of the kev family.** It is the best 0.6B checkpoint under a frozen, checksummed evaluation protocol: the 4B/8B recipe's data (`decision-v7`) at lr 1e-4, three seeds (transfer 0.613 / 0.605 / **0.620**), after eight one-knob mutations and three seeds of the previous data found nothing better than 0.61. Out of domain it is a 0.6B model — use `kev-4b` for accuracy; use this one where memory or latency rule the 4B out, and measure on your own data.
|
| 50 |
|
| 51 |
+
- Hub: `jaredpalmer/kev-0.6b` (this repo; trial `v7-06b/02-trial-2`, seed 2 of 3)
|
| 52 |
- Code, suites, results, and the full research log: [github.com/jaredpalmer/kev](https://github.com/jaredpalmer/kev) — see `PLAN.md`, `runs/leaderboard.md`, and `evals/v4/*/manifest.json`
|
| 53 |
|
| 54 |
## What changed since kev-0.5b
|
|
|
|
| 56 |
| | kev-0.5b | kev-0.6b (this) |
|
| 57 |
|---|---|---|
|
| 58 |
| backbone | Qwen2.5-0.5B | Qwen3-0.6B-Base |
|
| 59 |
+
| training records | 9,000 (six sources) | 12,576 (ten public sources + 896 policy minimal pairs + 1,680 records from 60 random rule structures) |
|
| 60 |
| none-of-the-above | augmentation fix only | + minimal pairs: same state rendered with the true option present and removed |
|
| 61 |
+
| in-distribution accuracy (decision-v4 dev) | 0.712 | **0.801** |
|
| 62 |
+
| out-of-domain accuracy (transfer-v4 dev) | 0.575 | **0.620** |
|
| 63 |
+
| none-option present, accuracy | 0.25 (transfer-v1) | 0.80 |
|
| 64 |
+
| seeds behind the number | 1 | 3 (transfer 0.605–0.620) |
|
| 65 |
|
| 66 |
+
Jev (`typesafe-ai/jev` via Vercel AI Gateway) on the same frozen development sets: **0.845** in-distribution, **0.857** out-of-domain. Per-source transfer accuracy for this checkpoint: QNLI 0.85, SciQ 0.93, TweetEval-offensive 0.69, PAWS 0.59, Emotion 0.49, MMLU 0.50; held-out policy structures near chance.
|
| 67 |
|
| 68 |
## Known limits
|
| 69 |
|
| 70 |
- **Out of domain it is a 0.6B model.** Transfer accuracy is flat at ~0.60 across every hyperparameter we tried (eight one-knob mutations, three seeds). The same recipe at 4B reaches 0.72–0.75 and at 8B 0.74–0.77; capacity, not data, is the bottleneck at this size.
|
| 71 |
- **Held-out policy reasoning fails**: on programmatic policy pairs whose rule structure was never trained, both-siblings-correct is 6–11% (kev-4b 0.73, Jev 0.86).
|
| 72 |
- **Ordinal hedging**: on 3-level Score questions with date arithmetic it collapses to the middle level.
|
| 73 |
+
- Confident-error rate out of domain is 11% (≥0.9 confidence and wrong); raw ECE 0.09 in-domain, 0.15 out of domain. Probabilities are usable in-domain; treat them as advisory elsewhere.
|
| 74 |
+
- **Locked test, read once** (`runs/locked/kev-06b-v7-ungated/`): in-distribution accuracy **0.808** (Brier 0.266, ECE 0.089), out-of-domain **0.642** (Brier 0.483, ECE 0.128, confident errors 7.9%). This partition will not be read again for this checkpoint.
|
| 75 |
|
| 76 |
## Architecture
|
| 77 |
|
|
|
|
| 79 |
|
| 80 |
## Training
|
| 81 |
|
| 82 |
+
Frozen suite `evals/v7/decision-v7` (manifest pins dataset and base-model revisions): 10,000 public records (1,000 per source), 896 policy minimal-pair records over nine template families, and 1,680 records from 60 randomly generated rule structures, two epochs, LoRA r=16 on attention and MLP projections at lr 1e-4, pointer head from scratch, cross-entropy on the option distribution, bf16 autocast with fp32 master weights on one H100 (~12 min). Augmentation: option permutation, none-of-the-above insertion, distractors, and none minimal pairs on 25% of Choice records. No Jev outputs were used for training.
|
| 83 |
|
| 84 |
## Evaluation protocol
|
| 85 |
|
adapter_model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 40419816
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:deaab63b61f95d628503831e8b336e1c3e334b9a42d3530d979a5149b429c492
|
| 3 |
size 40419816
|
head.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ce6cd9ffc54db41c179b65a33d60973dc8280c29e886218eb3ec07dc28d6f28b
|
| 3 |
+
size 2102399
|
provenance.json
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
{
|
| 2 |
"config": {
|
| 3 |
"epochs": 2,
|
| 4 |
-
"seed":
|
| 5 |
-
"lr": 0.
|
| 6 |
"lora": 16,
|
| 7 |
"accum": 1,
|
| 8 |
"batch": 8,
|
|
@@ -13,34 +13,41 @@
|
|
| 13 |
"p_none_distract": 0.12,
|
| 14 |
"p_distract": 0.15,
|
| 15 |
"p_none_pair": 0.25,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
"base": "Qwen/Qwen3-0.6B-Base",
|
| 17 |
"dtype": "bf16"
|
| 18 |
},
|
| 19 |
-
"config_sha256": "
|
| 20 |
-
"suite_sha256": "
|
| 21 |
"source_hashes": {
|
| 22 |
"kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
|
|
|
|
| 23 |
"kev/api.py": "cdb0602684d798ccd3fc2c86be622f2a8e7bcdeee64d07f95604c7d3fd4701ec",
|
| 24 |
-
"kev/
|
|
|
|
| 25 |
"kev/compare.py": "bd0445f021de59e35c7bff9304e39dd2e1211e3877a453575594ae7b81b0ada4",
|
| 26 |
-
"kev/composition.py": "
|
| 27 |
-
"kev/contrastive.py": "
|
| 28 |
-
"kev/data.py": "
|
| 29 |
-
"kev/evaluate.py": "
|
| 30 |
-
"kev/experiment.py": "
|
| 31 |
"kev/jev.py": "e0213782359ba2f95adbf045ddaf0a008b08b4162bf6a9b66d4ce55fc91a51cb",
|
| 32 |
-
"kev/model.py": "
|
| 33 |
"kev/plot.py": "d689c7dd18cde7f9ea77cb4f55c50ff7da1880b21cb2ecf244348e45842a1382",
|
| 34 |
-
"kev/publish.py": "
|
| 35 |
-
"kev/serve.py": "
|
| 36 |
-
"kev/study_v3.py": "
|
| 37 |
-
"kev/suite.py": "
|
| 38 |
-
"kev/train.py": "
|
| 39 |
-
"modal_app.py": "
|
| 40 |
"pyproject.toml": "ce566ee11b68c0cb587bb591943641f403d761d820817693dae7590faacf9888",
|
| 41 |
"uv.lock": "846fdc64a8bd6279a17a4365f18cd4e92ba9622d50d378aafcb5be6604296138"
|
| 42 |
},
|
| 43 |
-
"git_commit": "
|
| 44 |
"platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.36",
|
| 45 |
"torch": "2.8.0+cu128",
|
| 46 |
"device": "cuda",
|
|
|
|
| 1 |
{
|
| 2 |
"config": {
|
| 3 |
"epochs": 2,
|
| 4 |
+
"seed": 2,
|
| 5 |
+
"lr": 0.0001,
|
| 6 |
"lora": 16,
|
| 7 |
"accum": 1,
|
| 8 |
"batch": 8,
|
|
|
|
| 13 |
"p_none_distract": 0.12,
|
| 14 |
"p_distract": 0.15,
|
| 15 |
"p_none_pair": 0.25,
|
| 16 |
+
"synthetic_repeat": 1,
|
| 17 |
+
"public_frac": 1.0,
|
| 18 |
+
"head_lr": 0.0,
|
| 19 |
+
"weight_decay": 0.01,
|
| 20 |
+
"anchor_w": 0.0,
|
| 21 |
"base": "Qwen/Qwen3-0.6B-Base",
|
| 22 |
"dtype": "bf16"
|
| 23 |
},
|
| 24 |
+
"config_sha256": "e0d2ec20a46cf1cfa71568e1c695fe1f1ae16650f0ec6808df3ca00af3cdc3df",
|
| 25 |
+
"suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2",
|
| 26 |
"source_hashes": {
|
| 27 |
"kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
|
| 28 |
+
"kev/anchors.py": "089d8a5493502bb26f540eb1c5e681780ca0bb01276073d4e1733211f15d0e10",
|
| 29 |
"kev/api.py": "cdb0602684d798ccd3fc2c86be622f2a8e7bcdeee64d07f95604c7d3fd4701ec",
|
| 30 |
+
"kev/autoresearch.py": "0a8aa6b57c1c9ba93d25b2cf631b03aed2e686e374c1148002eec48765b7b1fc",
|
| 31 |
+
"kev/benchmark.py": "4dccec28be6becf6c3dcc8c186e8e1908a6482f15f602a60b440cc9cb4ac8a59",
|
| 32 |
"kev/compare.py": "bd0445f021de59e35c7bff9304e39dd2e1211e3877a453575594ae7b81b0ada4",
|
| 33 |
+
"kev/composition.py": "f335ed17e18e0a544893db5e22b9a059e6ce1b2e14dbb863ac7d7bbf8f3e0536",
|
| 34 |
+
"kev/contrastive.py": "cbb979aa5d40265ad0e64695f94b281d91751fa405811ddfa8212fede111edcf",
|
| 35 |
+
"kev/data.py": "9729f8f497b02aca542915d6f1bea3ff55956629d7d9610370aea91ee66efd23",
|
| 36 |
+
"kev/evaluate.py": "2b15a093ae9085a2d3999d5f57cb6293ce93d101bcddd0ea252089538e8f3a07",
|
| 37 |
+
"kev/experiment.py": "c635394fe56c9aa11765f8bbf8f133dcf3e57cbe8f2bce4ebdc7b4e85fa536ff",
|
| 38 |
"kev/jev.py": "e0213782359ba2f95adbf045ddaf0a008b08b4162bf6a9b66d4ce55fc91a51cb",
|
| 39 |
+
"kev/model.py": "03f87dc4c081fa8dc1f31c192de8015ddb8213b235d5807b66aebd75b50cb407",
|
| 40 |
"kev/plot.py": "d689c7dd18cde7f9ea77cb4f55c50ff7da1880b21cb2ecf244348e45842a1382",
|
| 41 |
+
"kev/publish.py": "b88c710140e7717ba42d539fdbe785507ac8b9d56b673641ade61c4136bb7dc9",
|
| 42 |
+
"kev/serve.py": "570e995aa98b51dbbf867694276d29fc9b9f3f1e62f22bea0116ae51d6497003",
|
| 43 |
+
"kev/study_v3.py": "9fc44d44dad09a4f1ee29a7bcb2eb3c7aa373d09186d0e9533666b69ec95c401",
|
| 44 |
+
"kev/suite.py": "f221e5a50111bab24ada035ff68ea3c3067482ff1edda48b8d4ece769043a561",
|
| 45 |
+
"kev/train.py": "e0df6503506d6a0cb5ccd35ae38bb072a6103383adf7beab4269bb1a1f5fcdce",
|
| 46 |
+
"modal_app.py": "fcc64f410bdd456bf5514b70da4e57da1f3a2f89c70918df0a6e7003141f33b8",
|
| 47 |
"pyproject.toml": "ce566ee11b68c0cb587bb591943641f403d761d820817693dae7590faacf9888",
|
| 48 |
"uv.lock": "846fdc64a8bd6279a17a4365f18cd4e92ba9622d50d378aafcb5be6604296138"
|
| 49 |
},
|
| 50 |
+
"git_commit": "37cb2f6155afec1054de2cc78dfd7c7dd24a956d",
|
| 51 |
"platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.36",
|
| 52 |
"torch": "2.8.0+cu128",
|
| 53 |
"device": "cuda",
|
result.json
CHANGED
|
@@ -1,594 +1,1044 @@
|
|
| 1 |
{
|
| 2 |
-
"objective": -0.
|
| 3 |
"paired_flip": {
|
| 4 |
"pairs": 80,
|
| 5 |
"flip_rate": 0.7,
|
| 6 |
-
"both_correct_rate": 0.
|
| 7 |
"invariant_pairs": 32,
|
| 8 |
-
"invariance_rate": 0.
|
| 9 |
-
"invariant_both_correct_rate": 0.
|
| 10 |
},
|
| 11 |
"clean": {
|
| 12 |
"n": 1264,
|
| 13 |
-
"nll": 0.
|
| 14 |
-
"acc": 0.
|
| 15 |
-
"ece": 0.
|
| 16 |
-
"brier": 0.
|
| 17 |
-
"mean_conf": 0.
|
| 18 |
-
"confident_error_rate": 0.
|
| 19 |
-
"coverage_at_0_9": 0.
|
| 20 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
"selective": {
|
| 22 |
"0.5": {
|
| 23 |
"coverage": 0.5,
|
| 24 |
-
"accuracy": 0.
|
| 25 |
-
"confidence_cutoff": 0.
|
| 26 |
},
|
| 27 |
"0.8": {
|
| 28 |
"coverage": 0.8006329113924051,
|
| 29 |
-
"accuracy": 0.
|
| 30 |
-
"confidence_cutoff": 0.
|
| 31 |
}
|
| 32 |
},
|
| 33 |
-
"score_mae": 0.
|
| 34 |
-
"ranked_probability_score": 0.
|
| 35 |
},
|
| 36 |
"tasks": {
|
| 37 |
"agnews": {
|
| 38 |
"n": 80,
|
| 39 |
-
"nll": 0.
|
| 40 |
"acc": 0.8375,
|
| 41 |
-
"ece": 0.
|
| 42 |
-
"brier": 0.
|
| 43 |
-
"mean_conf": 0.
|
| 44 |
-
"confident_error_rate": 0.
|
| 45 |
"coverage_at_0_9": 0.9,
|
| 46 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
"selective": {
|
| 48 |
"0.5": {
|
| 49 |
"coverage": 0.5,
|
| 50 |
-
"accuracy": 0.
|
| 51 |
-
"confidence_cutoff": 0.
|
| 52 |
},
|
| 53 |
"0.8": {
|
| 54 |
"coverage": 0.8,
|
| 55 |
-
"accuracy": 0.
|
| 56 |
-
"confidence_cutoff": 0.
|
| 57 |
}
|
| 58 |
}
|
| 59 |
},
|
| 60 |
"agnews_yn": {
|
| 61 |
"n": 160,
|
| 62 |
-
"nll": 0.
|
| 63 |
-
"acc": 0.
|
| 64 |
-
"ece": 0.
|
| 65 |
-
"brier": 0.
|
| 66 |
-
"mean_conf": 0.
|
| 67 |
"confident_error_rate": 0.0625,
|
| 68 |
-
"coverage_at_0_9": 0.
|
| 69 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
"selective": {
|
| 71 |
"0.5": {
|
| 72 |
"coverage": 0.5,
|
| 73 |
"accuracy": 1.0,
|
| 74 |
-
"confidence_cutoff": 0.
|
| 75 |
},
|
| 76 |
"0.8": {
|
| 77 |
"coverage": 0.8,
|
| 78 |
"accuracy": 0.96875,
|
| 79 |
-
"confidence_cutoff": 0.
|
| 80 |
}
|
| 81 |
}
|
| 82 |
},
|
| 83 |
"amazon": {
|
| 84 |
"n": 80,
|
| 85 |
-
"nll": 1.
|
| 86 |
-
"acc": 0.
|
| 87 |
-
"ece": 0.
|
| 88 |
-
"brier": 0.
|
| 89 |
-
"mean_conf": 0.
|
| 90 |
-
"confident_error_rate": 0.
|
| 91 |
-
"coverage_at_0_9": 0.
|
| 92 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
"selective": {
|
| 94 |
"0.5": {
|
| 95 |
"coverage": 0.5,
|
| 96 |
-
"accuracy": 0.
|
| 97 |
-
"confidence_cutoff": 0.
|
| 98 |
},
|
| 99 |
"0.8": {
|
| 100 |
"coverage": 0.8,
|
| 101 |
-
"accuracy": 0.
|
| 102 |
-
"confidence_cutoff": 0.
|
| 103 |
}
|
| 104 |
},
|
| 105 |
-
"score_mae": 0.
|
| 106 |
-
"ranked_probability_score": 0.
|
| 107 |
},
|
| 108 |
"banking77": {
|
| 109 |
"n": 80,
|
| 110 |
-
"nll": 0.
|
| 111 |
-
"acc": 0.
|
| 112 |
-
"ece": 0.
|
| 113 |
-
"brier": 0.
|
| 114 |
-
"mean_conf": 0.
|
| 115 |
-
"confident_error_rate": 0.
|
| 116 |
-
"coverage_at_0_9": 0.
|
| 117 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 118 |
"selective": {
|
| 119 |
"0.5": {
|
| 120 |
"coverage": 0.5,
|
| 121 |
"accuracy": 0.975,
|
| 122 |
-
"confidence_cutoff": 0.
|
| 123 |
},
|
| 124 |
"0.8": {
|
| 125 |
"coverage": 0.8,
|
| 126 |
-
"accuracy": 0.
|
| 127 |
-
"confidence_cutoff": 0.
|
| 128 |
}
|
| 129 |
}
|
| 130 |
},
|
| 131 |
"boolq": {
|
| 132 |
"n": 80,
|
| 133 |
-
"nll": 0.
|
| 134 |
-
"acc": 0.
|
| 135 |
-
"ece": 0.
|
| 136 |
-
"brier": 0.
|
| 137 |
-
"mean_conf": 0.
|
| 138 |
-
"confident_error_rate": 0.
|
| 139 |
-
"coverage_at_0_9": 0.
|
| 140 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
"selective": {
|
| 142 |
"0.5": {
|
| 143 |
"coverage": 0.5,
|
| 144 |
-
"accuracy": 0.
|
| 145 |
-
"confidence_cutoff": 0.
|
| 146 |
},
|
| 147 |
"0.8": {
|
| 148 |
"coverage": 0.8,
|
| 149 |
-
"accuracy": 0.
|
| 150 |
-
"confidence_cutoff": 0.
|
| 151 |
}
|
| 152 |
}
|
| 153 |
},
|
| 154 |
"composition_atom": {
|
| 155 |
"n": 16,
|
| 156 |
-
"nll": 0.
|
| 157 |
"acc": 1.0,
|
| 158 |
-
"ece": 0.
|
| 159 |
-
"brier": 0.
|
| 160 |
-
"mean_conf": 0.
|
| 161 |
"confident_error_rate": 0.0,
|
| 162 |
-
"coverage_at_0_9": 0.
|
| 163 |
"accuracy_at_0_9": 1.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 164 |
"selective": {
|
| 165 |
"0.5": {
|
| 166 |
"coverage": 0.5,
|
| 167 |
"accuracy": 1.0,
|
| 168 |
-
"confidence_cutoff": 0.
|
| 169 |
},
|
| 170 |
"0.8": {
|
| 171 |
"coverage": 0.8125,
|
| 172 |
"accuracy": 1.0,
|
| 173 |
-
"confidence_cutoff": 0.
|
| 174 |
}
|
| 175 |
}
|
| 176 |
},
|
| 177 |
"composition_conditional": {
|
| 178 |
"n": 16,
|
| 179 |
-
"nll":
|
| 180 |
-
"acc": 0.
|
| 181 |
-
"ece": 0.
|
| 182 |
-
"brier": 0.
|
| 183 |
-
"mean_conf": 0.
|
| 184 |
-
"confident_error_rate": 0.
|
| 185 |
"coverage_at_0_9": 0.25,
|
| 186 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 187 |
"selective": {
|
| 188 |
"0.5": {
|
| 189 |
"coverage": 0.5,
|
| 190 |
-
"accuracy": 0.
|
| 191 |
-
"confidence_cutoff": 0.
|
| 192 |
},
|
| 193 |
"0.8": {
|
| 194 |
"coverage": 0.8125,
|
| 195 |
-
"accuracy": 0.
|
| 196 |
-
"confidence_cutoff": 0.
|
| 197 |
}
|
| 198 |
}
|
| 199 |
},
|
| 200 |
"composition_conjunction": {
|
| 201 |
"n": 16,
|
| 202 |
-
"nll": 0.
|
| 203 |
-
"acc": 0.
|
| 204 |
-
"ece": 0.
|
| 205 |
-
"brier": 0.
|
| 206 |
-
"mean_conf": 0.
|
| 207 |
"confident_error_rate": 0.0,
|
| 208 |
-
"coverage_at_0_9": 0.
|
| 209 |
"accuracy_at_0_9": 1.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 210 |
"selective": {
|
| 211 |
"0.5": {
|
| 212 |
-
"coverage": 0.
|
| 213 |
-
"accuracy": 0.
|
| 214 |
-
"confidence_cutoff": 0.
|
| 215 |
},
|
| 216 |
"0.8": {
|
| 217 |
-
"coverage": 0.
|
| 218 |
-
"accuracy": 0.
|
| 219 |
-
"confidence_cutoff": 0.
|
| 220 |
}
|
| 221 |
}
|
| 222 |
},
|
| 223 |
"composition_disjunction": {
|
| 224 |
"n": 16,
|
| 225 |
-
"nll": 0.
|
| 226 |
-
"acc": 0.
|
| 227 |
-
"ece": 0.
|
| 228 |
-
"brier": 0.
|
| 229 |
-
"mean_conf": 0.
|
| 230 |
"confident_error_rate": 0.0,
|
| 231 |
-
"coverage_at_0_9": 0.
|
| 232 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 233 |
"selective": {
|
| 234 |
"0.5": {
|
| 235 |
-
"coverage": 0.
|
| 236 |
-
"accuracy": 0.
|
| 237 |
-
"confidence_cutoff": 0.
|
| 238 |
},
|
| 239 |
"0.8": {
|
| 240 |
"coverage": 0.8125,
|
| 241 |
-
"accuracy": 0.
|
| 242 |
-
"confidence_cutoff": 0.
|
| 243 |
}
|
| 244 |
}
|
| 245 |
},
|
| 246 |
"composition_exception": {
|
| 247 |
"n": 16,
|
| 248 |
-
"nll": 0.
|
| 249 |
-
"acc": 0.
|
| 250 |
-
"ece": 0.
|
| 251 |
-
"brier": 0.
|
| 252 |
-
"mean_conf": 0.
|
| 253 |
"confident_error_rate": 0.0,
|
| 254 |
"coverage_at_0_9": 0.0,
|
| 255 |
"accuracy_at_0_9": null,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 256 |
"selective": {
|
| 257 |
"0.5": {
|
| 258 |
"coverage": 0.5,
|
| 259 |
"accuracy": 0.625,
|
| 260 |
-
"confidence_cutoff": 0.
|
| 261 |
},
|
| 262 |
"0.8": {
|
| 263 |
-
"coverage": 0.
|
| 264 |
-
"accuracy": 0.
|
| 265 |
-
"confidence_cutoff": 0.
|
| 266 |
}
|
| 267 |
}
|
| 268 |
},
|
| 269 |
"composition_negation": {
|
| 270 |
"n": 16,
|
| 271 |
-
"nll": 0.
|
| 272 |
-
"acc": 0.
|
| 273 |
-
"ece": 0.
|
| 274 |
-
"brier": 0.
|
| 275 |
-
"mean_conf": 0.
|
| 276 |
-
"confident_error_rate": 0.
|
| 277 |
-
"coverage_at_0_9": 0.
|
| 278 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 279 |
"selective": {
|
| 280 |
"0.5": {
|
| 281 |
"coverage": 0.5,
|
| 282 |
-
"accuracy":
|
| 283 |
-
"confidence_cutoff": 0.
|
| 284 |
},
|
| 285 |
"0.8": {
|
| 286 |
"coverage": 0.875,
|
| 287 |
-
"accuracy": 0.
|
| 288 |
-
"confidence_cutoff": 0.
|
| 289 |
}
|
| 290 |
}
|
| 291 |
},
|
| 292 |
"composition_nested_and": {
|
| 293 |
"n": 16,
|
| 294 |
-
"nll": 0.
|
| 295 |
-
"acc": 0.
|
| 296 |
-
"ece": 0.
|
| 297 |
-
"brier": 0.
|
| 298 |
-
"mean_conf": 0.
|
| 299 |
"confident_error_rate": 0.0,
|
| 300 |
"coverage_at_0_9": 0.0,
|
| 301 |
"accuracy_at_0_9": null,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 302 |
"selective": {
|
| 303 |
"0.5": {
|
| 304 |
-
"coverage": 0.
|
| 305 |
-
"accuracy": 0.
|
| 306 |
-
"confidence_cutoff": 0.
|
| 307 |
},
|
| 308 |
"0.8": {
|
| 309 |
"coverage": 0.8125,
|
| 310 |
-
"accuracy": 0.
|
| 311 |
-
"confidence_cutoff": 0.
|
| 312 |
}
|
| 313 |
}
|
| 314 |
},
|
| 315 |
"composition_nested_or": {
|
| 316 |
"n": 16,
|
| 317 |
-
"nll": 0.
|
| 318 |
-
"acc": 0.
|
| 319 |
-
"ece": 0.
|
| 320 |
-
"brier": 0.
|
| 321 |
-
"mean_conf": 0.
|
| 322 |
"confident_error_rate": 0.0,
|
| 323 |
"coverage_at_0_9": 0.0625,
|
| 324 |
"accuracy_at_0_9": 1.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 325 |
"selective": {
|
| 326 |
"0.5": {
|
| 327 |
-
"coverage": 0.
|
| 328 |
-
"accuracy": 0.
|
| 329 |
-
"confidence_cutoff": 0.
|
| 330 |
},
|
| 331 |
"0.8": {
|
| 332 |
-
"coverage": 0.
|
| 333 |
-
"accuracy": 0.
|
| 334 |
-
"confidence_cutoff": 0.
|
| 335 |
}
|
| 336 |
}
|
| 337 |
},
|
| 338 |
"contrastive_age_eligibility": {
|
| 339 |
"n": 24,
|
| 340 |
-
"nll": 0.
|
| 341 |
"acc": 1.0,
|
| 342 |
-
"ece": 0.
|
| 343 |
-
"brier":
|
| 344 |
-
"mean_conf": 0.
|
| 345 |
"confident_error_rate": 0.0,
|
| 346 |
"coverage_at_0_9": 1.0,
|
| 347 |
"accuracy_at_0_9": 1.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 348 |
"selective": {
|
| 349 |
"0.5": {
|
| 350 |
"coverage": 0.5,
|
| 351 |
"accuracy": 1.0,
|
| 352 |
-
"confidence_cutoff": 0.
|
| 353 |
},
|
| 354 |
"0.8": {
|
| 355 |
"coverage": 0.8333333333333334,
|
| 356 |
"accuracy": 1.0,
|
| 357 |
-
"confidence_cutoff": 0.
|
| 358 |
}
|
| 359 |
}
|
| 360 |
},
|
| 361 |
"contrastive_quantity_limit": {
|
| 362 |
"n": 24,
|
| 363 |
-
"nll": 0.
|
| 364 |
-
"acc":
|
| 365 |
-
"ece": 0.
|
| 366 |
-
"brier": 0.
|
| 367 |
-
"mean_conf": 0.
|
| 368 |
"confident_error_rate": 0.0,
|
| 369 |
"coverage_at_0_9": 0.6666666666666666,
|
| 370 |
"accuracy_at_0_9": 1.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 371 |
"selective": {
|
| 372 |
"0.5": {
|
| 373 |
"coverage": 0.5,
|
| 374 |
"accuracy": 1.0,
|
| 375 |
-
"confidence_cutoff": 0.
|
| 376 |
},
|
| 377 |
"0.8": {
|
| 378 |
"coverage": 0.8333333333333334,
|
| 379 |
"accuracy": 1.0,
|
| 380 |
-
"confidence_cutoff": 0.
|
| 381 |
}
|
| 382 |
}
|
| 383 |
},
|
| 384 |
"contrastive_return_window": {
|
| 385 |
"n": 24,
|
| 386 |
-
"nll": 0.
|
| 387 |
"acc": 0.75,
|
| 388 |
-
"ece": 0.
|
| 389 |
-
"brier": 0.
|
| 390 |
-
"mean_conf": 0.
|
| 391 |
"confident_error_rate": 0.0,
|
| 392 |
-
"coverage_at_0_9": 0.
|
| 393 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 394 |
"selective": {
|
| 395 |
"0.5": {
|
| 396 |
"coverage": 0.5,
|
| 397 |
-
"accuracy": 0.
|
| 398 |
-
"confidence_cutoff": 0.
|
| 399 |
},
|
| 400 |
"0.8": {
|
| 401 |
"coverage": 0.8333333333333334,
|
| 402 |
-
"accuracy": 0.
|
| 403 |
-
"confidence_cutoff": 0.
|
| 404 |
}
|
| 405 |
}
|
| 406 |
},
|
| 407 |
"contrastive_spend_threshold": {
|
| 408 |
"n": 24,
|
| 409 |
-
"nll": 5.
|
| 410 |
"acc": 1.0,
|
| 411 |
-
"ece": 5.
|
| 412 |
-
"brier":
|
| 413 |
-
"mean_conf": 0.
|
| 414 |
"confident_error_rate": 0.0,
|
| 415 |
"coverage_at_0_9": 1.0,
|
| 416 |
"accuracy_at_0_9": 1.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 417 |
"selective": {
|
| 418 |
"0.5": {
|
| 419 |
"coverage": 0.5,
|
| 420 |
"accuracy": 1.0,
|
| 421 |
-
"confidence_cutoff": 0.
|
| 422 |
},
|
| 423 |
"0.8": {
|
| 424 |
"coverage": 0.8333333333333334,
|
| 425 |
"accuracy": 1.0,
|
| 426 |
-
"confidence_cutoff": 0.
|
| 427 |
}
|
| 428 |
}
|
| 429 |
},
|
| 430 |
"dbpedia14": {
|
| 431 |
"n": 80,
|
| 432 |
-
"nll": 0.
|
| 433 |
"acc": 0.95,
|
| 434 |
-
"ece": 0.
|
| 435 |
-
"brier": 0.
|
| 436 |
-
"mean_conf": 0.
|
| 437 |
-
"confident_error_rate": 0.
|
| 438 |
-
"coverage_at_0_9": 0.
|
| 439 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 440 |
"selective": {
|
| 441 |
"0.5": {
|
| 442 |
"coverage": 0.5,
|
| 443 |
-
"accuracy":
|
| 444 |
-
"confidence_cutoff": 0.
|
| 445 |
},
|
| 446 |
"0.8": {
|
| 447 |
"coverage": 0.8,
|
| 448 |
"accuracy": 0.984375,
|
| 449 |
-
"confidence_cutoff": 0.
|
| 450 |
}
|
| 451 |
}
|
| 452 |
},
|
| 453 |
"imdb": {
|
| 454 |
"n": 80,
|
| 455 |
-
"nll": 0.
|
| 456 |
-
"acc": 0.
|
| 457 |
-
"ece": 0.
|
| 458 |
-
"brier": 0.
|
| 459 |
-
"mean_conf": 0.
|
| 460 |
-
"confident_error_rate": 0.
|
| 461 |
-
"coverage_at_0_9": 0.
|
| 462 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 463 |
"selective": {
|
| 464 |
"0.5": {
|
| 465 |
"coverage": 0.5,
|
| 466 |
"accuracy": 0.975,
|
| 467 |
-
"confidence_cutoff": 0.
|
| 468 |
},
|
| 469 |
"0.8": {
|
| 470 |
"coverage": 0.8,
|
| 471 |
-
"accuracy": 0.
|
| 472 |
-
"confidence_cutoff": 0.
|
| 473 |
}
|
| 474 |
}
|
| 475 |
},
|
| 476 |
"mnli": {
|
| 477 |
"n": 80,
|
| 478 |
-
"nll": 1.
|
| 479 |
-
"acc": 0.
|
| 480 |
-
"ece": 0.
|
| 481 |
-
"brier": 0.
|
| 482 |
-
"mean_conf": 0.
|
| 483 |
-
"confident_error_rate": 0.
|
| 484 |
"coverage_at_0_9": 0.8625,
|
| 485 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 486 |
"selective": {
|
| 487 |
"0.5": {
|
| 488 |
"coverage": 0.5,
|
| 489 |
-
"accuracy": 0.
|
| 490 |
-
"confidence_cutoff": 0.
|
| 491 |
},
|
| 492 |
"0.8": {
|
| 493 |
"coverage": 0.8,
|
| 494 |
-
"accuracy": 0.
|
| 495 |
-
"confidence_cutoff": 0.
|
| 496 |
}
|
| 497 |
}
|
| 498 |
},
|
| 499 |
"sst5": {
|
| 500 |
"n": 80,
|
| 501 |
-
"nll": 1.
|
| 502 |
-
"acc": 0.
|
| 503 |
-
"ece": 0.
|
| 504 |
-
"brier": 0.
|
| 505 |
-
"mean_conf": 0.
|
| 506 |
-
"confident_error_rate": 0.
|
| 507 |
-
"coverage_at_0_9": 0.
|
| 508 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 509 |
"selective": {
|
| 510 |
"0.5": {
|
| 511 |
"coverage": 0.5,
|
| 512 |
-
"accuracy": 0.
|
| 513 |
-
"confidence_cutoff": 0.
|
| 514 |
},
|
| 515 |
"0.8": {
|
| 516 |
"coverage": 0.8,
|
| 517 |
-
"accuracy": 0.
|
| 518 |
-
"confidence_cutoff": 0.
|
| 519 |
}
|
| 520 |
},
|
| 521 |
-
"score_mae": 0.
|
| 522 |
-
"ranked_probability_score": 0.
|
| 523 |
},
|
| 524 |
"trec": {
|
| 525 |
"n": 80,
|
| 526 |
-
"nll": 0.
|
| 527 |
-
"acc": 0.
|
| 528 |
-
"ece": 0.
|
| 529 |
-
"brier": 0.
|
| 530 |
-
"mean_conf": 0.
|
| 531 |
"confident_error_rate": 0.0375,
|
| 532 |
-
"coverage_at_0_9": 0.
|
| 533 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 534 |
"selective": {
|
| 535 |
"0.5": {
|
| 536 |
"coverage": 0.5,
|
| 537 |
-
"accuracy":
|
| 538 |
-
"confidence_cutoff": 0.
|
| 539 |
},
|
| 540 |
"0.8": {
|
| 541 |
"coverage": 0.8,
|
| 542 |
-
"accuracy":
|
| 543 |
-
"confidence_cutoff": 0.
|
| 544 |
}
|
| 545 |
}
|
| 546 |
},
|
| 547 |
"yelp": {
|
| 548 |
"n": 80,
|
| 549 |
-
"nll": 0.
|
| 550 |
"acc": 0.6875,
|
| 551 |
-
"ece": 0.
|
| 552 |
-
"brier": 0.
|
| 553 |
-
"mean_conf": 0.
|
| 554 |
"confident_error_rate": 0.025,
|
| 555 |
-
"coverage_at_0_9": 0.
|
| 556 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 557 |
"selective": {
|
| 558 |
"0.5": {
|
| 559 |
"coverage": 0.5,
|
| 560 |
-
"accuracy": 0.
|
| 561 |
-
"confidence_cutoff": 0.
|
| 562 |
},
|
| 563 |
"0.8": {
|
| 564 |
"coverage": 0.8,
|
| 565 |
-
"accuracy": 0.
|
| 566 |
-
"confidence_cutoff": 0.
|
| 567 |
}
|
| 568 |
},
|
| 569 |
-
"score_mae": 0.
|
| 570 |
-
"ranked_probability_score": 0.
|
| 571 |
},
|
| 572 |
"yelp_yn": {
|
| 573 |
"n": 80,
|
| 574 |
-
"nll": 0.
|
| 575 |
"acc": 0.9,
|
| 576 |
-
"ece": 0.
|
| 577 |
-
"brier": 0.
|
| 578 |
-
"mean_conf": 0.
|
| 579 |
-
"confident_error_rate": 0.
|
| 580 |
-
"coverage_at_0_9": 0.
|
| 581 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 582 |
"selective": {
|
| 583 |
"0.5": {
|
| 584 |
"coverage": 0.5,
|
| 585 |
"accuracy": 0.975,
|
| 586 |
-
"confidence_cutoff": 0.
|
| 587 |
},
|
| 588 |
"0.8": {
|
| 589 |
"coverage": 0.8,
|
| 590 |
-
"accuracy": 0.
|
| 591 |
-
"confidence_cutoff": 0.
|
| 592 |
}
|
| 593 |
}
|
| 594 |
}
|
|
@@ -596,95 +1046,167 @@
|
|
| 596 |
"variants": {
|
| 597 |
"clean": {
|
| 598 |
"n": 1264,
|
| 599 |
-
"nll": 0.
|
| 600 |
-
"acc": 0.
|
| 601 |
-
"ece": 0.
|
| 602 |
-
"brier": 0.
|
| 603 |
-
"mean_conf": 0.
|
| 604 |
-
"confident_error_rate": 0.
|
| 605 |
-
"coverage_at_0_9": 0.
|
| 606 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 607 |
"selective": {
|
| 608 |
"0.5": {
|
| 609 |
"coverage": 0.5,
|
| 610 |
-
"accuracy": 0.
|
| 611 |
-
"confidence_cutoff": 0.
|
| 612 |
},
|
| 613 |
"0.8": {
|
| 614 |
"coverage": 0.8006329113924051,
|
| 615 |
-
"accuracy": 0.
|
| 616 |
-
"confidence_cutoff": 0.
|
| 617 |
}
|
| 618 |
},
|
| 619 |
-
"score_mae": 0.
|
| 620 |
-
"ranked_probability_score": 0.
|
| 621 |
},
|
| 622 |
"none_absent": {
|
| 623 |
"n": 60,
|
| 624 |
-
"nll": 0.
|
| 625 |
-
"acc": 0.
|
| 626 |
-
"ece": 0.
|
| 627 |
-
"brier": 0.
|
| 628 |
-
"mean_conf": 0.
|
| 629 |
-
"confident_error_rate": 0.
|
| 630 |
-
"coverage_at_0_9": 0.
|
| 631 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 632 |
"selective": {
|
| 633 |
"0.5": {
|
| 634 |
"coverage": 0.5,
|
| 635 |
-
"accuracy": 0.
|
| 636 |
-
"confidence_cutoff": 0.
|
| 637 |
},
|
| 638 |
"0.8": {
|
| 639 |
"coverage": 0.8,
|
| 640 |
-
"accuracy": 0.
|
| 641 |
-
"confidence_cutoff": 0.
|
| 642 |
}
|
| 643 |
}
|
| 644 |
},
|
| 645 |
"none_present": {
|
| 646 |
"n": 60,
|
| 647 |
-
"nll": 0.
|
| 648 |
-
"acc": 0.
|
| 649 |
-
"ece": 0.
|
| 650 |
-
"brier": 0.
|
| 651 |
-
"mean_conf": 0.
|
| 652 |
-
"confident_error_rate": 0.
|
| 653 |
-
"coverage_at_0_9": 0.
|
| 654 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 655 |
"selective": {
|
| 656 |
"0.5": {
|
| 657 |
"coverage": 0.5,
|
| 658 |
-
"accuracy": 0.
|
| 659 |
-
"confidence_cutoff": 0.
|
| 660 |
},
|
| 661 |
"0.8": {
|
| 662 |
"coverage": 0.8,
|
| 663 |
-
"accuracy": 0.
|
| 664 |
-
"confidence_cutoff": 0.
|
| 665 |
}
|
| 666 |
}
|
| 667 |
},
|
| 668 |
"permuted": {
|
| 669 |
"n": 84,
|
| 670 |
-
"nll": 0.
|
| 671 |
-
"acc": 0.
|
| 672 |
-
"ece": 0.
|
| 673 |
-
"brier": 0.
|
| 674 |
-
"mean_conf": 0.
|
| 675 |
-
"confident_error_rate": 0.
|
| 676 |
-
"coverage_at_0_9": 0.
|
| 677 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 678 |
"selective": {
|
| 679 |
"0.5": {
|
| 680 |
"coverage": 0.5,
|
| 681 |
"accuracy": 0.9761904761904762,
|
| 682 |
-
"confidence_cutoff": 0.
|
| 683 |
},
|
| 684 |
"0.8": {
|
| 685 |
"coverage": 0.8095238095238095,
|
| 686 |
-
"accuracy": 0.
|
| 687 |
-
"confidence_cutoff": 0.
|
| 688 |
}
|
| 689 |
}
|
| 690 |
}
|
|
@@ -692,34 +1214,52 @@
|
|
| 692 |
"heldout_tasks": {},
|
| 693 |
"permutation": {
|
| 694 |
"n": 60,
|
| 695 |
-
"mean_max_delta": 0.
|
| 696 |
-
"flip_rate": 0.
|
| 697 |
},
|
| 698 |
-
"temperature": 1.
|
| 699 |
"calibrated_clean": {
|
| 700 |
"n": 1264,
|
| 701 |
-
"nll": 0.
|
| 702 |
-
"acc": 0.
|
| 703 |
-
"ece": 0.
|
| 704 |
-
"brier": 0.
|
| 705 |
-
"mean_conf": 0.
|
| 706 |
-
"confident_error_rate": 0.
|
| 707 |
-
"coverage_at_0_9": 0.
|
| 708 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 709 |
"selective": {
|
| 710 |
"0.5": {
|
| 711 |
"coverage": 0.5,
|
| 712 |
-
"accuracy": 0.
|
| 713 |
-
"confidence_cutoff": 0.
|
| 714 |
},
|
| 715 |
"0.8": {
|
| 716 |
"coverage": 0.8006329113924051,
|
| 717 |
-
"accuracy": 0.
|
| 718 |
-
"confidence_cutoff": 0.
|
| 719 |
}
|
| 720 |
},
|
| 721 |
-
"score_mae": 0.
|
| 722 |
-
"ranked_probability_score": 0.
|
| 723 |
},
|
| 724 |
"metric_policy": {
|
| 725 |
"nll_floor": 1e-09,
|
|
@@ -736,297 +1276,513 @@
|
|
| 736 |
"truncated_records": 0
|
| 737 |
},
|
| 738 |
"latency_ms": {
|
| 739 |
-
"median":
|
| 740 |
-
"p95":
|
| 741 |
},
|
| 742 |
"transfer": {
|
| 743 |
-
"objective": -
|
| 744 |
"paired_flip": {
|
| 745 |
"pairs": 64,
|
| 746 |
-
"flip_rate": 0.
|
| 747 |
-
"both_correct_rate": 0.
|
| 748 |
"invariant_pairs": 24,
|
| 749 |
"invariance_rate": 0.9583333333333334,
|
| 750 |
-
"invariant_both_correct_rate": 0.
|
| 751 |
},
|
| 752 |
"clean": {
|
| 753 |
"n": 656,
|
| 754 |
-
"nll": 0.
|
| 755 |
-
"acc": 0.
|
| 756 |
-
"ece": 0.
|
| 757 |
-
"brier": 0.
|
| 758 |
-
"mean_conf": 0.
|
| 759 |
-
"confident_error_rate": 0.
|
| 760 |
-
"coverage_at_0_9": 0.
|
| 761 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 762 |
"selective": {
|
| 763 |
"0.5": {
|
| 764 |
"coverage": 0.5,
|
| 765 |
-
"accuracy": 0.
|
| 766 |
-
"confidence_cutoff": 0.
|
| 767 |
},
|
| 768 |
"0.8": {
|
| 769 |
"coverage": 0.8003048780487805,
|
| 770 |
-
"accuracy": 0.
|
| 771 |
-
"confidence_cutoff": 0.
|
| 772 |
}
|
| 773 |
},
|
| 774 |
-
"score_mae": 0.
|
| 775 |
-
"ranked_probability_score": 0.
|
| 776 |
},
|
| 777 |
"tasks": {
|
| 778 |
"composition_held_and_or": {
|
| 779 |
"n": 32,
|
| 780 |
-
"nll": 0.
|
| 781 |
-
"acc": 0.
|
| 782 |
-
"ece": 0.
|
| 783 |
-
"brier": 0.
|
| 784 |
-
"mean_conf": 0.
|
| 785 |
-
"confident_error_rate": 0.
|
| 786 |
-
"coverage_at_0_9": 0.
|
| 787 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 788 |
"selective": {
|
| 789 |
"0.5": {
|
| 790 |
-
"coverage": 0.
|
| 791 |
-
"accuracy": 0.
|
| 792 |
-
"confidence_cutoff": 0.
|
| 793 |
},
|
| 794 |
"0.8": {
|
| 795 |
"coverage": 0.84375,
|
| 796 |
-
"accuracy": 0.
|
| 797 |
-
"confidence_cutoff": 0.
|
| 798 |
}
|
| 799 |
}
|
| 800 |
},
|
| 801 |
"composition_held_conditional": {
|
| 802 |
"n": 32,
|
| 803 |
-
"nll": 0.
|
| 804 |
-
"acc": 0.
|
| 805 |
-
"ece": 0.
|
| 806 |
-
"brier": 0.
|
| 807 |
-
"mean_conf": 0.
|
| 808 |
-
"confident_error_rate": 0.
|
| 809 |
-
"coverage_at_0_9": 0.
|
| 810 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 811 |
"selective": {
|
| 812 |
"0.5": {
|
| 813 |
"coverage": 0.5,
|
| 814 |
-
"accuracy": 0.
|
| 815 |
-
"confidence_cutoff": 0.
|
| 816 |
},
|
| 817 |
"0.8": {
|
| 818 |
"coverage": 0.8125,
|
| 819 |
-
"accuracy": 0.
|
| 820 |
-
"confidence_cutoff": 0.
|
| 821 |
}
|
| 822 |
}
|
| 823 |
},
|
| 824 |
"composition_held_or_not": {
|
| 825 |
"n": 32,
|
| 826 |
-
"nll": 0.
|
| 827 |
-
"acc": 0.
|
| 828 |
-
"ece": 0.
|
| 829 |
-
"brier": 0.
|
| 830 |
-
"mean_conf": 0.
|
| 831 |
"confident_error_rate": 0.0,
|
| 832 |
"coverage_at_0_9": 0.0,
|
| 833 |
"accuracy_at_0_9": null,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 834 |
"selective": {
|
| 835 |
"0.5": {
|
| 836 |
"coverage": 0.5,
|
| 837 |
-
"accuracy": 0.
|
| 838 |
-
"confidence_cutoff": 0.
|
| 839 |
},
|
| 840 |
"0.8": {
|
| 841 |
"coverage": 0.8125,
|
| 842 |
-
"accuracy": 0.
|
| 843 |
-
"confidence_cutoff": 0.
|
| 844 |
}
|
| 845 |
}
|
| 846 |
},
|
| 847 |
"contrastive_authorization": {
|
| 848 |
"n": 40,
|
| 849 |
-
"nll": 1.
|
| 850 |
-
"acc": 0.
|
| 851 |
-
"ece": 0.
|
| 852 |
-
"brier": 0.
|
| 853 |
-
"mean_conf": 0.
|
| 854 |
-
"confident_error_rate": 0.
|
| 855 |
-
"coverage_at_0_9": 0.
|
| 856 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 857 |
"selective": {
|
| 858 |
"0.5": {
|
| 859 |
"coverage": 0.5,
|
| 860 |
-
"accuracy":
|
| 861 |
-
"confidence_cutoff": 0.
|
| 862 |
},
|
| 863 |
"0.8": {
|
| 864 |
"coverage": 0.8,
|
| 865 |
"accuracy": 0.625,
|
| 866 |
-
"confidence_cutoff": 0.
|
| 867 |
}
|
| 868 |
}
|
| 869 |
},
|
| 870 |
"contrastive_deadline": {
|
| 871 |
"n": 40,
|
| 872 |
-
"nll":
|
| 873 |
-
"acc": 0.
|
| 874 |
-
"ece": 0.
|
| 875 |
-
"brier":
|
| 876 |
-
"mean_conf": 0.
|
| 877 |
-
"confident_error_rate": 0.
|
| 878 |
-
"coverage_at_0_9": 0.
|
| 879 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 880 |
"selective": {
|
| 881 |
"0.5": {
|
| 882 |
"coverage": 0.5,
|
| 883 |
-
"accuracy": 0.
|
| 884 |
-
"confidence_cutoff": 0.
|
| 885 |
},
|
| 886 |
"0.8": {
|
| 887 |
"coverage": 0.8,
|
| 888 |
"accuracy": 0.28125,
|
| 889 |
-
"confidence_cutoff": 0.
|
| 890 |
}
|
| 891 |
},
|
| 892 |
-
"score_mae": 0.
|
| 893 |
-
"ranked_probability_score": 0.
|
| 894 |
},
|
| 895 |
"emotion": {
|
| 896 |
"n": 80,
|
| 897 |
-
"nll": 1.
|
| 898 |
-
"acc": 0.
|
| 899 |
-
"ece": 0.
|
| 900 |
-
"brier": 0.
|
| 901 |
-
"mean_conf": 0.
|
| 902 |
-
"confident_error_rate": 0.
|
| 903 |
-
"coverage_at_0_9": 0.
|
| 904 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 905 |
"selective": {
|
| 906 |
"0.5": {
|
| 907 |
"coverage": 0.5,
|
| 908 |
-
"accuracy": 0.
|
| 909 |
-
"confidence_cutoff": 0.
|
| 910 |
},
|
| 911 |
"0.8": {
|
| 912 |
"coverage": 0.8,
|
| 913 |
-
"accuracy": 0.
|
| 914 |
-
"confidence_cutoff": 0.
|
| 915 |
}
|
| 916 |
}
|
| 917 |
},
|
| 918 |
"mmlu": {
|
| 919 |
"n": 80,
|
| 920 |
-
"nll": 1.
|
| 921 |
-
"acc": 0.
|
| 922 |
-
"ece": 0.
|
| 923 |
-
"brier": 0.
|
| 924 |
-
"mean_conf": 0.
|
| 925 |
-
"confident_error_rate": 0.
|
| 926 |
-
"coverage_at_0_9": 0.
|
| 927 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 928 |
"selective": {
|
| 929 |
"0.5": {
|
| 930 |
"coverage": 0.5,
|
| 931 |
-
"accuracy": 0.
|
| 932 |
-
"confidence_cutoff": 0.
|
| 933 |
},
|
| 934 |
"0.8": {
|
| 935 |
"coverage": 0.8,
|
| 936 |
-
"accuracy": 0.
|
| 937 |
-
"confidence_cutoff": 0.
|
| 938 |
}
|
| 939 |
}
|
| 940 |
},
|
| 941 |
"paws": {
|
| 942 |
"n": 80,
|
| 943 |
-
"nll": 0.
|
| 944 |
-
"acc": 0.
|
| 945 |
-
"ece": 0.
|
| 946 |
-
"brier": 0.
|
| 947 |
-
"mean_conf": 0.
|
| 948 |
"confident_error_rate": 0.0125,
|
| 949 |
"coverage_at_0_9": 0.0125,
|
| 950 |
"accuracy_at_0_9": 0.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 951 |
"selective": {
|
| 952 |
"0.5": {
|
| 953 |
"coverage": 0.5,
|
| 954 |
-
"accuracy": 0.
|
| 955 |
-
"confidence_cutoff": 0.
|
| 956 |
},
|
| 957 |
"0.8": {
|
| 958 |
"coverage": 0.8,
|
| 959 |
-
"accuracy": 0.
|
| 960 |
-
"confidence_cutoff": 0.
|
| 961 |
}
|
| 962 |
}
|
| 963 |
},
|
| 964 |
"qnli": {
|
| 965 |
"n": 80,
|
| 966 |
-
"nll": 0.
|
| 967 |
-
"acc": 0.
|
| 968 |
-
"ece": 0.
|
| 969 |
-
"brier": 0.
|
| 970 |
-
"mean_conf": 0.
|
| 971 |
-
"confident_error_rate": 0.
|
| 972 |
-
"coverage_at_0_9": 0.
|
| 973 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 974 |
"selective": {
|
| 975 |
"0.5": {
|
| 976 |
"coverage": 0.5,
|
| 977 |
-
"accuracy": 0.
|
| 978 |
-
"confidence_cutoff": 0.
|
| 979 |
},
|
| 980 |
"0.8": {
|
| 981 |
"coverage": 0.8,
|
| 982 |
-
"accuracy": 0.
|
| 983 |
-
"confidence_cutoff": 0.
|
| 984 |
}
|
| 985 |
}
|
| 986 |
},
|
| 987 |
"sciq": {
|
| 988 |
"n": 80,
|
| 989 |
-
"nll": 0.
|
| 990 |
-
"acc": 0.
|
| 991 |
-
"ece": 0.
|
| 992 |
-
"brier": 0.
|
| 993 |
-
"mean_conf": 0.
|
| 994 |
-
"confident_error_rate": 0.
|
| 995 |
-
"coverage_at_0_9": 0.
|
| 996 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 997 |
"selective": {
|
| 998 |
"0.5": {
|
| 999 |
"coverage": 0.5,
|
| 1000 |
"accuracy": 1.0,
|
| 1001 |
-
"confidence_cutoff": 0.
|
| 1002 |
},
|
| 1003 |
"0.8": {
|
| 1004 |
"coverage": 0.8,
|
| 1005 |
-
"accuracy": 0.
|
| 1006 |
-
"confidence_cutoff": 0.
|
| 1007 |
}
|
| 1008 |
}
|
| 1009 |
},
|
| 1010 |
"tweet_offensive": {
|
| 1011 |
"n": 80,
|
| 1012 |
-
"nll": 0.
|
| 1013 |
-
"acc": 0.
|
| 1014 |
-
"ece": 0.
|
| 1015 |
-
"brier": 0.
|
| 1016 |
-
"mean_conf": 0.
|
| 1017 |
-
"confident_error_rate": 0.
|
| 1018 |
-
"coverage_at_0_9": 0.
|
| 1019 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1020 |
"selective": {
|
| 1021 |
"0.5": {
|
| 1022 |
"coverage": 0.5,
|
| 1023 |
"accuracy": 0.8,
|
| 1024 |
-
"confidence_cutoff": 0.
|
| 1025 |
},
|
| 1026 |
"0.8": {
|
| 1027 |
"coverage": 0.8,
|
| 1028 |
-
"accuracy": 0.
|
| 1029 |
-
"confidence_cutoff": 0.
|
| 1030 |
}
|
| 1031 |
}
|
| 1032 |
}
|
|
@@ -1034,95 +1790,167 @@
|
|
| 1034 |
"variants": {
|
| 1035 |
"clean": {
|
| 1036 |
"n": 656,
|
| 1037 |
-
"nll": 0.
|
| 1038 |
-
"acc": 0.
|
| 1039 |
-
"ece": 0.
|
| 1040 |
-
"brier": 0.
|
| 1041 |
-
"mean_conf": 0.
|
| 1042 |
-
"confident_error_rate": 0.
|
| 1043 |
-
"coverage_at_0_9": 0.
|
| 1044 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1045 |
"selective": {
|
| 1046 |
"0.5": {
|
| 1047 |
"coverage": 0.5,
|
| 1048 |
-
"accuracy": 0.
|
| 1049 |
-
"confidence_cutoff": 0.
|
| 1050 |
},
|
| 1051 |
"0.8": {
|
| 1052 |
"coverage": 0.8003048780487805,
|
| 1053 |
-
"accuracy": 0.
|
| 1054 |
-
"confidence_cutoff": 0.
|
| 1055 |
}
|
| 1056 |
},
|
| 1057 |
-
"score_mae": 0.
|
| 1058 |
-
"ranked_probability_score": 0.
|
| 1059 |
},
|
| 1060 |
"none_absent": {
|
| 1061 |
"n": 36,
|
| 1062 |
-
"nll": 0.
|
| 1063 |
-
"acc": 0.
|
| 1064 |
-
"ece": 0.
|
| 1065 |
-
"brier": 0.
|
| 1066 |
-
"mean_conf": 0.
|
| 1067 |
"confident_error_rate": 0.05555555555555555,
|
| 1068 |
-
"coverage_at_0_9": 0.
|
| 1069 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1070 |
"selective": {
|
| 1071 |
"0.5": {
|
| 1072 |
"coverage": 0.5,
|
| 1073 |
"accuracy": 0.8888888888888888,
|
| 1074 |
-
"confidence_cutoff": 0.
|
| 1075 |
},
|
| 1076 |
"0.8": {
|
| 1077 |
"coverage": 0.8055555555555556,
|
| 1078 |
-
"accuracy": 0.
|
| 1079 |
-
"confidence_cutoff": 0.
|
| 1080 |
}
|
| 1081 |
}
|
| 1082 |
},
|
| 1083 |
"none_present": {
|
| 1084 |
"n": 36,
|
| 1085 |
-
"nll":
|
| 1086 |
-
"acc": 0.
|
| 1087 |
-
"ece": 0.
|
| 1088 |
-
"brier": 0.
|
| 1089 |
-
"mean_conf": 0.
|
| 1090 |
-
"confident_error_rate": 0.
|
| 1091 |
-
"coverage_at_0_9": 0.
|
| 1092 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1093 |
"selective": {
|
| 1094 |
"0.5": {
|
| 1095 |
"coverage": 0.5,
|
| 1096 |
"accuracy": 0.6111111111111112,
|
| 1097 |
-
"confidence_cutoff": 0.
|
| 1098 |
},
|
| 1099 |
"0.8": {
|
| 1100 |
"coverage": 0.8055555555555556,
|
| 1101 |
-
"accuracy": 0.
|
| 1102 |
-
"confidence_cutoff": 0.
|
| 1103 |
}
|
| 1104 |
}
|
| 1105 |
},
|
| 1106 |
"permuted": {
|
| 1107 |
"n": 36,
|
| 1108 |
-
"nll": 0.
|
| 1109 |
"acc": 0.6944444444444444,
|
| 1110 |
-
"ece": 0.
|
| 1111 |
-
"brier": 0.
|
| 1112 |
-
"mean_conf": 0.
|
| 1113 |
-
"confident_error_rate": 0.
|
| 1114 |
-
"coverage_at_0_9": 0.
|
| 1115 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1116 |
"selective": {
|
| 1117 |
"0.5": {
|
| 1118 |
"coverage": 0.5,
|
| 1119 |
-
"accuracy": 0.
|
| 1120 |
-
"confidence_cutoff": 0.
|
| 1121 |
},
|
| 1122 |
"0.8": {
|
| 1123 |
"coverage": 0.8055555555555556,
|
| 1124 |
-
"accuracy": 0.
|
| 1125 |
-
"confidence_cutoff": 0.
|
| 1126 |
}
|
| 1127 |
}
|
| 1128 |
}
|
|
@@ -1130,290 +1958,506 @@
|
|
| 1130 |
"heldout_tasks": {
|
| 1131 |
"composition_held_and_or": {
|
| 1132 |
"n": 32,
|
| 1133 |
-
"nll": 0.
|
| 1134 |
-
"acc": 0.
|
| 1135 |
-
"ece": 0.
|
| 1136 |
-
"brier": 0.
|
| 1137 |
-
"mean_conf": 0.
|
| 1138 |
-
"confident_error_rate": 0.
|
| 1139 |
-
"coverage_at_0_9": 0.
|
| 1140 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1141 |
"selective": {
|
| 1142 |
"0.5": {
|
| 1143 |
-
"coverage": 0.
|
| 1144 |
-
"accuracy": 0.
|
| 1145 |
-
"confidence_cutoff": 0.
|
| 1146 |
},
|
| 1147 |
"0.8": {
|
| 1148 |
"coverage": 0.84375,
|
| 1149 |
-
"accuracy": 0.
|
| 1150 |
-
"confidence_cutoff": 0.
|
| 1151 |
}
|
| 1152 |
}
|
| 1153 |
},
|
| 1154 |
"composition_held_conditional": {
|
| 1155 |
"n": 32,
|
| 1156 |
-
"nll": 0.
|
| 1157 |
-
"acc": 0.
|
| 1158 |
-
"ece": 0.
|
| 1159 |
-
"brier": 0.
|
| 1160 |
-
"mean_conf": 0.
|
| 1161 |
-
"confident_error_rate": 0.
|
| 1162 |
-
"coverage_at_0_9": 0.
|
| 1163 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1164 |
"selective": {
|
| 1165 |
"0.5": {
|
| 1166 |
"coverage": 0.5,
|
| 1167 |
-
"accuracy": 0.
|
| 1168 |
-
"confidence_cutoff": 0.
|
| 1169 |
},
|
| 1170 |
"0.8": {
|
| 1171 |
"coverage": 0.8125,
|
| 1172 |
-
"accuracy": 0.
|
| 1173 |
-
"confidence_cutoff": 0.
|
| 1174 |
}
|
| 1175 |
}
|
| 1176 |
},
|
| 1177 |
"composition_held_or_not": {
|
| 1178 |
"n": 32,
|
| 1179 |
-
"nll": 0.
|
| 1180 |
-
"acc": 0.
|
| 1181 |
-
"ece": 0.
|
| 1182 |
-
"brier": 0.
|
| 1183 |
-
"mean_conf": 0.
|
| 1184 |
"confident_error_rate": 0.0,
|
| 1185 |
"coverage_at_0_9": 0.0,
|
| 1186 |
"accuracy_at_0_9": null,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1187 |
"selective": {
|
| 1188 |
"0.5": {
|
| 1189 |
"coverage": 0.5,
|
| 1190 |
-
"accuracy": 0.
|
| 1191 |
-
"confidence_cutoff": 0.
|
| 1192 |
},
|
| 1193 |
"0.8": {
|
| 1194 |
"coverage": 0.8125,
|
| 1195 |
-
"accuracy": 0.
|
| 1196 |
-
"confidence_cutoff": 0.
|
| 1197 |
}
|
| 1198 |
}
|
| 1199 |
},
|
| 1200 |
"contrastive_authorization": {
|
| 1201 |
"n": 40,
|
| 1202 |
-
"nll": 1.
|
| 1203 |
-
"acc": 0.
|
| 1204 |
-
"ece": 0.
|
| 1205 |
-
"brier": 0.
|
| 1206 |
-
"mean_conf": 0.
|
| 1207 |
-
"confident_error_rate": 0.
|
| 1208 |
-
"coverage_at_0_9": 0.
|
| 1209 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1210 |
"selective": {
|
| 1211 |
"0.5": {
|
| 1212 |
"coverage": 0.5,
|
| 1213 |
-
"accuracy":
|
| 1214 |
-
"confidence_cutoff": 0.
|
| 1215 |
},
|
| 1216 |
"0.8": {
|
| 1217 |
"coverage": 0.8,
|
| 1218 |
"accuracy": 0.625,
|
| 1219 |
-
"confidence_cutoff": 0.
|
| 1220 |
}
|
| 1221 |
}
|
| 1222 |
},
|
| 1223 |
"contrastive_deadline": {
|
| 1224 |
"n": 40,
|
| 1225 |
-
"nll":
|
| 1226 |
-
"acc": 0.
|
| 1227 |
-
"ece": 0.
|
| 1228 |
-
"brier":
|
| 1229 |
-
"mean_conf": 0.
|
| 1230 |
-
"confident_error_rate": 0.
|
| 1231 |
-
"coverage_at_0_9": 0.
|
| 1232 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1233 |
"selective": {
|
| 1234 |
"0.5": {
|
| 1235 |
"coverage": 0.5,
|
| 1236 |
-
"accuracy": 0.
|
| 1237 |
-
"confidence_cutoff": 0.
|
| 1238 |
},
|
| 1239 |
"0.8": {
|
| 1240 |
"coverage": 0.8,
|
| 1241 |
"accuracy": 0.28125,
|
| 1242 |
-
"confidence_cutoff": 0.
|
| 1243 |
}
|
| 1244 |
},
|
| 1245 |
-
"score_mae": 0.
|
| 1246 |
-
"ranked_probability_score": 0.
|
| 1247 |
},
|
| 1248 |
"emotion": {
|
| 1249 |
"n": 80,
|
| 1250 |
-
"nll": 1.
|
| 1251 |
-
"acc": 0.
|
| 1252 |
-
"ece": 0.
|
| 1253 |
-
"brier": 0.
|
| 1254 |
-
"mean_conf": 0.
|
| 1255 |
-
"confident_error_rate": 0.
|
| 1256 |
-
"coverage_at_0_9": 0.
|
| 1257 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1258 |
"selective": {
|
| 1259 |
"0.5": {
|
| 1260 |
"coverage": 0.5,
|
| 1261 |
-
"accuracy": 0.
|
| 1262 |
-
"confidence_cutoff": 0.
|
| 1263 |
},
|
| 1264 |
"0.8": {
|
| 1265 |
"coverage": 0.8,
|
| 1266 |
-
"accuracy": 0.
|
| 1267 |
-
"confidence_cutoff": 0.
|
| 1268 |
}
|
| 1269 |
}
|
| 1270 |
},
|
| 1271 |
"mmlu": {
|
| 1272 |
"n": 80,
|
| 1273 |
-
"nll": 1.
|
| 1274 |
-
"acc": 0.
|
| 1275 |
-
"ece": 0.
|
| 1276 |
-
"brier": 0.
|
| 1277 |
-
"mean_conf": 0.
|
| 1278 |
-
"confident_error_rate": 0.
|
| 1279 |
-
"coverage_at_0_9": 0.
|
| 1280 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1281 |
"selective": {
|
| 1282 |
"0.5": {
|
| 1283 |
"coverage": 0.5,
|
| 1284 |
-
"accuracy": 0.
|
| 1285 |
-
"confidence_cutoff": 0.
|
| 1286 |
},
|
| 1287 |
"0.8": {
|
| 1288 |
"coverage": 0.8,
|
| 1289 |
-
"accuracy": 0.
|
| 1290 |
-
"confidence_cutoff": 0.
|
| 1291 |
}
|
| 1292 |
}
|
| 1293 |
},
|
| 1294 |
"paws": {
|
| 1295 |
"n": 80,
|
| 1296 |
-
"nll": 0.
|
| 1297 |
-
"acc": 0.
|
| 1298 |
-
"ece": 0.
|
| 1299 |
-
"brier": 0.
|
| 1300 |
-
"mean_conf": 0.
|
| 1301 |
"confident_error_rate": 0.0125,
|
| 1302 |
"coverage_at_0_9": 0.0125,
|
| 1303 |
"accuracy_at_0_9": 0.0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1304 |
"selective": {
|
| 1305 |
"0.5": {
|
| 1306 |
"coverage": 0.5,
|
| 1307 |
-
"accuracy": 0.
|
| 1308 |
-
"confidence_cutoff": 0.
|
| 1309 |
},
|
| 1310 |
"0.8": {
|
| 1311 |
"coverage": 0.8,
|
| 1312 |
-
"accuracy": 0.
|
| 1313 |
-
"confidence_cutoff": 0.
|
| 1314 |
}
|
| 1315 |
}
|
| 1316 |
},
|
| 1317 |
"qnli": {
|
| 1318 |
"n": 80,
|
| 1319 |
-
"nll": 0.
|
| 1320 |
-
"acc": 0.
|
| 1321 |
-
"ece": 0.
|
| 1322 |
-
"brier": 0.
|
| 1323 |
-
"mean_conf": 0.
|
| 1324 |
-
"confident_error_rate": 0.
|
| 1325 |
-
"coverage_at_0_9": 0.
|
| 1326 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1327 |
"selective": {
|
| 1328 |
"0.5": {
|
| 1329 |
"coverage": 0.5,
|
| 1330 |
-
"accuracy": 0.
|
| 1331 |
-
"confidence_cutoff": 0.
|
| 1332 |
},
|
| 1333 |
"0.8": {
|
| 1334 |
"coverage": 0.8,
|
| 1335 |
-
"accuracy": 0.
|
| 1336 |
-
"confidence_cutoff": 0.
|
| 1337 |
}
|
| 1338 |
}
|
| 1339 |
},
|
| 1340 |
"sciq": {
|
| 1341 |
"n": 80,
|
| 1342 |
-
"nll": 0.
|
| 1343 |
-
"acc": 0.
|
| 1344 |
-
"ece": 0.
|
| 1345 |
-
"brier": 0.
|
| 1346 |
-
"mean_conf": 0.
|
| 1347 |
-
"confident_error_rate": 0.
|
| 1348 |
-
"coverage_at_0_9": 0.
|
| 1349 |
-
"accuracy_at_0_9":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1350 |
"selective": {
|
| 1351 |
"0.5": {
|
| 1352 |
"coverage": 0.5,
|
| 1353 |
"accuracy": 1.0,
|
| 1354 |
-
"confidence_cutoff": 0.
|
| 1355 |
},
|
| 1356 |
"0.8": {
|
| 1357 |
"coverage": 0.8,
|
| 1358 |
-
"accuracy": 0.
|
| 1359 |
-
"confidence_cutoff": 0.
|
| 1360 |
}
|
| 1361 |
}
|
| 1362 |
},
|
| 1363 |
"tweet_offensive": {
|
| 1364 |
"n": 80,
|
| 1365 |
-
"nll": 0.
|
| 1366 |
-
"acc": 0.
|
| 1367 |
-
"ece": 0.
|
| 1368 |
-
"brier": 0.
|
| 1369 |
-
"mean_conf": 0.
|
| 1370 |
-
"confident_error_rate": 0.
|
| 1371 |
-
"coverage_at_0_9": 0.
|
| 1372 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1373 |
"selective": {
|
| 1374 |
"0.5": {
|
| 1375 |
"coverage": 0.5,
|
| 1376 |
"accuracy": 0.8,
|
| 1377 |
-
"confidence_cutoff": 0.
|
| 1378 |
},
|
| 1379 |
"0.8": {
|
| 1380 |
"coverage": 0.8,
|
| 1381 |
-
"accuracy": 0.
|
| 1382 |
-
"confidence_cutoff": 0.
|
| 1383 |
}
|
| 1384 |
}
|
| 1385 |
}
|
| 1386 |
},
|
| 1387 |
"permutation": {
|
| 1388 |
"n": 36,
|
| 1389 |
-
"mean_max_delta": 0.
|
| 1390 |
-
"flip_rate": 0.
|
| 1391 |
},
|
| 1392 |
-
"temperature": 1.
|
| 1393 |
"calibrated_clean": {
|
| 1394 |
"n": 656,
|
| 1395 |
-
"nll": 0.
|
| 1396 |
-
"acc": 0.
|
| 1397 |
-
"ece": 0.
|
| 1398 |
-
"brier": 0.
|
| 1399 |
-
"mean_conf": 0.
|
| 1400 |
-
"confident_error_rate": 0.
|
| 1401 |
-
"coverage_at_0_9": 0.
|
| 1402 |
-
"accuracy_at_0_9": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1403 |
"selective": {
|
| 1404 |
"0.5": {
|
| 1405 |
"coverage": 0.5,
|
| 1406 |
-
"accuracy": 0.
|
| 1407 |
-
"confidence_cutoff": 0.
|
| 1408 |
},
|
| 1409 |
"0.8": {
|
| 1410 |
"coverage": 0.8003048780487805,
|
| 1411 |
-
"accuracy": 0.
|
| 1412 |
-
"confidence_cutoff": 0.
|
| 1413 |
}
|
| 1414 |
},
|
| 1415 |
-
"score_mae": 0.
|
| 1416 |
-
"ranked_probability_score": 0.
|
| 1417 |
},
|
| 1418 |
"metric_policy": {
|
| 1419 |
"nll_floor": 1e-09,
|
|
@@ -1430,16 +2474,16 @@
|
|
| 1430 |
"truncated_records": 0
|
| 1431 |
},
|
| 1432 |
"latency_ms": {
|
| 1433 |
-
"median":
|
| 1434 |
-
"p95":
|
| 1435 |
},
|
| 1436 |
"suite_sha256": "31677c2256b406222e7d94ffdc0a02a70ce05746b9efe307876024c4e77291d1"
|
| 1437 |
},
|
| 1438 |
"provenance": {
|
| 1439 |
"config": {
|
| 1440 |
"epochs": 2,
|
| 1441 |
-
"seed":
|
| 1442 |
-
"lr": 0.
|
| 1443 |
"lora": 16,
|
| 1444 |
"accum": 1,
|
| 1445 |
"batch": 8,
|
|
@@ -1450,34 +2494,41 @@
|
|
| 1450 |
"p_none_distract": 0.12,
|
| 1451 |
"p_distract": 0.15,
|
| 1452 |
"p_none_pair": 0.25,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1453 |
"base": "Qwen/Qwen3-0.6B-Base",
|
| 1454 |
"dtype": "bf16"
|
| 1455 |
},
|
| 1456 |
-
"config_sha256": "
|
| 1457 |
-
"suite_sha256": "
|
| 1458 |
"source_hashes": {
|
| 1459 |
"kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
|
|
|
|
| 1460 |
"kev/api.py": "cdb0602684d798ccd3fc2c86be622f2a8e7bcdeee64d07f95604c7d3fd4701ec",
|
| 1461 |
-
"kev/
|
|
|
|
| 1462 |
"kev/compare.py": "bd0445f021de59e35c7bff9304e39dd2e1211e3877a453575594ae7b81b0ada4",
|
| 1463 |
-
"kev/composition.py": "
|
| 1464 |
-
"kev/contrastive.py": "
|
| 1465 |
-
"kev/data.py": "
|
| 1466 |
-
"kev/evaluate.py": "
|
| 1467 |
-
"kev/experiment.py": "
|
| 1468 |
"kev/jev.py": "e0213782359ba2f95adbf045ddaf0a008b08b4162bf6a9b66d4ce55fc91a51cb",
|
| 1469 |
-
"kev/model.py": "
|
| 1470 |
"kev/plot.py": "d689c7dd18cde7f9ea77cb4f55c50ff7da1880b21cb2ecf244348e45842a1382",
|
| 1471 |
-
"kev/publish.py": "
|
| 1472 |
-
"kev/serve.py": "
|
| 1473 |
-
"kev/study_v3.py": "
|
| 1474 |
-
"kev/suite.py": "
|
| 1475 |
-
"kev/train.py": "
|
| 1476 |
-
"modal_app.py": "
|
| 1477 |
"pyproject.toml": "ce566ee11b68c0cb587bb591943641f403d761d820817693dae7590faacf9888",
|
| 1478 |
"uv.lock": "846fdc64a8bd6279a17a4365f18cd4e92ba9622d50d378aafcb5be6604296138"
|
| 1479 |
},
|
| 1480 |
-
"git_commit": "
|
| 1481 |
"platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.36",
|
| 1482 |
"torch": "2.8.0+cu128",
|
| 1483 |
"device": "cuda",
|
|
@@ -1487,7 +2538,7 @@
|
|
| 1487 |
"mechanism_checks": {
|
| 1488 |
"n": 8,
|
| 1489 |
"packed_max_delta": 0.0,
|
| 1490 |
-
"sibling_max_delta":
|
| 1491 |
"tolerance": 0.001,
|
| 1492 |
"passed": true
|
| 1493 |
},
|
|
@@ -1498,25 +2549,25 @@
|
|
| 1498 |
"isolation_and_packing": true,
|
| 1499 |
"transfer_complete": true,
|
| 1500 |
"heldout_pairs_at_least_70pct": false,
|
| 1501 |
-
"transfer_confident_errors_below_10pct":
|
| 1502 |
},
|
| 1503 |
"policy": "Research screening only: 1e-3 isolation; 5pp task regression; 70% heldout pair correctness; <=10% confident errors; no automatic release."
|
| 1504 |
},
|
| 1505 |
-
"wall_seconds": 727.
|
| 1506 |
"promotable": false,
|
| 1507 |
"test_evaluated": false,
|
| 1508 |
"training_resources": {
|
| 1509 |
-
"wall_seconds":
|
| 1510 |
-
"records_seen":
|
| 1511 |
-
"requested_records":
|
| 1512 |
"truncated_records": 0,
|
| 1513 |
"rejected_records": 0,
|
| 1514 |
-
"optimizer_steps":
|
| 1515 |
-
"forward_tokens":
|
| 1516 |
-
"peak_device_bytes":
|
| 1517 |
"device": "cuda",
|
| 1518 |
"dtype": "bf16",
|
| 1519 |
"batch": 8,
|
| 1520 |
-
"peak_rss_bytes":
|
| 1521 |
}
|
| 1522 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"objective": -0.6195928202255924,
|
| 3 |
"paired_flip": {
|
| 4 |
"pairs": 80,
|
| 5 |
"flip_rate": 0.7,
|
| 6 |
+
"both_correct_rate": 0.675,
|
| 7 |
"invariant_pairs": 32,
|
| 8 |
+
"invariance_rate": 0.9375,
|
| 9 |
+
"invariant_both_correct_rate": 0.5625
|
| 10 |
},
|
| 11 |
"clean": {
|
| 12 |
"n": 1264,
|
| 13 |
+
"nll": 0.6291677798298513,
|
| 14 |
+
"acc": 0.8006329113924051,
|
| 15 |
+
"ece": 0.0856521215240488,
|
| 16 |
+
"brier": 0.2969903053792275,
|
| 17 |
+
"mean_conf": 0.8857679456333062,
|
| 18 |
+
"confident_error_rate": 0.05458860759493671,
|
| 19 |
+
"coverage_at_0_9": 0.6566455696202531,
|
| 20 |
+
"accuracy_at_0_9": 0.9168674698795181,
|
| 21 |
+
"confidence_bias": 0.08513503424090108,
|
| 22 |
+
"top_bins": {
|
| 23 |
+
"0.9": {
|
| 24 |
+
"n": 830,
|
| 25 |
+
"errors": 69,
|
| 26 |
+
"error_rate": 0.08313253012048193
|
| 27 |
+
},
|
| 28 |
+
"0.95": {
|
| 29 |
+
"n": 758,
|
| 30 |
+
"errors": 50,
|
| 31 |
+
"error_rate": 0.06596306068601583
|
| 32 |
+
},
|
| 33 |
+
"0.99": {
|
| 34 |
+
"n": 633,
|
| 35 |
+
"errors": 29,
|
| 36 |
+
"error_rate": 0.045813586097946286
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
"selective": {
|
| 40 |
"0.5": {
|
| 41 |
"coverage": 0.5,
|
| 42 |
+
"accuracy": 0.9541139240506329,
|
| 43 |
+
"confidence_cutoff": 0.9903846647088901
|
| 44 |
},
|
| 45 |
"0.8": {
|
| 46 |
"coverage": 0.8006329113924051,
|
| 47 |
+
"accuracy": 0.8685770750988142,
|
| 48 |
+
"confidence_cutoff": 0.7193367680537224
|
| 49 |
}
|
| 50 |
},
|
| 51 |
+
"score_mae": 0.5034937052883923,
|
| 52 |
+
"ranked_probability_score": 0.08453946806779199
|
| 53 |
},
|
| 54 |
"tasks": {
|
| 55 |
"agnews": {
|
| 56 |
"n": 80,
|
| 57 |
+
"nll": 0.7864014332399729,
|
| 58 |
"acc": 0.8375,
|
| 59 |
+
"ece": 0.1341834060888242,
|
| 60 |
+
"brier": 0.2712856163517074,
|
| 61 |
+
"mean_conf": 0.9615480819159415,
|
| 62 |
+
"confident_error_rate": 0.1,
|
| 63 |
"coverage_at_0_9": 0.9,
|
| 64 |
+
"accuracy_at_0_9": 0.8888888888888888,
|
| 65 |
+
"confidence_bias": 0.12404808191594152,
|
| 66 |
+
"top_bins": {
|
| 67 |
+
"0.9": {
|
| 68 |
+
"n": 72,
|
| 69 |
+
"errors": 8,
|
| 70 |
+
"error_rate": 0.1111111111111111
|
| 71 |
+
},
|
| 72 |
+
"0.95": {
|
| 73 |
+
"n": 70,
|
| 74 |
+
"errors": 8,
|
| 75 |
+
"error_rate": 0.11428571428571428
|
| 76 |
+
},
|
| 77 |
+
"0.99": {
|
| 78 |
+
"n": 65,
|
| 79 |
+
"errors": 5,
|
| 80 |
+
"error_rate": 0.07692307692307693
|
| 81 |
+
}
|
| 82 |
+
},
|
| 83 |
"selective": {
|
| 84 |
"0.5": {
|
| 85 |
"coverage": 0.5,
|
| 86 |
+
"accuracy": 0.975,
|
| 87 |
+
"confidence_cutoff": 0.9996674420870753
|
| 88 |
},
|
| 89 |
"0.8": {
|
| 90 |
"coverage": 0.8,
|
| 91 |
+
"accuracy": 0.921875,
|
| 92 |
+
"confidence_cutoff": 0.991168628413978
|
| 93 |
}
|
| 94 |
}
|
| 95 |
},
|
| 96 |
"agnews_yn": {
|
| 97 |
"n": 160,
|
| 98 |
+
"nll": 0.28958017568669225,
|
| 99 |
+
"acc": 0.925,
|
| 100 |
+
"ece": 0.07096175529746375,
|
| 101 |
+
"brier": 0.13742498968382408,
|
| 102 |
+
"mean_conf": 0.9793689306266323,
|
| 103 |
"confident_error_rate": 0.0625,
|
| 104 |
+
"coverage_at_0_9": 0.95,
|
| 105 |
+
"accuracy_at_0_9": 0.9342105263157895,
|
| 106 |
+
"confidence_bias": 0.054368930626632284,
|
| 107 |
+
"top_bins": {
|
| 108 |
+
"0.9": {
|
| 109 |
+
"n": 152,
|
| 110 |
+
"errors": 10,
|
| 111 |
+
"error_rate": 0.06578947368421052
|
| 112 |
+
},
|
| 113 |
+
"0.95": {
|
| 114 |
+
"n": 147,
|
| 115 |
+
"errors": 9,
|
| 116 |
+
"error_rate": 0.061224489795918366
|
| 117 |
+
},
|
| 118 |
+
"0.99": {
|
| 119 |
+
"n": 121,
|
| 120 |
+
"errors": 4,
|
| 121 |
+
"error_rate": 0.03305785123966942
|
| 122 |
+
}
|
| 123 |
+
},
|
| 124 |
"selective": {
|
| 125 |
"0.5": {
|
| 126 |
"coverage": 0.5,
|
| 127 |
"accuracy": 1.0,
|
| 128 |
+
"confidence_cutoff": 0.9979671931165864
|
| 129 |
},
|
| 130 |
"0.8": {
|
| 131 |
"coverage": 0.8,
|
| 132 |
"accuracy": 0.96875,
|
| 133 |
+
"confidence_cutoff": 0.985054612159729
|
| 134 |
}
|
| 135 |
}
|
| 136 |
},
|
| 137 |
"amazon": {
|
| 138 |
"n": 80,
|
| 139 |
+
"nll": 1.0211323457815606,
|
| 140 |
+
"acc": 0.525,
|
| 141 |
+
"ece": 0.17043635212876354,
|
| 142 |
+
"brier": 0.5737438123490131,
|
| 143 |
+
"mean_conf": 0.6954363521287635,
|
| 144 |
+
"confident_error_rate": 0.0125,
|
| 145 |
+
"coverage_at_0_9": 0.2,
|
| 146 |
+
"accuracy_at_0_9": 0.9375,
|
| 147 |
+
"confidence_bias": 0.1704363521287635,
|
| 148 |
+
"top_bins": {
|
| 149 |
+
"0.9": {
|
| 150 |
+
"n": 16,
|
| 151 |
+
"errors": 1,
|
| 152 |
+
"error_rate": 0.0625
|
| 153 |
+
},
|
| 154 |
+
"0.95": {
|
| 155 |
+
"n": 7,
|
| 156 |
+
"errors": 0,
|
| 157 |
+
"error_rate": 0.0
|
| 158 |
+
},
|
| 159 |
+
"0.99": {
|
| 160 |
+
"n": 1,
|
| 161 |
+
"errors": 0,
|
| 162 |
+
"error_rate": 0.0
|
| 163 |
+
}
|
| 164 |
+
},
|
| 165 |
"selective": {
|
| 166 |
"0.5": {
|
| 167 |
"coverage": 0.5,
|
| 168 |
+
"accuracy": 0.7,
|
| 169 |
+
"confidence_cutoff": 0.6880432373732731
|
| 170 |
},
|
| 171 |
"0.8": {
|
| 172 |
"coverage": 0.8,
|
| 173 |
+
"accuracy": 0.578125,
|
| 174 |
+
"confidence_cutoff": 0.5195248263669912
|
| 175 |
}
|
| 176 |
},
|
| 177 |
+
"score_mae": 0.5581328526284086,
|
| 178 |
+
"ranked_probability_score": 0.09321312844170443
|
| 179 |
},
|
| 180 |
"banking77": {
|
| 181 |
"n": 80,
|
| 182 |
+
"nll": 0.6011434192795363,
|
| 183 |
+
"acc": 0.85,
|
| 184 |
+
"ece": 0.11207010433881469,
|
| 185 |
+
"brier": 0.24538362925906454,
|
| 186 |
+
"mean_conf": 0.9050673931399228,
|
| 187 |
+
"confident_error_rate": 0.025,
|
| 188 |
+
"coverage_at_0_9": 0.725,
|
| 189 |
+
"accuracy_at_0_9": 0.9655172413793104,
|
| 190 |
+
"confidence_bias": 0.055067393139922816,
|
| 191 |
+
"top_bins": {
|
| 192 |
+
"0.9": {
|
| 193 |
+
"n": 58,
|
| 194 |
+
"errors": 2,
|
| 195 |
+
"error_rate": 0.034482758620689655
|
| 196 |
+
},
|
| 197 |
+
"0.95": {
|
| 198 |
+
"n": 51,
|
| 199 |
+
"errors": 1,
|
| 200 |
+
"error_rate": 0.0196078431372549
|
| 201 |
+
},
|
| 202 |
+
"0.99": {
|
| 203 |
+
"n": 45,
|
| 204 |
+
"errors": 1,
|
| 205 |
+
"error_rate": 0.022222222222222223
|
| 206 |
+
}
|
| 207 |
+
},
|
| 208 |
"selective": {
|
| 209 |
"0.5": {
|
| 210 |
"coverage": 0.5,
|
| 211 |
"accuracy": 0.975,
|
| 212 |
+
"confidence_cutoff": 0.9951459634189641
|
| 213 |
},
|
| 214 |
"0.8": {
|
| 215 |
"coverage": 0.8,
|
| 216 |
+
"accuracy": 0.9375,
|
| 217 |
+
"confidence_cutoff": 0.8619084283796796
|
| 218 |
}
|
| 219 |
}
|
| 220 |
},
|
| 221 |
"boolq": {
|
| 222 |
"n": 80,
|
| 223 |
+
"nll": 0.7926256706822462,
|
| 224 |
+
"acc": 0.6875,
|
| 225 |
+
"ece": 0.21899707326050627,
|
| 226 |
+
"brier": 0.47274578916596577,
|
| 227 |
+
"mean_conf": 0.9064970732605063,
|
| 228 |
+
"confident_error_rate": 0.15,
|
| 229 |
+
"coverage_at_0_9": 0.7125,
|
| 230 |
+
"accuracy_at_0_9": 0.7894736842105263,
|
| 231 |
+
"confidence_bias": 0.21899707326050633,
|
| 232 |
+
"top_bins": {
|
| 233 |
+
"0.9": {
|
| 234 |
+
"n": 57,
|
| 235 |
+
"errors": 12,
|
| 236 |
+
"error_rate": 0.21052631578947367
|
| 237 |
+
},
|
| 238 |
+
"0.95": {
|
| 239 |
+
"n": 42,
|
| 240 |
+
"errors": 7,
|
| 241 |
+
"error_rate": 0.16666666666666666
|
| 242 |
+
},
|
| 243 |
+
"0.99": {
|
| 244 |
+
"n": 17,
|
| 245 |
+
"errors": 1,
|
| 246 |
+
"error_rate": 0.058823529411764705
|
| 247 |
+
}
|
| 248 |
+
},
|
| 249 |
"selective": {
|
| 250 |
"0.5": {
|
| 251 |
"coverage": 0.5,
|
| 252 |
+
"accuracy": 0.85,
|
| 253 |
+
"confidence_cutoff": 0.959244272692001
|
| 254 |
},
|
| 255 |
"0.8": {
|
| 256 |
"coverage": 0.8,
|
| 257 |
+
"accuracy": 0.765625,
|
| 258 |
+
"confidence_cutoff": 0.851481020450592
|
| 259 |
}
|
| 260 |
}
|
| 261 |
},
|
| 262 |
"composition_atom": {
|
| 263 |
"n": 16,
|
| 264 |
+
"nll": 0.17544967350851692,
|
| 265 |
"acc": 1.0,
|
| 266 |
+
"ece": 0.15428788409075606,
|
| 267 |
+
"brier": 0.0683359851651861,
|
| 268 |
+
"mean_conf": 0.845712115909244,
|
| 269 |
"confident_error_rate": 0.0,
|
| 270 |
+
"coverage_at_0_9": 0.375,
|
| 271 |
"accuracy_at_0_9": 1.0,
|
| 272 |
+
"confidence_bias": -0.154287884090756,
|
| 273 |
+
"top_bins": {
|
| 274 |
+
"0.9": {
|
| 275 |
+
"n": 6,
|
| 276 |
+
"errors": 0,
|
| 277 |
+
"error_rate": 0.0
|
| 278 |
+
},
|
| 279 |
+
"0.95": {
|
| 280 |
+
"n": 1,
|
| 281 |
+
"errors": 0,
|
| 282 |
+
"error_rate": 0.0
|
| 283 |
+
},
|
| 284 |
+
"0.99": {
|
| 285 |
+
"n": 0,
|
| 286 |
+
"errors": 0,
|
| 287 |
+
"error_rate": null
|
| 288 |
+
}
|
| 289 |
+
},
|
| 290 |
"selective": {
|
| 291 |
"0.5": {
|
| 292 |
"coverage": 0.5,
|
| 293 |
"accuracy": 1.0,
|
| 294 |
+
"confidence_cutoff": 0.892630225859271
|
| 295 |
},
|
| 296 |
"0.8": {
|
| 297 |
"coverage": 0.8125,
|
| 298 |
"accuracy": 1.0,
|
| 299 |
+
"confidence_cutoff": 0.7610612142402295
|
| 300 |
}
|
| 301 |
}
|
| 302 |
},
|
| 303 |
"composition_conditional": {
|
| 304 |
"n": 16,
|
| 305 |
+
"nll": 1.3744812880136639,
|
| 306 |
+
"acc": 0.4375,
|
| 307 |
+
"ece": 0.35408401289309854,
|
| 308 |
+
"brier": 0.7696203733778755,
|
| 309 |
+
"mean_conf": 0.758628032911951,
|
| 310 |
+
"confident_error_rate": 0.1875,
|
| 311 |
"coverage_at_0_9": 0.25,
|
| 312 |
+
"accuracy_at_0_9": 0.25,
|
| 313 |
+
"confidence_bias": 0.321128032911951,
|
| 314 |
+
"top_bins": {
|
| 315 |
+
"0.9": {
|
| 316 |
+
"n": 4,
|
| 317 |
+
"errors": 3,
|
| 318 |
+
"error_rate": 0.75
|
| 319 |
+
},
|
| 320 |
+
"0.95": {
|
| 321 |
+
"n": 4,
|
| 322 |
+
"errors": 3,
|
| 323 |
+
"error_rate": 0.75
|
| 324 |
+
},
|
| 325 |
+
"0.99": {
|
| 326 |
+
"n": 0,
|
| 327 |
+
"errors": 0,
|
| 328 |
+
"error_rate": null
|
| 329 |
+
}
|
| 330 |
+
},
|
| 331 |
"selective": {
|
| 332 |
"0.5": {
|
| 333 |
"coverage": 0.5,
|
| 334 |
+
"accuracy": 0.5,
|
| 335 |
+
"confidence_cutoff": 0.7288850384286069
|
| 336 |
},
|
| 337 |
"0.8": {
|
| 338 |
"coverage": 0.8125,
|
| 339 |
+
"accuracy": 0.46153846153846156,
|
| 340 |
+
"confidence_cutoff": 0.6750231780104882
|
| 341 |
}
|
| 342 |
}
|
| 343 |
},
|
| 344 |
"composition_conjunction": {
|
| 345 |
"n": 16,
|
| 346 |
+
"nll": 0.6403887325711459,
|
| 347 |
+
"acc": 0.625,
|
| 348 |
+
"ece": 0.25641329128434887,
|
| 349 |
+
"brier": 0.46730044476926313,
|
| 350 |
+
"mean_conf": 0.7820054162647625,
|
| 351 |
"confident_error_rate": 0.0,
|
| 352 |
+
"coverage_at_0_9": 0.25,
|
| 353 |
"accuracy_at_0_9": 1.0,
|
| 354 |
+
"confidence_bias": 0.1570054162647625,
|
| 355 |
+
"top_bins": {
|
| 356 |
+
"0.9": {
|
| 357 |
+
"n": 4,
|
| 358 |
+
"errors": 0,
|
| 359 |
+
"error_rate": 0.0
|
| 360 |
+
},
|
| 361 |
+
"0.95": {
|
| 362 |
+
"n": 4,
|
| 363 |
+
"errors": 0,
|
| 364 |
+
"error_rate": 0.0
|
| 365 |
+
},
|
| 366 |
+
"0.99": {
|
| 367 |
+
"n": 0,
|
| 368 |
+
"errors": 0,
|
| 369 |
+
"error_rate": null
|
| 370 |
+
}
|
| 371 |
+
},
|
| 372 |
"selective": {
|
| 373 |
"0.5": {
|
| 374 |
+
"coverage": 0.5,
|
| 375 |
+
"accuracy": 0.75,
|
| 376 |
+
"confidence_cutoff": 0.8147355196664405
|
| 377 |
},
|
| 378 |
"0.8": {
|
| 379 |
+
"coverage": 0.875,
|
| 380 |
+
"accuracy": 0.6428571428571429,
|
| 381 |
+
"confidence_cutoff": 0.5991041660308838
|
| 382 |
}
|
| 383 |
}
|
| 384 |
},
|
| 385 |
"composition_disjunction": {
|
| 386 |
"n": 16,
|
| 387 |
+
"nll": 0.5552568033665892,
|
| 388 |
+
"acc": 0.875,
|
| 389 |
+
"ece": 0.31967711321382625,
|
| 390 |
+
"brier": 0.367756998202627,
|
| 391 |
+
"mean_conf": 0.6459422496768941,
|
| 392 |
"confident_error_rate": 0.0,
|
| 393 |
+
"coverage_at_0_9": 0.0,
|
| 394 |
+
"accuracy_at_0_9": null,
|
| 395 |
+
"confidence_bias": -0.22905775032310594,
|
| 396 |
+
"top_bins": {
|
| 397 |
+
"0.9": {
|
| 398 |
+
"n": 0,
|
| 399 |
+
"errors": 0,
|
| 400 |
+
"error_rate": null
|
| 401 |
+
},
|
| 402 |
+
"0.95": {
|
| 403 |
+
"n": 0,
|
| 404 |
+
"errors": 0,
|
| 405 |
+
"error_rate": null
|
| 406 |
+
},
|
| 407 |
+
"0.99": {
|
| 408 |
+
"n": 0,
|
| 409 |
+
"errors": 0,
|
| 410 |
+
"error_rate": null
|
| 411 |
+
}
|
| 412 |
+
},
|
| 413 |
"selective": {
|
| 414 |
"0.5": {
|
| 415 |
+
"coverage": 0.5625,
|
| 416 |
+
"accuracy": 0.7777777777777778,
|
| 417 |
+
"confidence_cutoff": 0.6318208167499547
|
| 418 |
},
|
| 419 |
"0.8": {
|
| 420 |
"coverage": 0.8125,
|
| 421 |
+
"accuracy": 0.8461538461538461,
|
| 422 |
+
"confidence_cutoff": 0.5909334421157837
|
| 423 |
}
|
| 424 |
}
|
| 425 |
},
|
| 426 |
"composition_exception": {
|
| 427 |
"n": 16,
|
| 428 |
+
"nll": 0.7163920274103871,
|
| 429 |
+
"acc": 0.5,
|
| 430 |
+
"ece": 0.19532563513541545,
|
| 431 |
+
"brier": 0.5249127733122687,
|
| 432 |
+
"mean_conf": 0.6791466190343678,
|
| 433 |
"confident_error_rate": 0.0,
|
| 434 |
"coverage_at_0_9": 0.0,
|
| 435 |
"accuracy_at_0_9": null,
|
| 436 |
+
"confidence_bias": 0.1791466190343678,
|
| 437 |
+
"top_bins": {
|
| 438 |
+
"0.9": {
|
| 439 |
+
"n": 0,
|
| 440 |
+
"errors": 0,
|
| 441 |
+
"error_rate": null
|
| 442 |
+
},
|
| 443 |
+
"0.95": {
|
| 444 |
+
"n": 0,
|
| 445 |
+
"errors": 0,
|
| 446 |
+
"error_rate": null
|
| 447 |
+
},
|
| 448 |
+
"0.99": {
|
| 449 |
+
"n": 0,
|
| 450 |
+
"errors": 0,
|
| 451 |
+
"error_rate": null
|
| 452 |
+
}
|
| 453 |
+
},
|
| 454 |
"selective": {
|
| 455 |
"0.5": {
|
| 456 |
"coverage": 0.5,
|
| 457 |
"accuracy": 0.625,
|
| 458 |
+
"confidence_cutoff": 0.7307368734698556
|
| 459 |
},
|
| 460 |
"0.8": {
|
| 461 |
+
"coverage": 0.8125,
|
| 462 |
+
"accuracy": 0.5384615384615384,
|
| 463 |
+
"confidence_cutoff": 0.613737243823327
|
| 464 |
}
|
| 465 |
}
|
| 466 |
},
|
| 467 |
"composition_negation": {
|
| 468 |
"n": 16,
|
| 469 |
+
"nll": 0.8374503395176269,
|
| 470 |
+
"acc": 0.5625,
|
| 471 |
+
"ece": 0.2493486695490849,
|
| 472 |
+
"brier": 0.5606965580917005,
|
| 473 |
+
"mean_conf": 0.7918363963811793,
|
| 474 |
+
"confident_error_rate": 0.125,
|
| 475 |
+
"coverage_at_0_9": 0.3125,
|
| 476 |
+
"accuracy_at_0_9": 0.6,
|
| 477 |
+
"confidence_bias": 0.2293363963811793,
|
| 478 |
+
"top_bins": {
|
| 479 |
+
"0.9": {
|
| 480 |
+
"n": 5,
|
| 481 |
+
"errors": 2,
|
| 482 |
+
"error_rate": 0.4
|
| 483 |
+
},
|
| 484 |
+
"0.95": {
|
| 485 |
+
"n": 3,
|
| 486 |
+
"errors": 0,
|
| 487 |
+
"error_rate": 0.0
|
| 488 |
+
},
|
| 489 |
+
"0.99": {
|
| 490 |
+
"n": 0,
|
| 491 |
+
"errors": 0,
|
| 492 |
+
"error_rate": null
|
| 493 |
+
}
|
| 494 |
+
},
|
| 495 |
"selective": {
|
| 496 |
"0.5": {
|
| 497 |
"coverage": 0.5,
|
| 498 |
+
"accuracy": 0.625,
|
| 499 |
+
"confidence_cutoff": 0.7943201542746998
|
| 500 |
},
|
| 501 |
"0.8": {
|
| 502 |
"coverage": 0.875,
|
| 503 |
+
"accuracy": 0.5714285714285714,
|
| 504 |
+
"confidence_cutoff": 0.5952281114595003
|
| 505 |
}
|
| 506 |
}
|
| 507 |
},
|
| 508 |
"composition_nested_and": {
|
| 509 |
"n": 16,
|
| 510 |
+
"nll": 0.7179540381326612,
|
| 511 |
+
"acc": 0.625,
|
| 512 |
+
"ece": 0.2572182318076006,
|
| 513 |
+
"brier": 0.524083497689676,
|
| 514 |
+
"mean_conf": 0.6047320565187797,
|
| 515 |
"confident_error_rate": 0.0,
|
| 516 |
"coverage_at_0_9": 0.0,
|
| 517 |
"accuracy_at_0_9": null,
|
| 518 |
+
"confidence_bias": -0.020267943481220296,
|
| 519 |
+
"top_bins": {
|
| 520 |
+
"0.9": {
|
| 521 |
+
"n": 0,
|
| 522 |
+
"errors": 0,
|
| 523 |
+
"error_rate": null
|
| 524 |
+
},
|
| 525 |
+
"0.95": {
|
| 526 |
+
"n": 0,
|
| 527 |
+
"errors": 0,
|
| 528 |
+
"error_rate": null
|
| 529 |
+
},
|
| 530 |
+
"0.99": {
|
| 531 |
+
"n": 0,
|
| 532 |
+
"errors": 0,
|
| 533 |
+
"error_rate": null
|
| 534 |
+
}
|
| 535 |
+
},
|
| 536 |
"selective": {
|
| 537 |
"0.5": {
|
| 538 |
+
"coverage": 0.5,
|
| 539 |
+
"accuracy": 0.5,
|
| 540 |
+
"confidence_cutoff": 0.6389932442011462
|
| 541 |
},
|
| 542 |
"0.8": {
|
| 543 |
"coverage": 0.8125,
|
| 544 |
+
"accuracy": 0.5384615384615384,
|
| 545 |
+
"confidence_cutoff": 0.5100217613584328
|
| 546 |
}
|
| 547 |
}
|
| 548 |
},
|
| 549 |
"composition_nested_or": {
|
| 550 |
"n": 16,
|
| 551 |
+
"nll": 0.9453449583955086,
|
| 552 |
+
"acc": 0.375,
|
| 553 |
+
"ece": 0.40631057670363724,
|
| 554 |
+
"brier": 0.7081135645715555,
|
| 555 |
+
"mean_conf": 0.6947151587067877,
|
| 556 |
"confident_error_rate": 0.0,
|
| 557 |
"coverage_at_0_9": 0.0625,
|
| 558 |
"accuracy_at_0_9": 1.0,
|
| 559 |
+
"confidence_bias": 0.3197151587067877,
|
| 560 |
+
"top_bins": {
|
| 561 |
+
"0.9": {
|
| 562 |
+
"n": 1,
|
| 563 |
+
"errors": 0,
|
| 564 |
+
"error_rate": 0.0
|
| 565 |
+
},
|
| 566 |
+
"0.95": {
|
| 567 |
+
"n": 0,
|
| 568 |
+
"errors": 0,
|
| 569 |
+
"error_rate": null
|
| 570 |
+
},
|
| 571 |
+
"0.99": {
|
| 572 |
+
"n": 0,
|
| 573 |
+
"errors": 0,
|
| 574 |
+
"error_rate": null
|
| 575 |
+
}
|
| 576 |
+
},
|
| 577 |
"selective": {
|
| 578 |
"0.5": {
|
| 579 |
+
"coverage": 0.5625,
|
| 580 |
+
"accuracy": 0.3333333333333333,
|
| 581 |
+
"confidence_cutoff": 0.6655829350476582
|
| 582 |
},
|
| 583 |
"0.8": {
|
| 584 |
+
"coverage": 0.8125,
|
| 585 |
+
"accuracy": 0.38461538461538464,
|
| 586 |
+
"confidence_cutoff": 0.6321162989104897
|
| 587 |
}
|
| 588 |
}
|
| 589 |
},
|
| 590 |
"contrastive_age_eligibility": {
|
| 591 |
"n": 24,
|
| 592 |
+
"nll": 0.00026480341349052907,
|
| 593 |
"acc": 1.0,
|
| 594 |
+
"ece": 0.0002647253283631912,
|
| 595 |
+
"brier": 3.121045316073063e-07,
|
| 596 |
+
"mean_conf": 0.9997352746716368,
|
| 597 |
"confident_error_rate": 0.0,
|
| 598 |
"coverage_at_0_9": 1.0,
|
| 599 |
"accuracy_at_0_9": 1.0,
|
| 600 |
+
"confidence_bias": -0.0002647253283631912,
|
| 601 |
+
"top_bins": {
|
| 602 |
+
"0.9": {
|
| 603 |
+
"n": 24,
|
| 604 |
+
"errors": 0,
|
| 605 |
+
"error_rate": 0.0
|
| 606 |
+
},
|
| 607 |
+
"0.95": {
|
| 608 |
+
"n": 24,
|
| 609 |
+
"errors": 0,
|
| 610 |
+
"error_rate": 0.0
|
| 611 |
+
},
|
| 612 |
+
"0.99": {
|
| 613 |
+
"n": 24,
|
| 614 |
+
"errors": 0,
|
| 615 |
+
"error_rate": 0.0
|
| 616 |
+
}
|
| 617 |
+
},
|
| 618 |
"selective": {
|
| 619 |
"0.5": {
|
| 620 |
"coverage": 0.5,
|
| 621 |
"accuracy": 1.0,
|
| 622 |
+
"confidence_cutoff": 0.9998386801646693
|
| 623 |
},
|
| 624 |
"0.8": {
|
| 625 |
"coverage": 0.8333333333333334,
|
| 626 |
"accuracy": 1.0,
|
| 627 |
+
"confidence_cutoff": 0.9996550610048963
|
| 628 |
}
|
| 629 |
}
|
| 630 |
},
|
| 631 |
"contrastive_quantity_limit": {
|
| 632 |
"n": 24,
|
| 633 |
+
"nll": 0.13676344280632033,
|
| 634 |
+
"acc": 1.0,
|
| 635 |
+
"ece": 0.10861937817026238,
|
| 636 |
+
"brier": 0.07985279266022248,
|
| 637 |
+
"mean_conf": 0.8913806218297377,
|
| 638 |
"confident_error_rate": 0.0,
|
| 639 |
"coverage_at_0_9": 0.6666666666666666,
|
| 640 |
"accuracy_at_0_9": 1.0,
|
| 641 |
+
"confidence_bias": -0.10861937817026235,
|
| 642 |
+
"top_bins": {
|
| 643 |
+
"0.9": {
|
| 644 |
+
"n": 16,
|
| 645 |
+
"errors": 0,
|
| 646 |
+
"error_rate": 0.0
|
| 647 |
+
},
|
| 648 |
+
"0.95": {
|
| 649 |
+
"n": 15,
|
| 650 |
+
"errors": 0,
|
| 651 |
+
"error_rate": 0.0
|
| 652 |
+
},
|
| 653 |
+
"0.99": {
|
| 654 |
+
"n": 14,
|
| 655 |
+
"errors": 0,
|
| 656 |
+
"error_rate": 0.0
|
| 657 |
+
}
|
| 658 |
+
},
|
| 659 |
"selective": {
|
| 660 |
"0.5": {
|
| 661 |
"coverage": 0.5,
|
| 662 |
"accuracy": 1.0,
|
| 663 |
+
"confidence_cutoff": 0.9999416002075856
|
| 664 |
},
|
| 665 |
"0.8": {
|
| 666 |
"coverage": 0.8333333333333334,
|
| 667 |
"accuracy": 1.0,
|
| 668 |
+
"confidence_cutoff": 0.6180640991789241
|
| 669 |
}
|
| 670 |
}
|
| 671 |
},
|
| 672 |
"contrastive_return_window": {
|
| 673 |
"n": 24,
|
| 674 |
+
"nll": 0.5040842387089451,
|
| 675 |
"acc": 0.75,
|
| 676 |
+
"ece": 0.06457848152805973,
|
| 677 |
+
"brier": 0.3360148460720538,
|
| 678 |
+
"mean_conf": 0.7234365399748733,
|
| 679 |
"confident_error_rate": 0.0,
|
| 680 |
+
"coverage_at_0_9": 0.08333333333333333,
|
| 681 |
+
"accuracy_at_0_9": 1.0,
|
| 682 |
+
"confidence_bias": -0.02656346002512666,
|
| 683 |
+
"top_bins": {
|
| 684 |
+
"0.9": {
|
| 685 |
+
"n": 2,
|
| 686 |
+
"errors": 0,
|
| 687 |
+
"error_rate": 0.0
|
| 688 |
+
},
|
| 689 |
+
"0.95": {
|
| 690 |
+
"n": 0,
|
| 691 |
+
"errors": 0,
|
| 692 |
+
"error_rate": null
|
| 693 |
+
},
|
| 694 |
+
"0.99": {
|
| 695 |
+
"n": 0,
|
| 696 |
+
"errors": 0,
|
| 697 |
+
"error_rate": null
|
| 698 |
+
}
|
| 699 |
+
},
|
| 700 |
"selective": {
|
| 701 |
"0.5": {
|
| 702 |
"coverage": 0.5,
|
| 703 |
+
"accuracy": 0.9166666666666666,
|
| 704 |
+
"confidence_cutoff": 0.7820771802709806
|
| 705 |
},
|
| 706 |
"0.8": {
|
| 707 |
"coverage": 0.8333333333333334,
|
| 708 |
+
"accuracy": 0.8,
|
| 709 |
+
"confidence_cutoff": 0.5465864105995419
|
| 710 |
}
|
| 711 |
}
|
| 712 |
},
|
| 713 |
"contrastive_spend_threshold": {
|
| 714 |
"n": 24,
|
| 715 |
+
"nll": 5.5880089314245744e-06,
|
| 716 |
"acc": 1.0,
|
| 717 |
+
"ece": 5.587984092891851e-06,
|
| 718 |
+
"brier": 7.903295619923303e-11,
|
| 719 |
+
"mean_conf": 0.9999944120159071,
|
| 720 |
"confident_error_rate": 0.0,
|
| 721 |
"coverage_at_0_9": 1.0,
|
| 722 |
"accuracy_at_0_9": 1.0,
|
| 723 |
+
"confidence_bias": -5.587984092891851e-06,
|
| 724 |
+
"top_bins": {
|
| 725 |
+
"0.9": {
|
| 726 |
+
"n": 24,
|
| 727 |
+
"errors": 0,
|
| 728 |
+
"error_rate": 0.0
|
| 729 |
+
},
|
| 730 |
+
"0.95": {
|
| 731 |
+
"n": 24,
|
| 732 |
+
"errors": 0,
|
| 733 |
+
"error_rate": 0.0
|
| 734 |
+
},
|
| 735 |
+
"0.99": {
|
| 736 |
+
"n": 24,
|
| 737 |
+
"errors": 0,
|
| 738 |
+
"error_rate": 0.0
|
| 739 |
+
}
|
| 740 |
+
},
|
| 741 |
"selective": {
|
| 742 |
"0.5": {
|
| 743 |
"coverage": 0.5,
|
| 744 |
"accuracy": 1.0,
|
| 745 |
+
"confidence_cutoff": 0.9999954891846687
|
| 746 |
},
|
| 747 |
"0.8": {
|
| 748 |
"coverage": 0.8333333333333334,
|
| 749 |
"accuracy": 1.0,
|
| 750 |
+
"confidence_cutoff": 0.9999925488462089
|
| 751 |
}
|
| 752 |
}
|
| 753 |
},
|
| 754 |
"dbpedia14": {
|
| 755 |
"n": 80,
|
| 756 |
+
"nll": 0.3111186530104228,
|
| 757 |
"acc": 0.95,
|
| 758 |
+
"ece": 0.03998458072307311,
|
| 759 |
+
"brier": 0.08129993082712443,
|
| 760 |
+
"mean_conf": 0.9899845807230727,
|
| 761 |
+
"confident_error_rate": 0.0125,
|
| 762 |
+
"coverage_at_0_9": 0.9625,
|
| 763 |
+
"accuracy_at_0_9": 0.987012987012987,
|
| 764 |
+
"confidence_bias": 0.03998458072307276,
|
| 765 |
+
"top_bins": {
|
| 766 |
+
"0.9": {
|
| 767 |
+
"n": 77,
|
| 768 |
+
"errors": 1,
|
| 769 |
+
"error_rate": 0.012987012987012988
|
| 770 |
+
},
|
| 771 |
+
"0.95": {
|
| 772 |
+
"n": 76,
|
| 773 |
+
"errors": 1,
|
| 774 |
+
"error_rate": 0.013157894736842105
|
| 775 |
+
},
|
| 776 |
+
"0.99": {
|
| 777 |
+
"n": 76,
|
| 778 |
+
"errors": 1,
|
| 779 |
+
"error_rate": 0.013157894736842105
|
| 780 |
+
}
|
| 781 |
+
},
|
| 782 |
"selective": {
|
| 783 |
"0.5": {
|
| 784 |
"coverage": 0.5,
|
| 785 |
+
"accuracy": 0.975,
|
| 786 |
+
"confidence_cutoff": 0.9999996225650949
|
| 787 |
},
|
| 788 |
"0.8": {
|
| 789 |
"coverage": 0.8,
|
| 790 |
"accuracy": 0.984375,
|
| 791 |
+
"confidence_cutoff": 0.9999897353322881
|
| 792 |
}
|
| 793 |
}
|
| 794 |
},
|
| 795 |
"imdb": {
|
| 796 |
"n": 80,
|
| 797 |
+
"nll": 0.4484578250582059,
|
| 798 |
+
"acc": 0.8875,
|
| 799 |
+
"ece": 0.08444057248938229,
|
| 800 |
+
"brier": 0.1911023370422809,
|
| 801 |
+
"mean_conf": 0.9719405724893821,
|
| 802 |
+
"confident_error_rate": 0.075,
|
| 803 |
+
"coverage_at_0_9": 0.9,
|
| 804 |
+
"accuracy_at_0_9": 0.9166666666666666,
|
| 805 |
+
"confidence_bias": 0.08444057248938219,
|
| 806 |
+
"top_bins": {
|
| 807 |
+
"0.9": {
|
| 808 |
+
"n": 72,
|
| 809 |
+
"errors": 6,
|
| 810 |
+
"error_rate": 0.08333333333333333
|
| 811 |
+
},
|
| 812 |
+
"0.95": {
|
| 813 |
+
"n": 68,
|
| 814 |
+
"errors": 4,
|
| 815 |
+
"error_rate": 0.058823529411764705
|
| 816 |
+
},
|
| 817 |
+
"0.99": {
|
| 818 |
+
"n": 61,
|
| 819 |
+
"errors": 3,
|
| 820 |
+
"error_rate": 0.04918032786885246
|
| 821 |
+
}
|
| 822 |
+
},
|
| 823 |
"selective": {
|
| 824 |
"0.5": {
|
| 825 |
"coverage": 0.5,
|
| 826 |
"accuracy": 0.975,
|
| 827 |
+
"confidence_cutoff": 0.9993848541794427
|
| 828 |
},
|
| 829 |
"0.8": {
|
| 830 |
"coverage": 0.8,
|
| 831 |
+
"accuracy": 0.9375,
|
| 832 |
+
"confidence_cutoff": 0.9861624817616435
|
| 833 |
}
|
| 834 |
}
|
| 835 |
},
|
| 836 |
"mnli": {
|
| 837 |
"n": 80,
|
| 838 |
+
"nll": 1.1887200313270914,
|
| 839 |
+
"acc": 0.775,
|
| 840 |
+
"ece": 0.19684266390281957,
|
| 841 |
+
"brier": 0.3819567555781473,
|
| 842 |
+
"mean_conf": 0.9518911079981226,
|
| 843 |
+
"confident_error_rate": 0.1625,
|
| 844 |
"coverage_at_0_9": 0.8625,
|
| 845 |
+
"accuracy_at_0_9": 0.8115942028985508,
|
| 846 |
+
"confidence_bias": 0.17689110799812258,
|
| 847 |
+
"top_bins": {
|
| 848 |
+
"0.9": {
|
| 849 |
+
"n": 69,
|
| 850 |
+
"errors": 13,
|
| 851 |
+
"error_rate": 0.18840579710144928
|
| 852 |
+
},
|
| 853 |
+
"0.95": {
|
| 854 |
+
"n": 65,
|
| 855 |
+
"errors": 12,
|
| 856 |
+
"error_rate": 0.18461538461538463
|
| 857 |
+
},
|
| 858 |
+
"0.99": {
|
| 859 |
+
"n": 57,
|
| 860 |
+
"errors": 9,
|
| 861 |
+
"error_rate": 0.15789473684210525
|
| 862 |
+
}
|
| 863 |
+
},
|
| 864 |
"selective": {
|
| 865 |
"0.5": {
|
| 866 |
"coverage": 0.5,
|
| 867 |
+
"accuracy": 0.875,
|
| 868 |
+
"confidence_cutoff": 0.997983986925514
|
| 869 |
},
|
| 870 |
"0.8": {
|
| 871 |
"coverage": 0.8,
|
| 872 |
+
"accuracy": 0.8125,
|
| 873 |
+
"confidence_cutoff": 0.9592310105623559
|
| 874 |
}
|
| 875 |
}
|
| 876 |
},
|
| 877 |
"sst5": {
|
| 878 |
"n": 80,
|
| 879 |
+
"nll": 1.156142626053144,
|
| 880 |
+
"acc": 0.6125,
|
| 881 |
+
"ece": 0.12302780677382641,
|
| 882 |
+
"brier": 0.5836811013115014,
|
| 883 |
+
"mean_conf": 0.7002388293788752,
|
| 884 |
+
"confident_error_rate": 0.0375,
|
| 885 |
+
"coverage_at_0_9": 0.05,
|
| 886 |
+
"accuracy_at_0_9": 0.25,
|
| 887 |
+
"confidence_bias": 0.08773882937887512,
|
| 888 |
+
"top_bins": {
|
| 889 |
+
"0.9": {
|
| 890 |
+
"n": 4,
|
| 891 |
+
"errors": 3,
|
| 892 |
+
"error_rate": 0.75
|
| 893 |
+
},
|
| 894 |
+
"0.95": {
|
| 895 |
+
"n": 1,
|
| 896 |
+
"errors": 0,
|
| 897 |
+
"error_rate": 0.0
|
| 898 |
+
},
|
| 899 |
+
"0.99": {
|
| 900 |
+
"n": 0,
|
| 901 |
+
"errors": 0,
|
| 902 |
+
"error_rate": null
|
| 903 |
+
}
|
| 904 |
+
},
|
| 905 |
"selective": {
|
| 906 |
"0.5": {
|
| 907 |
"coverage": 0.5,
|
| 908 |
+
"accuracy": 0.625,
|
| 909 |
+
"confidence_cutoff": 0.7093760579306325
|
| 910 |
},
|
| 911 |
"0.8": {
|
| 912 |
"coverage": 0.8,
|
| 913 |
+
"accuracy": 0.609375,
|
| 914 |
+
"confidence_cutoff": 0.5718410142922629
|
| 915 |
}
|
| 916 |
},
|
| 917 |
+
"score_mae": 0.5252797338208348,
|
| 918 |
+
"ranked_probability_score": 0.0888736425195075
|
| 919 |
},
|
| 920 |
"trec": {
|
| 921 |
"n": 80,
|
| 922 |
+
"nll": 0.4675955248386523,
|
| 923 |
+
"acc": 0.9625,
|
| 924 |
+
"ece": 0.04368193449615696,
|
| 925 |
+
"brier": 0.08067220345789725,
|
| 926 |
+
"mean_conf": 0.9919145370476684,
|
| 927 |
"confident_error_rate": 0.0375,
|
| 928 |
+
"coverage_at_0_9": 0.975,
|
| 929 |
+
"accuracy_at_0_9": 0.9615384615384616,
|
| 930 |
+
"confidence_bias": 0.029414537047668388,
|
| 931 |
+
"top_bins": {
|
| 932 |
+
"0.9": {
|
| 933 |
+
"n": 78,
|
| 934 |
+
"errors": 3,
|
| 935 |
+
"error_rate": 0.038461538461538464
|
| 936 |
+
},
|
| 937 |
+
"0.95": {
|
| 938 |
+
"n": 78,
|
| 939 |
+
"errors": 3,
|
| 940 |
+
"error_rate": 0.038461538461538464
|
| 941 |
+
},
|
| 942 |
+
"0.99": {
|
| 943 |
+
"n": 76,
|
| 944 |
+
"errors": 3,
|
| 945 |
+
"error_rate": 0.039473684210526314
|
| 946 |
+
}
|
| 947 |
+
},
|
| 948 |
"selective": {
|
| 949 |
"0.5": {
|
| 950 |
"coverage": 0.5,
|
| 951 |
+
"accuracy": 0.975,
|
| 952 |
+
"confidence_cutoff": 0.9999972575307172
|
| 953 |
},
|
| 954 |
"0.8": {
|
| 955 |
"coverage": 0.8,
|
| 956 |
+
"accuracy": 0.953125,
|
| 957 |
+
"confidence_cutoff": 0.999901472932086
|
| 958 |
}
|
| 959 |
}
|
| 960 |
},
|
| 961 |
"yelp": {
|
| 962 |
"n": 80,
|
| 963 |
+
"nll": 0.9023602009894471,
|
| 964 |
"acc": 0.6875,
|
| 965 |
+
"ece": 0.08356422024067008,
|
| 966 |
+
"brier": 0.4622395750295082,
|
| 967 |
+
"mean_conf": 0.7560037403007092,
|
| 968 |
"confident_error_rate": 0.025,
|
| 969 |
+
"coverage_at_0_9": 0.225,
|
| 970 |
+
"accuracy_at_0_9": 0.8888888888888888,
|
| 971 |
+
"confidence_bias": 0.06850374030070916,
|
| 972 |
+
"top_bins": {
|
| 973 |
+
"0.9": {
|
| 974 |
+
"n": 18,
|
| 975 |
+
"errors": 2,
|
| 976 |
+
"error_rate": 0.1111111111111111
|
| 977 |
+
},
|
| 978 |
+
"0.95": {
|
| 979 |
+
"n": 10,
|
| 980 |
+
"errors": 0,
|
| 981 |
+
"error_rate": 0.0
|
| 982 |
+
},
|
| 983 |
+
"0.99": {
|
| 984 |
+
"n": 1,
|
| 985 |
+
"errors": 0,
|
| 986 |
+
"error_rate": 0.0
|
| 987 |
+
}
|
| 988 |
+
},
|
| 989 |
"selective": {
|
| 990 |
"0.5": {
|
| 991 |
"coverage": 0.5,
|
| 992 |
+
"accuracy": 0.825,
|
| 993 |
+
"confidence_cutoff": 0.7533232786063864
|
| 994 |
},
|
| 995 |
"0.8": {
|
| 996 |
"coverage": 0.8,
|
| 997 |
+
"accuracy": 0.75,
|
| 998 |
+
"confidence_cutoff": 0.6044538628392925
|
| 999 |
}
|
| 1000 |
},
|
| 1001 |
+
"score_mae": 0.42706852941593343,
|
| 1002 |
+
"ranked_probability_score": 0.07153163324216405
|
| 1003 |
},
|
| 1004 |
"yelp_yn": {
|
| 1005 |
"n": 80,
|
| 1006 |
+
"nll": 0.3011138456134589,
|
| 1007 |
"acc": 0.9,
|
| 1008 |
+
"ece": 0.06096574774156942,
|
| 1009 |
+
"brier": 0.15056167094115347,
|
| 1010 |
+
"mean_conf": 0.9609657477415695,
|
| 1011 |
+
"confident_error_rate": 0.0375,
|
| 1012 |
+
"coverage_at_0_9": 0.8875,
|
| 1013 |
+
"accuracy_at_0_9": 0.9577464788732394,
|
| 1014 |
+
"confidence_bias": 0.06096574774156949,
|
| 1015 |
+
"top_bins": {
|
| 1016 |
+
"0.9": {
|
| 1017 |
+
"n": 71,
|
| 1018 |
+
"errors": 3,
|
| 1019 |
+
"error_rate": 0.04225352112676056
|
| 1020 |
+
},
|
| 1021 |
+
"0.95": {
|
| 1022 |
+
"n": 68,
|
| 1023 |
+
"errors": 2,
|
| 1024 |
+
"error_rate": 0.029411764705882353
|
| 1025 |
+
},
|
| 1026 |
+
"0.99": {
|
| 1027 |
+
"n": 51,
|
| 1028 |
+
"errors": 2,
|
| 1029 |
+
"error_rate": 0.0392156862745098
|
| 1030 |
+
}
|
| 1031 |
+
},
|
| 1032 |
"selective": {
|
| 1033 |
"0.5": {
|
| 1034 |
"coverage": 0.5,
|
| 1035 |
"accuracy": 0.975,
|
| 1036 |
+
"confidence_cutoff": 0.9978126800890552
|
| 1037 |
},
|
| 1038 |
"0.8": {
|
| 1039 |
"coverage": 0.8,
|
| 1040 |
+
"accuracy": 0.96875,
|
| 1041 |
+
"confidence_cutoff": 0.9756043927703096
|
| 1042 |
}
|
| 1043 |
}
|
| 1044 |
}
|
|
|
|
| 1046 |
"variants": {
|
| 1047 |
"clean": {
|
| 1048 |
"n": 1264,
|
| 1049 |
+
"nll": 0.6291677798298513,
|
| 1050 |
+
"acc": 0.8006329113924051,
|
| 1051 |
+
"ece": 0.0856521215240488,
|
| 1052 |
+
"brier": 0.2969903053792275,
|
| 1053 |
+
"mean_conf": 0.8857679456333062,
|
| 1054 |
+
"confident_error_rate": 0.05458860759493671,
|
| 1055 |
+
"coverage_at_0_9": 0.6566455696202531,
|
| 1056 |
+
"accuracy_at_0_9": 0.9168674698795181,
|
| 1057 |
+
"confidence_bias": 0.08513503424090108,
|
| 1058 |
+
"top_bins": {
|
| 1059 |
+
"0.9": {
|
| 1060 |
+
"n": 830,
|
| 1061 |
+
"errors": 69,
|
| 1062 |
+
"error_rate": 0.08313253012048193
|
| 1063 |
+
},
|
| 1064 |
+
"0.95": {
|
| 1065 |
+
"n": 758,
|
| 1066 |
+
"errors": 50,
|
| 1067 |
+
"error_rate": 0.06596306068601583
|
| 1068 |
+
},
|
| 1069 |
+
"0.99": {
|
| 1070 |
+
"n": 633,
|
| 1071 |
+
"errors": 29,
|
| 1072 |
+
"error_rate": 0.045813586097946286
|
| 1073 |
+
}
|
| 1074 |
+
},
|
| 1075 |
"selective": {
|
| 1076 |
"0.5": {
|
| 1077 |
"coverage": 0.5,
|
| 1078 |
+
"accuracy": 0.9541139240506329,
|
| 1079 |
+
"confidence_cutoff": 0.9903846647088901
|
| 1080 |
},
|
| 1081 |
"0.8": {
|
| 1082 |
"coverage": 0.8006329113924051,
|
| 1083 |
+
"accuracy": 0.8685770750988142,
|
| 1084 |
+
"confidence_cutoff": 0.7193367680537224
|
| 1085 |
}
|
| 1086 |
},
|
| 1087 |
+
"score_mae": 0.5034937052883923,
|
| 1088 |
+
"ranked_probability_score": 0.08453946806779199
|
| 1089 |
},
|
| 1090 |
"none_absent": {
|
| 1091 |
"n": 60,
|
| 1092 |
+
"nll": 0.4604061524495717,
|
| 1093 |
+
"acc": 0.85,
|
| 1094 |
+
"ece": 0.11060768954216148,
|
| 1095 |
+
"brier": 0.22512676868916184,
|
| 1096 |
+
"mean_conf": 0.9466371284236123,
|
| 1097 |
+
"confident_error_rate": 0.08333333333333333,
|
| 1098 |
+
"coverage_at_0_9": 0.8333333333333334,
|
| 1099 |
+
"accuracy_at_0_9": 0.9,
|
| 1100 |
+
"confidence_bias": 0.09663712842361227,
|
| 1101 |
+
"top_bins": {
|
| 1102 |
+
"0.9": {
|
| 1103 |
+
"n": 50,
|
| 1104 |
+
"errors": 5,
|
| 1105 |
+
"error_rate": 0.1
|
| 1106 |
+
},
|
| 1107 |
+
"0.95": {
|
| 1108 |
+
"n": 47,
|
| 1109 |
+
"errors": 3,
|
| 1110 |
+
"error_rate": 0.06382978723404255
|
| 1111 |
+
},
|
| 1112 |
+
"0.99": {
|
| 1113 |
+
"n": 34,
|
| 1114 |
+
"errors": 1,
|
| 1115 |
+
"error_rate": 0.029411764705882353
|
| 1116 |
+
}
|
| 1117 |
+
},
|
| 1118 |
"selective": {
|
| 1119 |
"0.5": {
|
| 1120 |
"coverage": 0.5,
|
| 1121 |
+
"accuracy": 0.9666666666666667,
|
| 1122 |
+
"confidence_cutoff": 0.9939384823343665
|
| 1123 |
},
|
| 1124 |
"0.8": {
|
| 1125 |
"coverage": 0.8,
|
| 1126 |
+
"accuracy": 0.9166666666666666,
|
| 1127 |
+
"confidence_cutoff": 0.9280017588436554
|
| 1128 |
}
|
| 1129 |
}
|
| 1130 |
},
|
| 1131 |
"none_present": {
|
| 1132 |
"n": 60,
|
| 1133 |
+
"nll": 0.906454532923607,
|
| 1134 |
+
"acc": 0.8,
|
| 1135 |
+
"ece": 0.12046818710766105,
|
| 1136 |
+
"brier": 0.31172266907926255,
|
| 1137 |
+
"mean_conf": 0.9099585661005806,
|
| 1138 |
+
"confident_error_rate": 0.08333333333333333,
|
| 1139 |
+
"coverage_at_0_9": 0.7,
|
| 1140 |
+
"accuracy_at_0_9": 0.8809523809523809,
|
| 1141 |
+
"confidence_bias": 0.10995856610058052,
|
| 1142 |
+
"top_bins": {
|
| 1143 |
+
"0.9": {
|
| 1144 |
+
"n": 42,
|
| 1145 |
+
"errors": 5,
|
| 1146 |
+
"error_rate": 0.11904761904761904
|
| 1147 |
+
},
|
| 1148 |
+
"0.95": {
|
| 1149 |
+
"n": 35,
|
| 1150 |
+
"errors": 3,
|
| 1151 |
+
"error_rate": 0.08571428571428572
|
| 1152 |
+
},
|
| 1153 |
+
"0.99": {
|
| 1154 |
+
"n": 21,
|
| 1155 |
+
"errors": 1,
|
| 1156 |
+
"error_rate": 0.047619047619047616
|
| 1157 |
+
}
|
| 1158 |
+
},
|
| 1159 |
"selective": {
|
| 1160 |
"0.5": {
|
| 1161 |
"coverage": 0.5,
|
| 1162 |
+
"accuracy": 0.9333333333333333,
|
| 1163 |
+
"confidence_cutoff": 0.9652735907557973
|
| 1164 |
},
|
| 1165 |
"0.8": {
|
| 1166 |
"coverage": 0.8,
|
| 1167 |
+
"accuracy": 0.8958333333333334,
|
| 1168 |
+
"confidence_cutoff": 0.8530403587702203
|
| 1169 |
}
|
| 1170 |
}
|
| 1171 |
},
|
| 1172 |
"permuted": {
|
| 1173 |
"n": 84,
|
| 1174 |
+
"nll": 0.4639830274521712,
|
| 1175 |
+
"acc": 0.9166666666666666,
|
| 1176 |
+
"ece": 0.08051812654264506,
|
| 1177 |
+
"brier": 0.1518899638801877,
|
| 1178 |
+
"mean_conf": 0.9656925740801011,
|
| 1179 |
+
"confident_error_rate": 0.05952380952380952,
|
| 1180 |
+
"coverage_at_0_9": 0.9047619047619048,
|
| 1181 |
+
"accuracy_at_0_9": 0.9342105263157895,
|
| 1182 |
+
"confidence_bias": 0.04902590741343449,
|
| 1183 |
+
"top_bins": {
|
| 1184 |
+
"0.9": {
|
| 1185 |
+
"n": 76,
|
| 1186 |
+
"errors": 5,
|
| 1187 |
+
"error_rate": 0.06578947368421052
|
| 1188 |
+
},
|
| 1189 |
+
"0.95": {
|
| 1190 |
+
"n": 73,
|
| 1191 |
+
"errors": 4,
|
| 1192 |
+
"error_rate": 0.0547945205479452
|
| 1193 |
+
},
|
| 1194 |
+
"0.99": {
|
| 1195 |
+
"n": 66,
|
| 1196 |
+
"errors": 4,
|
| 1197 |
+
"error_rate": 0.06060606060606061
|
| 1198 |
+
}
|
| 1199 |
+
},
|
| 1200 |
"selective": {
|
| 1201 |
"0.5": {
|
| 1202 |
"coverage": 0.5,
|
| 1203 |
"accuracy": 0.9761904761904762,
|
| 1204 |
+
"confidence_cutoff": 0.9992487515625712
|
| 1205 |
},
|
| 1206 |
"0.8": {
|
| 1207 |
"coverage": 0.8095238095238095,
|
| 1208 |
+
"accuracy": 0.9411764705882353,
|
| 1209 |
+
"confidence_cutoff": 0.9828670677995383
|
| 1210 |
}
|
| 1211 |
}
|
| 1212 |
}
|
|
|
|
| 1214 |
"heldout_tasks": {},
|
| 1215 |
"permutation": {
|
| 1216 |
"n": 60,
|
| 1217 |
+
"mean_max_delta": 0.030460907247831088,
|
| 1218 |
+
"flip_rate": 0.06666666666666667
|
| 1219 |
},
|
| 1220 |
+
"temperature": 1.9318726578496908,
|
| 1221 |
"calibrated_clean": {
|
| 1222 |
"n": 1264,
|
| 1223 |
+
"nll": 0.5024173854081659,
|
| 1224 |
+
"acc": 0.8006329113924051,
|
| 1225 |
+
"ece": 0.03166951395049025,
|
| 1226 |
+
"brier": 0.2798369926110367,
|
| 1227 |
+
"mean_conf": 0.8090278226398868,
|
| 1228 |
+
"confident_error_rate": 0.02689873417721519,
|
| 1229 |
+
"coverage_at_0_9": 0.5087025316455697,
|
| 1230 |
+
"accuracy_at_0_9": 0.9471228615863142,
|
| 1231 |
+
"confidence_bias": 0.008394911247481707,
|
| 1232 |
+
"top_bins": {
|
| 1233 |
+
"0.9": {
|
| 1234 |
+
"n": 643,
|
| 1235 |
+
"errors": 34,
|
| 1236 |
+
"error_rate": 0.05287713841368585
|
| 1237 |
+
},
|
| 1238 |
+
"0.95": {
|
| 1239 |
+
"n": 532,
|
| 1240 |
+
"errors": 14,
|
| 1241 |
+
"error_rate": 0.02631578947368421
|
| 1242 |
+
},
|
| 1243 |
+
"0.99": {
|
| 1244 |
+
"n": 262,
|
| 1245 |
+
"errors": 5,
|
| 1246 |
+
"error_rate": 0.019083969465648856
|
| 1247 |
+
}
|
| 1248 |
+
},
|
| 1249 |
"selective": {
|
| 1250 |
"0.5": {
|
| 1251 |
"coverage": 0.5,
|
| 1252 |
+
"accuracy": 0.9493670886075949,
|
| 1253 |
+
"confidence_cutoff": 0.9051024319980593
|
| 1254 |
},
|
| 1255 |
"0.8": {
|
| 1256 |
"coverage": 0.8006329113924051,
|
| 1257 |
+
"accuracy": 0.8656126482213439,
|
| 1258 |
+
"confidence_cutoff": 0.5859782618246442
|
| 1259 |
}
|
| 1260 |
},
|
| 1261 |
+
"score_mae": 0.5272155953122818,
|
| 1262 |
+
"ranked_probability_score": 0.08158842149335112
|
| 1263 |
},
|
| 1264 |
"metric_policy": {
|
| 1265 |
"nll_floor": 1e-09,
|
|
|
|
| 1276 |
"truncated_records": 0
|
| 1277 |
},
|
| 1278 |
"latency_ms": {
|
| 1279 |
+
"median": 23.046538500011593,
|
| 1280 |
+
"p95": 34.68967335006141
|
| 1281 |
},
|
| 1282 |
"transfer": {
|
| 1283 |
+
"objective": -1.0504768775174653,
|
| 1284 |
"paired_flip": {
|
| 1285 |
"pairs": 64,
|
| 1286 |
+
"flip_rate": 0.125,
|
| 1287 |
+
"both_correct_rate": 0.078125,
|
| 1288 |
"invariant_pairs": 24,
|
| 1289 |
"invariance_rate": 0.9583333333333334,
|
| 1290 |
+
"invariant_both_correct_rate": 0.4583333333333333
|
| 1291 |
},
|
| 1292 |
"clean": {
|
| 1293 |
"n": 656,
|
| 1294 |
+
"nll": 0.9789452089606111,
|
| 1295 |
+
"acc": 0.6204268292682927,
|
| 1296 |
+
"ece": 0.1521695989629108,
|
| 1297 |
+
"brier": 0.5360083888306599,
|
| 1298 |
+
"mean_conf": 0.771367853277944,
|
| 1299 |
+
"confident_error_rate": 0.10823170731707317,
|
| 1300 |
+
"coverage_at_0_9": 0.4222560975609756,
|
| 1301 |
+
"accuracy_at_0_9": 0.7436823104693141,
|
| 1302 |
+
"confidence_bias": 0.15094102400965126,
|
| 1303 |
+
"top_bins": {
|
| 1304 |
+
"0.9": {
|
| 1305 |
+
"n": 277,
|
| 1306 |
+
"errors": 71,
|
| 1307 |
+
"error_rate": 0.2563176895306859
|
| 1308 |
+
},
|
| 1309 |
+
"0.95": {
|
| 1310 |
+
"n": 224,
|
| 1311 |
+
"errors": 51,
|
| 1312 |
+
"error_rate": 0.22767857142857142
|
| 1313 |
+
},
|
| 1314 |
+
"0.99": {
|
| 1315 |
+
"n": 76,
|
| 1316 |
+
"errors": 8,
|
| 1317 |
+
"error_rate": 0.10526315789473684
|
| 1318 |
+
}
|
| 1319 |
+
},
|
| 1320 |
"selective": {
|
| 1321 |
"0.5": {
|
| 1322 |
"coverage": 0.5,
|
| 1323 |
+
"accuracy": 0.7164634146341463,
|
| 1324 |
+
"confidence_cutoff": 0.821762097206147
|
| 1325 |
},
|
| 1326 |
"0.8": {
|
| 1327 |
"coverage": 0.8003048780487805,
|
| 1328 |
+
"accuracy": 0.6647619047619048,
|
| 1329 |
+
"confidence_cutoff": 0.5518338844726647
|
| 1330 |
}
|
| 1331 |
},
|
| 1332 |
+
"score_mae": 0.6509208807376698,
|
| 1333 |
+
"ranked_probability_score": 0.30494000794977794
|
| 1334 |
},
|
| 1335 |
"tasks": {
|
| 1336 |
"composition_held_and_or": {
|
| 1337 |
"n": 32,
|
| 1338 |
+
"nll": 0.7375551551827133,
|
| 1339 |
+
"acc": 0.5,
|
| 1340 |
+
"ece": 0.20788620213868944,
|
| 1341 |
+
"brier": 0.5276657618880333,
|
| 1342 |
+
"mean_conf": 0.6322889744455078,
|
| 1343 |
+
"confident_error_rate": 0.0,
|
| 1344 |
+
"coverage_at_0_9": 0.03125,
|
| 1345 |
+
"accuracy_at_0_9": 1.0,
|
| 1346 |
+
"confidence_bias": 0.1322889744455078,
|
| 1347 |
+
"top_bins": {
|
| 1348 |
+
"0.9": {
|
| 1349 |
+
"n": 1,
|
| 1350 |
+
"errors": 0,
|
| 1351 |
+
"error_rate": 0.0
|
| 1352 |
+
},
|
| 1353 |
+
"0.95": {
|
| 1354 |
+
"n": 0,
|
| 1355 |
+
"errors": 0,
|
| 1356 |
+
"error_rate": null
|
| 1357 |
+
},
|
| 1358 |
+
"0.99": {
|
| 1359 |
+
"n": 0,
|
| 1360 |
+
"errors": 0,
|
| 1361 |
+
"error_rate": null
|
| 1362 |
+
}
|
| 1363 |
+
},
|
| 1364 |
"selective": {
|
| 1365 |
"0.5": {
|
| 1366 |
+
"coverage": 0.5,
|
| 1367 |
+
"accuracy": 0.6875,
|
| 1368 |
+
"confidence_cutoff": 0.5948953628540039
|
| 1369 |
},
|
| 1370 |
"0.8": {
|
| 1371 |
"coverage": 0.84375,
|
| 1372 |
+
"accuracy": 0.48148148148148145,
|
| 1373 |
+
"confidence_cutoff": 0.5323349078793851
|
| 1374 |
}
|
| 1375 |
}
|
| 1376 |
},
|
| 1377 |
"composition_held_conditional": {
|
| 1378 |
"n": 32,
|
| 1379 |
+
"nll": 0.8014951063953045,
|
| 1380 |
+
"acc": 0.46875,
|
| 1381 |
+
"ece": 0.2740389039677988,
|
| 1382 |
+
"brier": 0.5929889640661177,
|
| 1383 |
+
"mean_conf": 0.7427889039677988,
|
| 1384 |
+
"confident_error_rate": 0.03125,
|
| 1385 |
+
"coverage_at_0_9": 0.21875,
|
| 1386 |
+
"accuracy_at_0_9": 0.8571428571428571,
|
| 1387 |
+
"confidence_bias": 0.27403890396779884,
|
| 1388 |
+
"top_bins": {
|
| 1389 |
+
"0.9": {
|
| 1390 |
+
"n": 7,
|
| 1391 |
+
"errors": 1,
|
| 1392 |
+
"error_rate": 0.14285714285714285
|
| 1393 |
+
},
|
| 1394 |
+
"0.95": {
|
| 1395 |
+
"n": 3,
|
| 1396 |
+
"errors": 0,
|
| 1397 |
+
"error_rate": 0.0
|
| 1398 |
+
},
|
| 1399 |
+
"0.99": {
|
| 1400 |
+
"n": 0,
|
| 1401 |
+
"errors": 0,
|
| 1402 |
+
"error_rate": null
|
| 1403 |
+
}
|
| 1404 |
+
},
|
| 1405 |
"selective": {
|
| 1406 |
"0.5": {
|
| 1407 |
"coverage": 0.5,
|
| 1408 |
+
"accuracy": 0.625,
|
| 1409 |
+
"confidence_cutoff": 0.7428816556930542
|
| 1410 |
},
|
| 1411 |
"0.8": {
|
| 1412 |
"coverage": 0.8125,
|
| 1413 |
+
"accuracy": 0.5,
|
| 1414 |
+
"confidence_cutoff": 0.6217024100841528
|
| 1415 |
}
|
| 1416 |
}
|
| 1417 |
},
|
| 1418 |
"composition_held_or_not": {
|
| 1419 |
"n": 32,
|
| 1420 |
+
"nll": 0.7994410561517606,
|
| 1421 |
+
"acc": 0.53125,
|
| 1422 |
+
"ece": 0.23994846307304188,
|
| 1423 |
+
"brier": 0.5828810243922846,
|
| 1424 |
+
"mean_conf": 0.6147556398926043,
|
| 1425 |
"confident_error_rate": 0.0,
|
| 1426 |
"coverage_at_0_9": 0.0,
|
| 1427 |
"accuracy_at_0_9": null,
|
| 1428 |
+
"confidence_bias": 0.08350563989260429,
|
| 1429 |
+
"top_bins": {
|
| 1430 |
+
"0.9": {
|
| 1431 |
+
"n": 0,
|
| 1432 |
+
"errors": 0,
|
| 1433 |
+
"error_rate": null
|
| 1434 |
+
},
|
| 1435 |
+
"0.95": {
|
| 1436 |
+
"n": 0,
|
| 1437 |
+
"errors": 0,
|
| 1438 |
+
"error_rate": null
|
| 1439 |
+
},
|
| 1440 |
+
"0.99": {
|
| 1441 |
+
"n": 0,
|
| 1442 |
+
"errors": 0,
|
| 1443 |
+
"error_rate": null
|
| 1444 |
+
}
|
| 1445 |
+
},
|
| 1446 |
"selective": {
|
| 1447 |
"0.5": {
|
| 1448 |
"coverage": 0.5,
|
| 1449 |
+
"accuracy": 0.375,
|
| 1450 |
+
"confidence_cutoff": 0.6124296370872747
|
| 1451 |
},
|
| 1452 |
"0.8": {
|
| 1453 |
"coverage": 0.8125,
|
| 1454 |
+
"accuracy": 0.5384615384615384,
|
| 1455 |
+
"confidence_cutoff": 0.5218212163991098
|
| 1456 |
}
|
| 1457 |
}
|
| 1458 |
},
|
| 1459 |
"contrastive_authorization": {
|
| 1460 |
"n": 40,
|
| 1461 |
+
"nll": 1.3085126106876184,
|
| 1462 |
+
"acc": 0.525,
|
| 1463 |
+
"ece": 0.40097863797629046,
|
| 1464 |
+
"brier": 0.7670783556054575,
|
| 1465 |
+
"mean_conf": 0.9259786379762904,
|
| 1466 |
+
"confident_error_rate": 0.325,
|
| 1467 |
+
"coverage_at_0_9": 0.825,
|
| 1468 |
+
"accuracy_at_0_9": 0.6060606060606061,
|
| 1469 |
+
"confidence_bias": 0.4009786379762904,
|
| 1470 |
+
"top_bins": {
|
| 1471 |
+
"0.9": {
|
| 1472 |
+
"n": 33,
|
| 1473 |
+
"errors": 13,
|
| 1474 |
+
"error_rate": 0.3939393939393939
|
| 1475 |
+
},
|
| 1476 |
+
"0.95": {
|
| 1477 |
+
"n": 30,
|
| 1478 |
+
"errors": 10,
|
| 1479 |
+
"error_rate": 0.3333333333333333
|
| 1480 |
+
},
|
| 1481 |
+
"0.99": {
|
| 1482 |
+
"n": 0,
|
| 1483 |
+
"errors": 0,
|
| 1484 |
+
"error_rate": null
|
| 1485 |
+
}
|
| 1486 |
+
},
|
| 1487 |
"selective": {
|
| 1488 |
"0.5": {
|
| 1489 |
"coverage": 0.5,
|
| 1490 |
+
"accuracy": 0.95,
|
| 1491 |
+
"confidence_cutoff": 0.973573315945716
|
| 1492 |
},
|
| 1493 |
"0.8": {
|
| 1494 |
"coverage": 0.8,
|
| 1495 |
"accuracy": 0.625,
|
| 1496 |
+
"confidence_cutoff": 0.935230381304252
|
| 1497 |
}
|
| 1498 |
}
|
| 1499 |
},
|
| 1500 |
"contrastive_deadline": {
|
| 1501 |
"n": 40,
|
| 1502 |
+
"nll": 2.9410876864668603,
|
| 1503 |
+
"acc": 0.375,
|
| 1504 |
+
"ece": 0.622533434407196,
|
| 1505 |
+
"brier": 1.2108976873827435,
|
| 1506 |
+
"mean_conf": 0.9483477456076063,
|
| 1507 |
+
"confident_error_rate": 0.6,
|
| 1508 |
+
"coverage_at_0_9": 0.9,
|
| 1509 |
+
"accuracy_at_0_9": 0.3333333333333333,
|
| 1510 |
+
"confidence_bias": 0.5733477456076063,
|
| 1511 |
+
"top_bins": {
|
| 1512 |
+
"0.9": {
|
| 1513 |
+
"n": 36,
|
| 1514 |
+
"errors": 24,
|
| 1515 |
+
"error_rate": 0.6666666666666666
|
| 1516 |
+
},
|
| 1517 |
+
"0.95": {
|
| 1518 |
+
"n": 31,
|
| 1519 |
+
"errors": 23,
|
| 1520 |
+
"error_rate": 0.7419354838709677
|
| 1521 |
+
},
|
| 1522 |
+
"0.99": {
|
| 1523 |
+
"n": 7,
|
| 1524 |
+
"errors": 4,
|
| 1525 |
+
"error_rate": 0.5714285714285714
|
| 1526 |
+
}
|
| 1527 |
+
},
|
| 1528 |
"selective": {
|
| 1529 |
"0.5": {
|
| 1530 |
"coverage": 0.5,
|
| 1531 |
+
"accuracy": 0.3,
|
| 1532 |
+
"confidence_cutoff": 0.9746268367408688
|
| 1533 |
},
|
| 1534 |
"0.8": {
|
| 1535 |
"coverage": 0.8,
|
| 1536 |
"accuracy": 0.28125,
|
| 1537 |
+
"confidence_cutoff": 0.9499374190141426
|
| 1538 |
}
|
| 1539 |
},
|
| 1540 |
+
"score_mae": 0.6509208807376698,
|
| 1541 |
+
"ranked_probability_score": 0.30494000794977794
|
| 1542 |
},
|
| 1543 |
"emotion": {
|
| 1544 |
"n": 80,
|
| 1545 |
+
"nll": 1.3401844583691325,
|
| 1546 |
+
"acc": 0.4875,
|
| 1547 |
+
"ece": 0.2024348539640546,
|
| 1548 |
+
"brier": 0.6865956742406507,
|
| 1549 |
+
"mean_conf": 0.6552581599189324,
|
| 1550 |
+
"confident_error_rate": 0.075,
|
| 1551 |
+
"coverage_at_0_9": 0.2125,
|
| 1552 |
+
"accuracy_at_0_9": 0.6470588235294118,
|
| 1553 |
+
"confidence_bias": 0.16775815991893245,
|
| 1554 |
+
"top_bins": {
|
| 1555 |
+
"0.9": {
|
| 1556 |
+
"n": 17,
|
| 1557 |
+
"errors": 6,
|
| 1558 |
+
"error_rate": 0.35294117647058826
|
| 1559 |
+
},
|
| 1560 |
+
"0.95": {
|
| 1561 |
+
"n": 12,
|
| 1562 |
+
"errors": 3,
|
| 1563 |
+
"error_rate": 0.25
|
| 1564 |
+
},
|
| 1565 |
+
"0.99": {
|
| 1566 |
+
"n": 5,
|
| 1567 |
+
"errors": 1,
|
| 1568 |
+
"error_rate": 0.2
|
| 1569 |
+
}
|
| 1570 |
+
},
|
| 1571 |
"selective": {
|
| 1572 |
"0.5": {
|
| 1573 |
"coverage": 0.5,
|
| 1574 |
+
"accuracy": 0.55,
|
| 1575 |
+
"confidence_cutoff": 0.654870567876276
|
| 1576 |
},
|
| 1577 |
"0.8": {
|
| 1578 |
"coverage": 0.8,
|
| 1579 |
+
"accuracy": 0.53125,
|
| 1580 |
+
"confidence_cutoff": 0.425271655038087
|
| 1581 |
}
|
| 1582 |
}
|
| 1583 |
},
|
| 1584 |
"mmlu": {
|
| 1585 |
"n": 80,
|
| 1586 |
+
"nll": 1.37276267509046,
|
| 1587 |
+
"acc": 0.5,
|
| 1588 |
+
"ece": 0.14707810693154655,
|
| 1589 |
+
"brier": 0.6666753052385594,
|
| 1590 |
+
"mean_conf": 0.6311081277400896,
|
| 1591 |
+
"confident_error_rate": 0.0375,
|
| 1592 |
+
"coverage_at_0_9": 0.1625,
|
| 1593 |
+
"accuracy_at_0_9": 0.7692307692307693,
|
| 1594 |
+
"confidence_bias": 0.13110812774008962,
|
| 1595 |
+
"top_bins": {
|
| 1596 |
+
"0.9": {
|
| 1597 |
+
"n": 13,
|
| 1598 |
+
"errors": 3,
|
| 1599 |
+
"error_rate": 0.23076923076923078
|
| 1600 |
+
},
|
| 1601 |
+
"0.95": {
|
| 1602 |
+
"n": 10,
|
| 1603 |
+
"errors": 2,
|
| 1604 |
+
"error_rate": 0.2
|
| 1605 |
+
},
|
| 1606 |
+
"0.99": {
|
| 1607 |
+
"n": 5,
|
| 1608 |
+
"errors": 1,
|
| 1609 |
+
"error_rate": 0.2
|
| 1610 |
+
}
|
| 1611 |
+
},
|
| 1612 |
"selective": {
|
| 1613 |
"0.5": {
|
| 1614 |
"coverage": 0.5,
|
| 1615 |
+
"accuracy": 0.6,
|
| 1616 |
+
"confidence_cutoff": 0.5743741671251661
|
| 1617 |
},
|
| 1618 |
"0.8": {
|
| 1619 |
"coverage": 0.8,
|
| 1620 |
+
"accuracy": 0.53125,
|
| 1621 |
+
"confidence_cutoff": 0.43554868910006816
|
| 1622 |
}
|
| 1623 |
}
|
| 1624 |
},
|
| 1625 |
"paws": {
|
| 1626 |
"n": 80,
|
| 1627 |
+
"nll": 0.6751459309347105,
|
| 1628 |
+
"acc": 0.5875,
|
| 1629 |
+
"ece": 0.0530374871173959,
|
| 1630 |
+
"brier": 0.47488187717818836,
|
| 1631 |
+
"mean_conf": 0.5908602187530765,
|
| 1632 |
"confident_error_rate": 0.0125,
|
| 1633 |
"coverage_at_0_9": 0.0125,
|
| 1634 |
"accuracy_at_0_9": 0.0,
|
| 1635 |
+
"confidence_bias": 0.0033602187530764605,
|
| 1636 |
+
"top_bins": {
|
| 1637 |
+
"0.9": {
|
| 1638 |
+
"n": 1,
|
| 1639 |
+
"errors": 1,
|
| 1640 |
+
"error_rate": 1.0
|
| 1641 |
+
},
|
| 1642 |
+
"0.95": {
|
| 1643 |
+
"n": 0,
|
| 1644 |
+
"errors": 0,
|
| 1645 |
+
"error_rate": null
|
| 1646 |
+
},
|
| 1647 |
+
"0.99": {
|
| 1648 |
+
"n": 0,
|
| 1649 |
+
"errors": 0,
|
| 1650 |
+
"error_rate": null
|
| 1651 |
+
}
|
| 1652 |
+
},
|
| 1653 |
"selective": {
|
| 1654 |
"0.5": {
|
| 1655 |
"coverage": 0.5,
|
| 1656 |
+
"accuracy": 0.65,
|
| 1657 |
+
"confidence_cutoff": 0.5645352938494261
|
| 1658 |
},
|
| 1659 |
"0.8": {
|
| 1660 |
"coverage": 0.8,
|
| 1661 |
+
"accuracy": 0.65625,
|
| 1662 |
+
"confidence_cutoff": 0.5327850142951314
|
| 1663 |
}
|
| 1664 |
}
|
| 1665 |
},
|
| 1666 |
"qnli": {
|
| 1667 |
"n": 80,
|
| 1668 |
+
"nll": 0.4965084025860714,
|
| 1669 |
+
"acc": 0.85,
|
| 1670 |
+
"ece": 0.08773846278340862,
|
| 1671 |
+
"brier": 0.26700316317551664,
|
| 1672 |
+
"mean_conf": 0.8941121420330482,
|
| 1673 |
+
"confident_error_rate": 0.075,
|
| 1674 |
+
"coverage_at_0_9": 0.65,
|
| 1675 |
+
"accuracy_at_0_9": 0.8846153846153846,
|
| 1676 |
+
"confidence_bias": 0.04411214203304825,
|
| 1677 |
+
"top_bins": {
|
| 1678 |
+
"0.9": {
|
| 1679 |
+
"n": 52,
|
| 1680 |
+
"errors": 6,
|
| 1681 |
+
"error_rate": 0.11538461538461539
|
| 1682 |
+
},
|
| 1683 |
+
"0.95": {
|
| 1684 |
+
"n": 38,
|
| 1685 |
+
"errors": 4,
|
| 1686 |
+
"error_rate": 0.10526315789473684
|
| 1687 |
+
},
|
| 1688 |
+
"0.99": {
|
| 1689 |
+
"n": 3,
|
| 1690 |
+
"errors": 1,
|
| 1691 |
+
"error_rate": 0.3333333333333333
|
| 1692 |
+
}
|
| 1693 |
+
},
|
| 1694 |
"selective": {
|
| 1695 |
"0.5": {
|
| 1696 |
"coverage": 0.5,
|
| 1697 |
+
"accuracy": 0.9,
|
| 1698 |
+
"confidence_cutoff": 0.9491957112907086
|
| 1699 |
},
|
| 1700 |
"0.8": {
|
| 1701 |
"coverage": 0.8,
|
| 1702 |
+
"accuracy": 0.875,
|
| 1703 |
+
"confidence_cutoff": 0.8221900340588386
|
| 1704 |
}
|
| 1705 |
}
|
| 1706 |
},
|
| 1707 |
"sciq": {
|
| 1708 |
"n": 80,
|
| 1709 |
+
"nll": 0.2663922795365539,
|
| 1710 |
+
"acc": 0.925,
|
| 1711 |
+
"ece": 0.08478984004891495,
|
| 1712 |
+
"brier": 0.1402940023600535,
|
| 1713 |
+
"mean_conf": 0.9375872823845735,
|
| 1714 |
+
"confident_error_rate": 0.05,
|
| 1715 |
+
"coverage_at_0_9": 0.8125,
|
| 1716 |
+
"accuracy_at_0_9": 0.9384615384615385,
|
| 1717 |
+
"confidence_bias": 0.012587282384573406,
|
| 1718 |
+
"top_bins": {
|
| 1719 |
+
"0.9": {
|
| 1720 |
+
"n": 65,
|
| 1721 |
+
"errors": 4,
|
| 1722 |
+
"error_rate": 0.06153846153846154
|
| 1723 |
+
},
|
| 1724 |
+
"0.95": {
|
| 1725 |
+
"n": 60,
|
| 1726 |
+
"errors": 1,
|
| 1727 |
+
"error_rate": 0.016666666666666666
|
| 1728 |
+
},
|
| 1729 |
+
"0.99": {
|
| 1730 |
+
"n": 49,
|
| 1731 |
+
"errors": 0,
|
| 1732 |
+
"error_rate": 0.0
|
| 1733 |
+
}
|
| 1734 |
+
},
|
| 1735 |
"selective": {
|
| 1736 |
"0.5": {
|
| 1737 |
"coverage": 0.5,
|
| 1738 |
"accuracy": 1.0,
|
| 1739 |
+
"confidence_cutoff": 0.9986336163388265
|
| 1740 |
},
|
| 1741 |
"0.8": {
|
| 1742 |
"coverage": 0.8,
|
| 1743 |
+
"accuracy": 0.9375,
|
| 1744 |
+
"confidence_cutoff": 0.9180914769789518
|
| 1745 |
}
|
| 1746 |
}
|
| 1747 |
},
|
| 1748 |
"tweet_offensive": {
|
| 1749 |
"n": 80,
|
| 1750 |
+
"nll": 0.8161602912909318,
|
| 1751 |
+
"acc": 0.6875,
|
| 1752 |
+
"ece": 0.21453663827286928,
|
| 1753 |
+
"brier": 0.4894164445857676,
|
| 1754 |
+
"mean_conf": 0.8831938669351078,
|
| 1755 |
+
"confident_error_rate": 0.1625,
|
| 1756 |
+
"coverage_at_0_9": 0.65,
|
| 1757 |
+
"accuracy_at_0_9": 0.75,
|
| 1758 |
+
"confidence_bias": 0.19569386693510782,
|
| 1759 |
+
"top_bins": {
|
| 1760 |
+
"0.9": {
|
| 1761 |
+
"n": 52,
|
| 1762 |
+
"errors": 13,
|
| 1763 |
+
"error_rate": 0.25
|
| 1764 |
+
},
|
| 1765 |
+
"0.95": {
|
| 1766 |
+
"n": 40,
|
| 1767 |
+
"errors": 8,
|
| 1768 |
+
"error_rate": 0.2
|
| 1769 |
+
},
|
| 1770 |
+
"0.99": {
|
| 1771 |
+
"n": 7,
|
| 1772 |
+
"errors": 1,
|
| 1773 |
+
"error_rate": 0.14285714285714285
|
| 1774 |
+
}
|
| 1775 |
+
},
|
| 1776 |
"selective": {
|
| 1777 |
"0.5": {
|
| 1778 |
"coverage": 0.5,
|
| 1779 |
"accuracy": 0.8,
|
| 1780 |
+
"confidence_cutoff": 0.9502587702067471
|
| 1781 |
},
|
| 1782 |
"0.8": {
|
| 1783 |
"coverage": 0.8,
|
| 1784 |
+
"accuracy": 0.71875,
|
| 1785 |
+
"confidence_cutoff": 0.821762097206147
|
| 1786 |
}
|
| 1787 |
}
|
| 1788 |
}
|
|
|
|
| 1790 |
"variants": {
|
| 1791 |
"clean": {
|
| 1792 |
"n": 656,
|
| 1793 |
+
"nll": 0.9789452089606111,
|
| 1794 |
+
"acc": 0.6204268292682927,
|
| 1795 |
+
"ece": 0.1521695989629108,
|
| 1796 |
+
"brier": 0.5360083888306599,
|
| 1797 |
+
"mean_conf": 0.771367853277944,
|
| 1798 |
+
"confident_error_rate": 0.10823170731707317,
|
| 1799 |
+
"coverage_at_0_9": 0.4222560975609756,
|
| 1800 |
+
"accuracy_at_0_9": 0.7436823104693141,
|
| 1801 |
+
"confidence_bias": 0.15094102400965126,
|
| 1802 |
+
"top_bins": {
|
| 1803 |
+
"0.9": {
|
| 1804 |
+
"n": 277,
|
| 1805 |
+
"errors": 71,
|
| 1806 |
+
"error_rate": 0.2563176895306859
|
| 1807 |
+
},
|
| 1808 |
+
"0.95": {
|
| 1809 |
+
"n": 224,
|
| 1810 |
+
"errors": 51,
|
| 1811 |
+
"error_rate": 0.22767857142857142
|
| 1812 |
+
},
|
| 1813 |
+
"0.99": {
|
| 1814 |
+
"n": 76,
|
| 1815 |
+
"errors": 8,
|
| 1816 |
+
"error_rate": 0.10526315789473684
|
| 1817 |
+
}
|
| 1818 |
+
},
|
| 1819 |
"selective": {
|
| 1820 |
"0.5": {
|
| 1821 |
"coverage": 0.5,
|
| 1822 |
+
"accuracy": 0.7164634146341463,
|
| 1823 |
+
"confidence_cutoff": 0.821762097206147
|
| 1824 |
},
|
| 1825 |
"0.8": {
|
| 1826 |
"coverage": 0.8003048780487805,
|
| 1827 |
+
"accuracy": 0.6647619047619048,
|
| 1828 |
+
"confidence_cutoff": 0.5518338844726647
|
| 1829 |
}
|
| 1830 |
},
|
| 1831 |
+
"score_mae": 0.6509208807376698,
|
| 1832 |
+
"ranked_probability_score": 0.30494000794977794
|
| 1833 |
},
|
| 1834 |
"none_absent": {
|
| 1835 |
"n": 36,
|
| 1836 |
+
"nll": 0.4556876721453974,
|
| 1837 |
+
"acc": 0.8888888888888888,
|
| 1838 |
+
"ece": 0.1367857535659568,
|
| 1839 |
+
"brier": 0.241037659869187,
|
| 1840 |
+
"mean_conf": 0.8319164028780932,
|
| 1841 |
"confident_error_rate": 0.05555555555555555,
|
| 1842 |
+
"coverage_at_0_9": 0.5555555555555556,
|
| 1843 |
+
"accuracy_at_0_9": 0.9,
|
| 1844 |
+
"confidence_bias": -0.05697248601079563,
|
| 1845 |
+
"top_bins": {
|
| 1846 |
+
"0.9": {
|
| 1847 |
+
"n": 20,
|
| 1848 |
+
"errors": 2,
|
| 1849 |
+
"error_rate": 0.1
|
| 1850 |
+
},
|
| 1851 |
+
"0.95": {
|
| 1852 |
+
"n": 16,
|
| 1853 |
+
"errors": 1,
|
| 1854 |
+
"error_rate": 0.0625
|
| 1855 |
+
},
|
| 1856 |
+
"0.99": {
|
| 1857 |
+
"n": 7,
|
| 1858 |
+
"errors": 0,
|
| 1859 |
+
"error_rate": 0.0
|
| 1860 |
+
}
|
| 1861 |
+
},
|
| 1862 |
"selective": {
|
| 1863 |
"0.5": {
|
| 1864 |
"coverage": 0.5,
|
| 1865 |
"accuracy": 0.8888888888888888,
|
| 1866 |
+
"confidence_cutoff": 0.9294200676744145
|
| 1867 |
},
|
| 1868 |
"0.8": {
|
| 1869 |
"coverage": 0.8055555555555556,
|
| 1870 |
+
"accuracy": 0.896551724137931,
|
| 1871 |
+
"confidence_cutoff": 0.708448785328515
|
| 1872 |
}
|
| 1873 |
}
|
| 1874 |
},
|
| 1875 |
"none_present": {
|
| 1876 |
"n": 36,
|
| 1877 |
+
"nll": 2.009091057715083,
|
| 1878 |
+
"acc": 0.5,
|
| 1879 |
+
"ece": 0.405110777515241,
|
| 1880 |
+
"brier": 0.8798406934000841,
|
| 1881 |
+
"mean_conf": 0.8581696393397298,
|
| 1882 |
+
"confident_error_rate": 0.3055555555555556,
|
| 1883 |
+
"coverage_at_0_9": 0.6944444444444444,
|
| 1884 |
+
"accuracy_at_0_9": 0.56,
|
| 1885 |
+
"confidence_bias": 0.3581696393397298,
|
| 1886 |
+
"top_bins": {
|
| 1887 |
+
"0.9": {
|
| 1888 |
+
"n": 25,
|
| 1889 |
+
"errors": 11,
|
| 1890 |
+
"error_rate": 0.44
|
| 1891 |
+
},
|
| 1892 |
+
"0.95": {
|
| 1893 |
+
"n": 17,
|
| 1894 |
+
"errors": 7,
|
| 1895 |
+
"error_rate": 0.4117647058823529
|
| 1896 |
+
},
|
| 1897 |
+
"0.99": {
|
| 1898 |
+
"n": 5,
|
| 1899 |
+
"errors": 2,
|
| 1900 |
+
"error_rate": 0.4
|
| 1901 |
+
}
|
| 1902 |
+
},
|
| 1903 |
"selective": {
|
| 1904 |
"0.5": {
|
| 1905 |
"coverage": 0.5,
|
| 1906 |
"accuracy": 0.6111111111111112,
|
| 1907 |
+
"confidence_cutoff": 0.9473120411566688
|
| 1908 |
},
|
| 1909 |
"0.8": {
|
| 1910 |
"coverage": 0.8055555555555556,
|
| 1911 |
+
"accuracy": 0.5172413793103449,
|
| 1912 |
+
"confidence_cutoff": 0.6828171065659605
|
| 1913 |
}
|
| 1914 |
}
|
| 1915 |
},
|
| 1916 |
"permuted": {
|
| 1917 |
"n": 36,
|
| 1918 |
+
"nll": 0.7104708508010783,
|
| 1919 |
"acc": 0.6944444444444444,
|
| 1920 |
+
"ece": 0.11763119975963233,
|
| 1921 |
+
"brier": 0.3882533889486839,
|
| 1922 |
+
"mean_conf": 0.7904855165061119,
|
| 1923 |
+
"confident_error_rate": 0.0,
|
| 1924 |
+
"coverage_at_0_9": 0.4444444444444444,
|
| 1925 |
+
"accuracy_at_0_9": 1.0,
|
| 1926 |
+
"confidence_bias": 0.09604107206166745,
|
| 1927 |
+
"top_bins": {
|
| 1928 |
+
"0.9": {
|
| 1929 |
+
"n": 16,
|
| 1930 |
+
"errors": 0,
|
| 1931 |
+
"error_rate": 0.0
|
| 1932 |
+
},
|
| 1933 |
+
"0.95": {
|
| 1934 |
+
"n": 13,
|
| 1935 |
+
"errors": 0,
|
| 1936 |
+
"error_rate": 0.0
|
| 1937 |
+
},
|
| 1938 |
+
"0.99": {
|
| 1939 |
+
"n": 8,
|
| 1940 |
+
"errors": 0,
|
| 1941 |
+
"error_rate": 0.0
|
| 1942 |
+
}
|
| 1943 |
+
},
|
| 1944 |
"selective": {
|
| 1945 |
"0.5": {
|
| 1946 |
"coverage": 0.5,
|
| 1947 |
+
"accuracy": 0.9444444444444444,
|
| 1948 |
+
"confidence_cutoff": 0.889561050715202
|
| 1949 |
},
|
| 1950 |
"0.8": {
|
| 1951 |
"coverage": 0.8055555555555556,
|
| 1952 |
+
"accuracy": 0.7931034482758621,
|
| 1953 |
+
"confidence_cutoff": 0.6079693765027671
|
| 1954 |
}
|
| 1955 |
}
|
| 1956 |
}
|
|
|
|
| 1958 |
"heldout_tasks": {
|
| 1959 |
"composition_held_and_or": {
|
| 1960 |
"n": 32,
|
| 1961 |
+
"nll": 0.7375551551827133,
|
| 1962 |
+
"acc": 0.5,
|
| 1963 |
+
"ece": 0.20788620213868944,
|
| 1964 |
+
"brier": 0.5276657618880333,
|
| 1965 |
+
"mean_conf": 0.6322889744455078,
|
| 1966 |
+
"confident_error_rate": 0.0,
|
| 1967 |
+
"coverage_at_0_9": 0.03125,
|
| 1968 |
+
"accuracy_at_0_9": 1.0,
|
| 1969 |
+
"confidence_bias": 0.1322889744455078,
|
| 1970 |
+
"top_bins": {
|
| 1971 |
+
"0.9": {
|
| 1972 |
+
"n": 1,
|
| 1973 |
+
"errors": 0,
|
| 1974 |
+
"error_rate": 0.0
|
| 1975 |
+
},
|
| 1976 |
+
"0.95": {
|
| 1977 |
+
"n": 0,
|
| 1978 |
+
"errors": 0,
|
| 1979 |
+
"error_rate": null
|
| 1980 |
+
},
|
| 1981 |
+
"0.99": {
|
| 1982 |
+
"n": 0,
|
| 1983 |
+
"errors": 0,
|
| 1984 |
+
"error_rate": null
|
| 1985 |
+
}
|
| 1986 |
+
},
|
| 1987 |
"selective": {
|
| 1988 |
"0.5": {
|
| 1989 |
+
"coverage": 0.5,
|
| 1990 |
+
"accuracy": 0.6875,
|
| 1991 |
+
"confidence_cutoff": 0.5948953628540039
|
| 1992 |
},
|
| 1993 |
"0.8": {
|
| 1994 |
"coverage": 0.84375,
|
| 1995 |
+
"accuracy": 0.48148148148148145,
|
| 1996 |
+
"confidence_cutoff": 0.5323349078793851
|
| 1997 |
}
|
| 1998 |
}
|
| 1999 |
},
|
| 2000 |
"composition_held_conditional": {
|
| 2001 |
"n": 32,
|
| 2002 |
+
"nll": 0.8014951063953045,
|
| 2003 |
+
"acc": 0.46875,
|
| 2004 |
+
"ece": 0.2740389039677988,
|
| 2005 |
+
"brier": 0.5929889640661177,
|
| 2006 |
+
"mean_conf": 0.7427889039677988,
|
| 2007 |
+
"confident_error_rate": 0.03125,
|
| 2008 |
+
"coverage_at_0_9": 0.21875,
|
| 2009 |
+
"accuracy_at_0_9": 0.8571428571428571,
|
| 2010 |
+
"confidence_bias": 0.27403890396779884,
|
| 2011 |
+
"top_bins": {
|
| 2012 |
+
"0.9": {
|
| 2013 |
+
"n": 7,
|
| 2014 |
+
"errors": 1,
|
| 2015 |
+
"error_rate": 0.14285714285714285
|
| 2016 |
+
},
|
| 2017 |
+
"0.95": {
|
| 2018 |
+
"n": 3,
|
| 2019 |
+
"errors": 0,
|
| 2020 |
+
"error_rate": 0.0
|
| 2021 |
+
},
|
| 2022 |
+
"0.99": {
|
| 2023 |
+
"n": 0,
|
| 2024 |
+
"errors": 0,
|
| 2025 |
+
"error_rate": null
|
| 2026 |
+
}
|
| 2027 |
+
},
|
| 2028 |
"selective": {
|
| 2029 |
"0.5": {
|
| 2030 |
"coverage": 0.5,
|
| 2031 |
+
"accuracy": 0.625,
|
| 2032 |
+
"confidence_cutoff": 0.7428816556930542
|
| 2033 |
},
|
| 2034 |
"0.8": {
|
| 2035 |
"coverage": 0.8125,
|
| 2036 |
+
"accuracy": 0.5,
|
| 2037 |
+
"confidence_cutoff": 0.6217024100841528
|
| 2038 |
}
|
| 2039 |
}
|
| 2040 |
},
|
| 2041 |
"composition_held_or_not": {
|
| 2042 |
"n": 32,
|
| 2043 |
+
"nll": 0.7994410561517606,
|
| 2044 |
+
"acc": 0.53125,
|
| 2045 |
+
"ece": 0.23994846307304188,
|
| 2046 |
+
"brier": 0.5828810243922846,
|
| 2047 |
+
"mean_conf": 0.6147556398926043,
|
| 2048 |
"confident_error_rate": 0.0,
|
| 2049 |
"coverage_at_0_9": 0.0,
|
| 2050 |
"accuracy_at_0_9": null,
|
| 2051 |
+
"confidence_bias": 0.08350563989260429,
|
| 2052 |
+
"top_bins": {
|
| 2053 |
+
"0.9": {
|
| 2054 |
+
"n": 0,
|
| 2055 |
+
"errors": 0,
|
| 2056 |
+
"error_rate": null
|
| 2057 |
+
},
|
| 2058 |
+
"0.95": {
|
| 2059 |
+
"n": 0,
|
| 2060 |
+
"errors": 0,
|
| 2061 |
+
"error_rate": null
|
| 2062 |
+
},
|
| 2063 |
+
"0.99": {
|
| 2064 |
+
"n": 0,
|
| 2065 |
+
"errors": 0,
|
| 2066 |
+
"error_rate": null
|
| 2067 |
+
}
|
| 2068 |
+
},
|
| 2069 |
"selective": {
|
| 2070 |
"0.5": {
|
| 2071 |
"coverage": 0.5,
|
| 2072 |
+
"accuracy": 0.375,
|
| 2073 |
+
"confidence_cutoff": 0.6124296370872747
|
| 2074 |
},
|
| 2075 |
"0.8": {
|
| 2076 |
"coverage": 0.8125,
|
| 2077 |
+
"accuracy": 0.5384615384615384,
|
| 2078 |
+
"confidence_cutoff": 0.5218212163991098
|
| 2079 |
}
|
| 2080 |
}
|
| 2081 |
},
|
| 2082 |
"contrastive_authorization": {
|
| 2083 |
"n": 40,
|
| 2084 |
+
"nll": 1.3085126106876184,
|
| 2085 |
+
"acc": 0.525,
|
| 2086 |
+
"ece": 0.40097863797629046,
|
| 2087 |
+
"brier": 0.7670783556054575,
|
| 2088 |
+
"mean_conf": 0.9259786379762904,
|
| 2089 |
+
"confident_error_rate": 0.325,
|
| 2090 |
+
"coverage_at_0_9": 0.825,
|
| 2091 |
+
"accuracy_at_0_9": 0.6060606060606061,
|
| 2092 |
+
"confidence_bias": 0.4009786379762904,
|
| 2093 |
+
"top_bins": {
|
| 2094 |
+
"0.9": {
|
| 2095 |
+
"n": 33,
|
| 2096 |
+
"errors": 13,
|
| 2097 |
+
"error_rate": 0.3939393939393939
|
| 2098 |
+
},
|
| 2099 |
+
"0.95": {
|
| 2100 |
+
"n": 30,
|
| 2101 |
+
"errors": 10,
|
| 2102 |
+
"error_rate": 0.3333333333333333
|
| 2103 |
+
},
|
| 2104 |
+
"0.99": {
|
| 2105 |
+
"n": 0,
|
| 2106 |
+
"errors": 0,
|
| 2107 |
+
"error_rate": null
|
| 2108 |
+
}
|
| 2109 |
+
},
|
| 2110 |
"selective": {
|
| 2111 |
"0.5": {
|
| 2112 |
"coverage": 0.5,
|
| 2113 |
+
"accuracy": 0.95,
|
| 2114 |
+
"confidence_cutoff": 0.973573315945716
|
| 2115 |
},
|
| 2116 |
"0.8": {
|
| 2117 |
"coverage": 0.8,
|
| 2118 |
"accuracy": 0.625,
|
| 2119 |
+
"confidence_cutoff": 0.935230381304252
|
| 2120 |
}
|
| 2121 |
}
|
| 2122 |
},
|
| 2123 |
"contrastive_deadline": {
|
| 2124 |
"n": 40,
|
| 2125 |
+
"nll": 2.9410876864668603,
|
| 2126 |
+
"acc": 0.375,
|
| 2127 |
+
"ece": 0.622533434407196,
|
| 2128 |
+
"brier": 1.2108976873827435,
|
| 2129 |
+
"mean_conf": 0.9483477456076063,
|
| 2130 |
+
"confident_error_rate": 0.6,
|
| 2131 |
+
"coverage_at_0_9": 0.9,
|
| 2132 |
+
"accuracy_at_0_9": 0.3333333333333333,
|
| 2133 |
+
"confidence_bias": 0.5733477456076063,
|
| 2134 |
+
"top_bins": {
|
| 2135 |
+
"0.9": {
|
| 2136 |
+
"n": 36,
|
| 2137 |
+
"errors": 24,
|
| 2138 |
+
"error_rate": 0.6666666666666666
|
| 2139 |
+
},
|
| 2140 |
+
"0.95": {
|
| 2141 |
+
"n": 31,
|
| 2142 |
+
"errors": 23,
|
| 2143 |
+
"error_rate": 0.7419354838709677
|
| 2144 |
+
},
|
| 2145 |
+
"0.99": {
|
| 2146 |
+
"n": 7,
|
| 2147 |
+
"errors": 4,
|
| 2148 |
+
"error_rate": 0.5714285714285714
|
| 2149 |
+
}
|
| 2150 |
+
},
|
| 2151 |
"selective": {
|
| 2152 |
"0.5": {
|
| 2153 |
"coverage": 0.5,
|
| 2154 |
+
"accuracy": 0.3,
|
| 2155 |
+
"confidence_cutoff": 0.9746268367408688
|
| 2156 |
},
|
| 2157 |
"0.8": {
|
| 2158 |
"coverage": 0.8,
|
| 2159 |
"accuracy": 0.28125,
|
| 2160 |
+
"confidence_cutoff": 0.9499374190141426
|
| 2161 |
}
|
| 2162 |
},
|
| 2163 |
+
"score_mae": 0.6509208807376698,
|
| 2164 |
+
"ranked_probability_score": 0.30494000794977794
|
| 2165 |
},
|
| 2166 |
"emotion": {
|
| 2167 |
"n": 80,
|
| 2168 |
+
"nll": 1.3401844583691325,
|
| 2169 |
+
"acc": 0.4875,
|
| 2170 |
+
"ece": 0.2024348539640546,
|
| 2171 |
+
"brier": 0.6865956742406507,
|
| 2172 |
+
"mean_conf": 0.6552581599189324,
|
| 2173 |
+
"confident_error_rate": 0.075,
|
| 2174 |
+
"coverage_at_0_9": 0.2125,
|
| 2175 |
+
"accuracy_at_0_9": 0.6470588235294118,
|
| 2176 |
+
"confidence_bias": 0.16775815991893245,
|
| 2177 |
+
"top_bins": {
|
| 2178 |
+
"0.9": {
|
| 2179 |
+
"n": 17,
|
| 2180 |
+
"errors": 6,
|
| 2181 |
+
"error_rate": 0.35294117647058826
|
| 2182 |
+
},
|
| 2183 |
+
"0.95": {
|
| 2184 |
+
"n": 12,
|
| 2185 |
+
"errors": 3,
|
| 2186 |
+
"error_rate": 0.25
|
| 2187 |
+
},
|
| 2188 |
+
"0.99": {
|
| 2189 |
+
"n": 5,
|
| 2190 |
+
"errors": 1,
|
| 2191 |
+
"error_rate": 0.2
|
| 2192 |
+
}
|
| 2193 |
+
},
|
| 2194 |
"selective": {
|
| 2195 |
"0.5": {
|
| 2196 |
"coverage": 0.5,
|
| 2197 |
+
"accuracy": 0.55,
|
| 2198 |
+
"confidence_cutoff": 0.654870567876276
|
| 2199 |
},
|
| 2200 |
"0.8": {
|
| 2201 |
"coverage": 0.8,
|
| 2202 |
+
"accuracy": 0.53125,
|
| 2203 |
+
"confidence_cutoff": 0.425271655038087
|
| 2204 |
}
|
| 2205 |
}
|
| 2206 |
},
|
| 2207 |
"mmlu": {
|
| 2208 |
"n": 80,
|
| 2209 |
+
"nll": 1.37276267509046,
|
| 2210 |
+
"acc": 0.5,
|
| 2211 |
+
"ece": 0.14707810693154655,
|
| 2212 |
+
"brier": 0.6666753052385594,
|
| 2213 |
+
"mean_conf": 0.6311081277400896,
|
| 2214 |
+
"confident_error_rate": 0.0375,
|
| 2215 |
+
"coverage_at_0_9": 0.1625,
|
| 2216 |
+
"accuracy_at_0_9": 0.7692307692307693,
|
| 2217 |
+
"confidence_bias": 0.13110812774008962,
|
| 2218 |
+
"top_bins": {
|
| 2219 |
+
"0.9": {
|
| 2220 |
+
"n": 13,
|
| 2221 |
+
"errors": 3,
|
| 2222 |
+
"error_rate": 0.23076923076923078
|
| 2223 |
+
},
|
| 2224 |
+
"0.95": {
|
| 2225 |
+
"n": 10,
|
| 2226 |
+
"errors": 2,
|
| 2227 |
+
"error_rate": 0.2
|
| 2228 |
+
},
|
| 2229 |
+
"0.99": {
|
| 2230 |
+
"n": 5,
|
| 2231 |
+
"errors": 1,
|
| 2232 |
+
"error_rate": 0.2
|
| 2233 |
+
}
|
| 2234 |
+
},
|
| 2235 |
"selective": {
|
| 2236 |
"0.5": {
|
| 2237 |
"coverage": 0.5,
|
| 2238 |
+
"accuracy": 0.6,
|
| 2239 |
+
"confidence_cutoff": 0.5743741671251661
|
| 2240 |
},
|
| 2241 |
"0.8": {
|
| 2242 |
"coverage": 0.8,
|
| 2243 |
+
"accuracy": 0.53125,
|
| 2244 |
+
"confidence_cutoff": 0.43554868910006816
|
| 2245 |
}
|
| 2246 |
}
|
| 2247 |
},
|
| 2248 |
"paws": {
|
| 2249 |
"n": 80,
|
| 2250 |
+
"nll": 0.6751459309347105,
|
| 2251 |
+
"acc": 0.5875,
|
| 2252 |
+
"ece": 0.0530374871173959,
|
| 2253 |
+
"brier": 0.47488187717818836,
|
| 2254 |
+
"mean_conf": 0.5908602187530765,
|
| 2255 |
"confident_error_rate": 0.0125,
|
| 2256 |
"coverage_at_0_9": 0.0125,
|
| 2257 |
"accuracy_at_0_9": 0.0,
|
| 2258 |
+
"confidence_bias": 0.0033602187530764605,
|
| 2259 |
+
"top_bins": {
|
| 2260 |
+
"0.9": {
|
| 2261 |
+
"n": 1,
|
| 2262 |
+
"errors": 1,
|
| 2263 |
+
"error_rate": 1.0
|
| 2264 |
+
},
|
| 2265 |
+
"0.95": {
|
| 2266 |
+
"n": 0,
|
| 2267 |
+
"errors": 0,
|
| 2268 |
+
"error_rate": null
|
| 2269 |
+
},
|
| 2270 |
+
"0.99": {
|
| 2271 |
+
"n": 0,
|
| 2272 |
+
"errors": 0,
|
| 2273 |
+
"error_rate": null
|
| 2274 |
+
}
|
| 2275 |
+
},
|
| 2276 |
"selective": {
|
| 2277 |
"0.5": {
|
| 2278 |
"coverage": 0.5,
|
| 2279 |
+
"accuracy": 0.65,
|
| 2280 |
+
"confidence_cutoff": 0.5645352938494261
|
| 2281 |
},
|
| 2282 |
"0.8": {
|
| 2283 |
"coverage": 0.8,
|
| 2284 |
+
"accuracy": 0.65625,
|
| 2285 |
+
"confidence_cutoff": 0.5327850142951314
|
| 2286 |
}
|
| 2287 |
}
|
| 2288 |
},
|
| 2289 |
"qnli": {
|
| 2290 |
"n": 80,
|
| 2291 |
+
"nll": 0.4965084025860714,
|
| 2292 |
+
"acc": 0.85,
|
| 2293 |
+
"ece": 0.08773846278340862,
|
| 2294 |
+
"brier": 0.26700316317551664,
|
| 2295 |
+
"mean_conf": 0.8941121420330482,
|
| 2296 |
+
"confident_error_rate": 0.075,
|
| 2297 |
+
"coverage_at_0_9": 0.65,
|
| 2298 |
+
"accuracy_at_0_9": 0.8846153846153846,
|
| 2299 |
+
"confidence_bias": 0.04411214203304825,
|
| 2300 |
+
"top_bins": {
|
| 2301 |
+
"0.9": {
|
| 2302 |
+
"n": 52,
|
| 2303 |
+
"errors": 6,
|
| 2304 |
+
"error_rate": 0.11538461538461539
|
| 2305 |
+
},
|
| 2306 |
+
"0.95": {
|
| 2307 |
+
"n": 38,
|
| 2308 |
+
"errors": 4,
|
| 2309 |
+
"error_rate": 0.10526315789473684
|
| 2310 |
+
},
|
| 2311 |
+
"0.99": {
|
| 2312 |
+
"n": 3,
|
| 2313 |
+
"errors": 1,
|
| 2314 |
+
"error_rate": 0.3333333333333333
|
| 2315 |
+
}
|
| 2316 |
+
},
|
| 2317 |
"selective": {
|
| 2318 |
"0.5": {
|
| 2319 |
"coverage": 0.5,
|
| 2320 |
+
"accuracy": 0.9,
|
| 2321 |
+
"confidence_cutoff": 0.9491957112907086
|
| 2322 |
},
|
| 2323 |
"0.8": {
|
| 2324 |
"coverage": 0.8,
|
| 2325 |
+
"accuracy": 0.875,
|
| 2326 |
+
"confidence_cutoff": 0.8221900340588386
|
| 2327 |
}
|
| 2328 |
}
|
| 2329 |
},
|
| 2330 |
"sciq": {
|
| 2331 |
"n": 80,
|
| 2332 |
+
"nll": 0.2663922795365539,
|
| 2333 |
+
"acc": 0.925,
|
| 2334 |
+
"ece": 0.08478984004891495,
|
| 2335 |
+
"brier": 0.1402940023600535,
|
| 2336 |
+
"mean_conf": 0.9375872823845735,
|
| 2337 |
+
"confident_error_rate": 0.05,
|
| 2338 |
+
"coverage_at_0_9": 0.8125,
|
| 2339 |
+
"accuracy_at_0_9": 0.9384615384615385,
|
| 2340 |
+
"confidence_bias": 0.012587282384573406,
|
| 2341 |
+
"top_bins": {
|
| 2342 |
+
"0.9": {
|
| 2343 |
+
"n": 65,
|
| 2344 |
+
"errors": 4,
|
| 2345 |
+
"error_rate": 0.06153846153846154
|
| 2346 |
+
},
|
| 2347 |
+
"0.95": {
|
| 2348 |
+
"n": 60,
|
| 2349 |
+
"errors": 1,
|
| 2350 |
+
"error_rate": 0.016666666666666666
|
| 2351 |
+
},
|
| 2352 |
+
"0.99": {
|
| 2353 |
+
"n": 49,
|
| 2354 |
+
"errors": 0,
|
| 2355 |
+
"error_rate": 0.0
|
| 2356 |
+
}
|
| 2357 |
+
},
|
| 2358 |
"selective": {
|
| 2359 |
"0.5": {
|
| 2360 |
"coverage": 0.5,
|
| 2361 |
"accuracy": 1.0,
|
| 2362 |
+
"confidence_cutoff": 0.9986336163388265
|
| 2363 |
},
|
| 2364 |
"0.8": {
|
| 2365 |
"coverage": 0.8,
|
| 2366 |
+
"accuracy": 0.9375,
|
| 2367 |
+
"confidence_cutoff": 0.9180914769789518
|
| 2368 |
}
|
| 2369 |
}
|
| 2370 |
},
|
| 2371 |
"tweet_offensive": {
|
| 2372 |
"n": 80,
|
| 2373 |
+
"nll": 0.8161602912909318,
|
| 2374 |
+
"acc": 0.6875,
|
| 2375 |
+
"ece": 0.21453663827286928,
|
| 2376 |
+
"brier": 0.4894164445857676,
|
| 2377 |
+
"mean_conf": 0.8831938669351078,
|
| 2378 |
+
"confident_error_rate": 0.1625,
|
| 2379 |
+
"coverage_at_0_9": 0.65,
|
| 2380 |
+
"accuracy_at_0_9": 0.75,
|
| 2381 |
+
"confidence_bias": 0.19569386693510782,
|
| 2382 |
+
"top_bins": {
|
| 2383 |
+
"0.9": {
|
| 2384 |
+
"n": 52,
|
| 2385 |
+
"errors": 13,
|
| 2386 |
+
"error_rate": 0.25
|
| 2387 |
+
},
|
| 2388 |
+
"0.95": {
|
| 2389 |
+
"n": 40,
|
| 2390 |
+
"errors": 8,
|
| 2391 |
+
"error_rate": 0.2
|
| 2392 |
+
},
|
| 2393 |
+
"0.99": {
|
| 2394 |
+
"n": 7,
|
| 2395 |
+
"errors": 1,
|
| 2396 |
+
"error_rate": 0.14285714285714285
|
| 2397 |
+
}
|
| 2398 |
+
},
|
| 2399 |
"selective": {
|
| 2400 |
"0.5": {
|
| 2401 |
"coverage": 0.5,
|
| 2402 |
"accuracy": 0.8,
|
| 2403 |
+
"confidence_cutoff": 0.9502587702067471
|
| 2404 |
},
|
| 2405 |
"0.8": {
|
| 2406 |
"coverage": 0.8,
|
| 2407 |
+
"accuracy": 0.71875,
|
| 2408 |
+
"confidence_cutoff": 0.821762097206147
|
| 2409 |
}
|
| 2410 |
}
|
| 2411 |
}
|
| 2412 |
},
|
| 2413 |
"permutation": {
|
| 2414 |
"n": 36,
|
| 2415 |
+
"mean_max_delta": 0.07562486874312503,
|
| 2416 |
+
"flip_rate": 0.1111111111111111
|
| 2417 |
},
|
| 2418 |
+
"temperature": 1.9318726578496908,
|
| 2419 |
"calibrated_clean": {
|
| 2420 |
"n": 656,
|
| 2421 |
+
"nll": 0.7970266863281446,
|
| 2422 |
+
"acc": 0.6204268292682927,
|
| 2423 |
+
"ece": 0.07284936542930612,
|
| 2424 |
+
"brier": 0.4891594782446207,
|
| 2425 |
+
"mean_conf": 0.6720420107384325,
|
| 2426 |
+
"confident_error_rate": 0.012195121951219513,
|
| 2427 |
+
"coverage_at_0_9": 0.12804878048780488,
|
| 2428 |
+
"accuracy_at_0_9": 0.9047619047619048,
|
| 2429 |
+
"confidence_bias": 0.05161518147013977,
|
| 2430 |
+
"top_bins": {
|
| 2431 |
+
"0.9": {
|
| 2432 |
+
"n": 84,
|
| 2433 |
+
"errors": 8,
|
| 2434 |
+
"error_rate": 0.09523809523809523
|
| 2435 |
+
},
|
| 2436 |
+
"0.95": {
|
| 2437 |
+
"n": 49,
|
| 2438 |
+
"errors": 1,
|
| 2439 |
+
"error_rate": 0.02040816326530612
|
| 2440 |
+
},
|
| 2441 |
+
"0.99": {
|
| 2442 |
+
"n": 16,
|
| 2443 |
+
"errors": 0,
|
| 2444 |
+
"error_rate": 0.0
|
| 2445 |
+
}
|
| 2446 |
+
},
|
| 2447 |
"selective": {
|
| 2448 |
"0.5": {
|
| 2449 |
"coverage": 0.5,
|
| 2450 |
+
"accuracy": 0.7134146341463414,
|
| 2451 |
+
"confidence_cutoff": 0.6578320406689322
|
| 2452 |
},
|
| 2453 |
"0.8": {
|
| 2454 |
"coverage": 0.8003048780487805,
|
| 2455 |
+
"accuracy": 0.6647619047619048,
|
| 2456 |
+
"confidence_cutoff": 0.5117883377953908
|
| 2457 |
}
|
| 2458 |
},
|
| 2459 |
+
"score_mae": 0.6682966727607406,
|
| 2460 |
+
"ranked_probability_score": 0.26850318644312043
|
| 2461 |
},
|
| 2462 |
"metric_policy": {
|
| 2463 |
"nll_floor": 1e-09,
|
|
|
|
| 2474 |
"truncated_records": 0
|
| 2475 |
},
|
| 2476 |
"latency_ms": {
|
| 2477 |
+
"median": 22.76046250000263,
|
| 2478 |
+
"p95": 25.7828150500643
|
| 2479 |
},
|
| 2480 |
"suite_sha256": "31677c2256b406222e7d94ffdc0a02a70ce05746b9efe307876024c4e77291d1"
|
| 2481 |
},
|
| 2482 |
"provenance": {
|
| 2483 |
"config": {
|
| 2484 |
"epochs": 2,
|
| 2485 |
+
"seed": 2,
|
| 2486 |
+
"lr": 0.0001,
|
| 2487 |
"lora": 16,
|
| 2488 |
"accum": 1,
|
| 2489 |
"batch": 8,
|
|
|
|
| 2494 |
"p_none_distract": 0.12,
|
| 2495 |
"p_distract": 0.15,
|
| 2496 |
"p_none_pair": 0.25,
|
| 2497 |
+
"synthetic_repeat": 1,
|
| 2498 |
+
"public_frac": 1.0,
|
| 2499 |
+
"head_lr": 0.0,
|
| 2500 |
+
"weight_decay": 0.01,
|
| 2501 |
+
"anchor_w": 0.0,
|
| 2502 |
"base": "Qwen/Qwen3-0.6B-Base",
|
| 2503 |
"dtype": "bf16"
|
| 2504 |
},
|
| 2505 |
+
"config_sha256": "e0d2ec20a46cf1cfa71568e1c695fe1f1ae16650f0ec6808df3ca00af3cdc3df",
|
| 2506 |
+
"suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2",
|
| 2507 |
"source_hashes": {
|
| 2508 |
"kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
|
| 2509 |
+
"kev/anchors.py": "089d8a5493502bb26f540eb1c5e681780ca0bb01276073d4e1733211f15d0e10",
|
| 2510 |
"kev/api.py": "cdb0602684d798ccd3fc2c86be622f2a8e7bcdeee64d07f95604c7d3fd4701ec",
|
| 2511 |
+
"kev/autoresearch.py": "0a8aa6b57c1c9ba93d25b2cf631b03aed2e686e374c1148002eec48765b7b1fc",
|
| 2512 |
+
"kev/benchmark.py": "4dccec28be6becf6c3dcc8c186e8e1908a6482f15f602a60b440cc9cb4ac8a59",
|
| 2513 |
"kev/compare.py": "bd0445f021de59e35c7bff9304e39dd2e1211e3877a453575594ae7b81b0ada4",
|
| 2514 |
+
"kev/composition.py": "f335ed17e18e0a544893db5e22b9a059e6ce1b2e14dbb863ac7d7bbf8f3e0536",
|
| 2515 |
+
"kev/contrastive.py": "cbb979aa5d40265ad0e64695f94b281d91751fa405811ddfa8212fede111edcf",
|
| 2516 |
+
"kev/data.py": "9729f8f497b02aca542915d6f1bea3ff55956629d7d9610370aea91ee66efd23",
|
| 2517 |
+
"kev/evaluate.py": "2b15a093ae9085a2d3999d5f57cb6293ce93d101bcddd0ea252089538e8f3a07",
|
| 2518 |
+
"kev/experiment.py": "c635394fe56c9aa11765f8bbf8f133dcf3e57cbe8f2bce4ebdc7b4e85fa536ff",
|
| 2519 |
"kev/jev.py": "e0213782359ba2f95adbf045ddaf0a008b08b4162bf6a9b66d4ce55fc91a51cb",
|
| 2520 |
+
"kev/model.py": "03f87dc4c081fa8dc1f31c192de8015ddb8213b235d5807b66aebd75b50cb407",
|
| 2521 |
"kev/plot.py": "d689c7dd18cde7f9ea77cb4f55c50ff7da1880b21cb2ecf244348e45842a1382",
|
| 2522 |
+
"kev/publish.py": "b88c710140e7717ba42d539fdbe785507ac8b9d56b673641ade61c4136bb7dc9",
|
| 2523 |
+
"kev/serve.py": "570e995aa98b51dbbf867694276d29fc9b9f3f1e62f22bea0116ae51d6497003",
|
| 2524 |
+
"kev/study_v3.py": "9fc44d44dad09a4f1ee29a7bcb2eb3c7aa373d09186d0e9533666b69ec95c401",
|
| 2525 |
+
"kev/suite.py": "f221e5a50111bab24ada035ff68ea3c3067482ff1edda48b8d4ece769043a561",
|
| 2526 |
+
"kev/train.py": "e0df6503506d6a0cb5ccd35ae38bb072a6103383adf7beab4269bb1a1f5fcdce",
|
| 2527 |
+
"modal_app.py": "fcc64f410bdd456bf5514b70da4e57da1f3a2f89c70918df0a6e7003141f33b8",
|
| 2528 |
"pyproject.toml": "ce566ee11b68c0cb587bb591943641f403d761d820817693dae7590faacf9888",
|
| 2529 |
"uv.lock": "846fdc64a8bd6279a17a4365f18cd4e92ba9622d50d378aafcb5be6604296138"
|
| 2530 |
},
|
| 2531 |
+
"git_commit": "37cb2f6155afec1054de2cc78dfd7c7dd24a956d",
|
| 2532 |
"platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.36",
|
| 2533 |
"torch": "2.8.0+cu128",
|
| 2534 |
"device": "cuda",
|
|
|
|
| 2538 |
"mechanism_checks": {
|
| 2539 |
"n": 8,
|
| 2540 |
"packed_max_delta": 0.0,
|
| 2541 |
+
"sibling_max_delta": 2.950429916381836e-06,
|
| 2542 |
"tolerance": 0.001,
|
| 2543 |
"passed": true
|
| 2544 |
},
|
|
|
|
| 2549 |
"isolation_and_packing": true,
|
| 2550 |
"transfer_complete": true,
|
| 2551 |
"heldout_pairs_at_least_70pct": false,
|
| 2552 |
+
"transfer_confident_errors_below_10pct": false
|
| 2553 |
},
|
| 2554 |
"policy": "Research screening only: 1e-3 isolation; 5pp task regression; 70% heldout pair correctness; <=10% confident errors; no automatic release."
|
| 2555 |
},
|
| 2556 |
+
"wall_seconds": 727.917370719,
|
| 2557 |
"promotable": false,
|
| 2558 |
"test_evaluated": false,
|
| 2559 |
"training_resources": {
|
| 2560 |
+
"wall_seconds": 638.1045656204224,
|
| 2561 |
+
"records_seen": 30370,
|
| 2562 |
+
"requested_records": 25152,
|
| 2563 |
"truncated_records": 0,
|
| 2564 |
"rejected_records": 0,
|
| 2565 |
+
"optimizer_steps": 3144,
|
| 2566 |
+
"forward_tokens": 5829284,
|
| 2567 |
+
"peak_device_bytes": 37059323392,
|
| 2568 |
"device": "cuda",
|
| 2569 |
"dtype": "bf16",
|
| 2570 |
"batch": 8,
|
| 2571 |
+
"peak_rss_bytes": 8731115520
|
| 2572 |
}
|
| 2573 |
}
|
train.log
CHANGED
|
@@ -1,275 +1,317 @@
|
|
| 1 |
device=cuda trainable params=10.6M
|
| 2 |
-
|
| 3 |
-
ep0 step 10/
|
| 4 |
-
ep0 step 20/
|
| 5 |
-
ep0 step 30/
|
| 6 |
-
ep0 step 40/
|
| 7 |
-
ep0 step 50/
|
| 8 |
-
ep0 step 60/
|
| 9 |
-
ep0 step 70/
|
| 10 |
-
ep0 step 80/
|
| 11 |
-
ep0 step 90/
|
| 12 |
-
ep0 step 100/
|
| 13 |
-
ep0 step 110/
|
| 14 |
-
ep0 step 120/
|
| 15 |
-
ep0 step 130/
|
| 16 |
-
ep0 step 140/
|
| 17 |
-
ep0 step 150/
|
| 18 |
-
ep0 step 160/
|
| 19 |
-
ep0 step 170/
|
| 20 |
-
ep0 step 180/
|
| 21 |
-
ep0 step 190/
|
| 22 |
-
ep0 step 200/
|
| 23 |
-
ep0 step 210/
|
| 24 |
-
ep0 step 220/
|
| 25 |
-
ep0 step 230/
|
| 26 |
-
ep0 step 240/
|
| 27 |
-
ep0 step 250/
|
| 28 |
-
ep0 step 260/
|
| 29 |
-
ep0 step 270/
|
| 30 |
-
ep0 step 280/
|
| 31 |
-
ep0 step 290/
|
| 32 |
-
ep0 step 300/
|
| 33 |
-
ep0 step 310/
|
| 34 |
-
ep0 step 320/
|
| 35 |
-
ep0 step 330/
|
| 36 |
-
ep0 step 340/
|
| 37 |
-
ep0 step 350/
|
| 38 |
-
ep0 step 360/
|
| 39 |
-
ep0 step 370/
|
| 40 |
-
ep0 step 380/
|
| 41 |
-
ep0 step 390/
|
| 42 |
-
ep0 step 400/
|
| 43 |
-
ep0 step 410/
|
| 44 |
-
ep0 step 420/
|
| 45 |
-
ep0 step 430/
|
| 46 |
-
ep0 step 440/
|
| 47 |
-
ep0 step 450/
|
| 48 |
-
ep0 step 460/
|
| 49 |
-
ep0 step 470/
|
| 50 |
-
ep0 step 480/
|
| 51 |
-
ep0 step 490/
|
| 52 |
-
ep0 step 500/
|
| 53 |
-
ep0 step 510/
|
| 54 |
-
ep0 step 520/
|
| 55 |
-
ep0 step 530/
|
| 56 |
-
ep0 step 540/
|
| 57 |
-
ep0 step 550/
|
| 58 |
-
ep0 step 560/
|
| 59 |
-
ep0 step 570/
|
| 60 |
-
ep0 step 580/
|
| 61 |
-
ep0 step 590/
|
| 62 |
-
ep0 step 600/
|
| 63 |
-
ep0 step 610/
|
| 64 |
-
ep0 step 620/
|
| 65 |
-
ep0 step 630/
|
| 66 |
-
ep0 step 640/
|
| 67 |
-
ep0 step 650/
|
| 68 |
-
ep0 step 660/
|
| 69 |
-
ep0 step 670/
|
| 70 |
-
ep0 step 680/
|
| 71 |
-
ep0 step 690/
|
| 72 |
-
ep0 step 700/
|
| 73 |
-
ep0 step 710/
|
| 74 |
-
ep0 step 720/
|
| 75 |
-
ep0 step 730/
|
| 76 |
-
ep0 step 740/
|
| 77 |
-
ep0 step 750/
|
| 78 |
-
ep0 step 760/
|
| 79 |
-
ep0 step 770/
|
| 80 |
-
ep0 step 780/
|
| 81 |
-
ep0 step 790/
|
| 82 |
-
ep0 step 800/
|
| 83 |
-
ep0 step 810/
|
| 84 |
-
ep0 step 820/
|
| 85 |
-
ep0 step 830/
|
| 86 |
-
ep0 step 840/
|
| 87 |
-
ep0 step 850/
|
| 88 |
-
ep0 step 860/
|
| 89 |
-
ep0 step 870/
|
| 90 |
-
ep0 step 880/
|
| 91 |
-
ep0 step 890/
|
| 92 |
-
ep0 step 900/
|
| 93 |
-
ep0 step 910/
|
| 94 |
-
ep0 step 920/
|
| 95 |
-
ep0 step 930/
|
| 96 |
-
ep0 step 940/
|
| 97 |
-
ep0 step 950/
|
| 98 |
-
ep0 step 960/
|
| 99 |
-
ep0 step 970/
|
| 100 |
-
ep0 step 980/
|
| 101 |
-
ep0 step 990/
|
| 102 |
-
ep0 step 1000/
|
| 103 |
-
ep0 step 1010/
|
| 104 |
-
ep0 step 1020/
|
| 105 |
-
ep0 step 1030/
|
| 106 |
-
ep0 step 1040/
|
| 107 |
-
ep0 step 1050/
|
| 108 |
-
ep0 step 1060/
|
| 109 |
-
ep0 step 1070/
|
| 110 |
-
ep0 step 1080/
|
| 111 |
-
ep0 step 1090/
|
| 112 |
-
ep0 step 1100/
|
| 113 |
-
ep0 step 1110/
|
| 114 |
-
ep0 step 1120/
|
| 115 |
-
ep0 step 1130/
|
| 116 |
-
ep0 step 1140/
|
| 117 |
-
ep0 step 1150/
|
| 118 |
-
ep0 step 1160/
|
| 119 |
-
ep0 step 1170/
|
| 120 |
-
ep0 step 1180/
|
| 121 |
-
ep0 step 1190/
|
| 122 |
-
ep0 step 1200/
|
| 123 |
-
ep0 step 1210/
|
| 124 |
-
ep0 step 1220/
|
| 125 |
-
ep0 step 1230/
|
| 126 |
-
ep0 step 1240/
|
| 127 |
-
ep0 step 1250/
|
| 128 |
-
ep0 step 1260/
|
| 129 |
-
ep0 step 1270/
|
| 130 |
-
ep0 step 1280/
|
| 131 |
-
ep0 step 1290/
|
| 132 |
-
ep0 step 1300/
|
| 133 |
-
ep0 step 1310/
|
| 134 |
-
ep0 step 1320/
|
| 135 |
-
ep0 step 1330/
|
| 136 |
-
ep0 step 1340/
|
| 137 |
-
ep0 step 1350/
|
| 138 |
-
ep0 step 1360/
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
|
| 153 |
-
|
| 154 |
-
|
| 155 |
-
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
| 160 |
-
ep1 step 1580/
|
| 161 |
-
ep1 step 1590/
|
| 162 |
-
ep1 step 1600/
|
| 163 |
-
ep1 step 1610/
|
| 164 |
-
ep1 step 1620/
|
| 165 |
-
ep1 step 1630/
|
| 166 |
-
ep1 step 1640/
|
| 167 |
-
ep1 step 1650/
|
| 168 |
-
ep1 step 1660/
|
| 169 |
-
ep1 step 1670/
|
| 170 |
-
ep1 step 1680/
|
| 171 |
-
ep1 step 1690/
|
| 172 |
-
ep1 step 1700/
|
| 173 |
-
ep1 step 1710/
|
| 174 |
-
ep1 step 1720/
|
| 175 |
-
ep1 step 1730/
|
| 176 |
-
ep1 step 1740/
|
| 177 |
-
ep1 step 1750/
|
| 178 |
-
ep1 step 1760/
|
| 179 |
-
ep1 step 1770/
|
| 180 |
-
ep1 step 1780/
|
| 181 |
-
ep1 step 1790/
|
| 182 |
-
ep1 step 1800/
|
| 183 |
-
ep1 step 1810/
|
| 184 |
-
ep1 step 1820/
|
| 185 |
-
ep1 step 1830/
|
| 186 |
-
ep1 step 1840/
|
| 187 |
-
ep1 step 1850/
|
| 188 |
-
ep1 step 1860/
|
| 189 |
-
ep1 step 1870/
|
| 190 |
-
ep1 step 1880/
|
| 191 |
-
ep1 step 1890/
|
| 192 |
-
ep1 step 1900/
|
| 193 |
-
ep1 step 1910/
|
| 194 |
-
ep1 step 1920/
|
| 195 |
-
ep1 step 1930/
|
| 196 |
-
ep1 step 1940/
|
| 197 |
-
ep1 step 1950/
|
| 198 |
-
ep1 step 1960/
|
| 199 |
-
ep1 step 1970/
|
| 200 |
-
ep1 step 1980/
|
| 201 |
-
ep1 step 1990/
|
| 202 |
-
ep1 step 2000/
|
| 203 |
-
ep1 step 2010/
|
| 204 |
-
ep1 step 2020/
|
| 205 |
-
ep1 step 2030/
|
| 206 |
-
ep1 step 2040/
|
| 207 |
-
ep1 step 2050/
|
| 208 |
-
ep1 step 2060/
|
| 209 |
-
ep1 step 2070/
|
| 210 |
-
ep1 step 2080/
|
| 211 |
-
ep1 step 2090/
|
| 212 |
-
ep1 step 2100/
|
| 213 |
-
ep1 step 2110/
|
| 214 |
-
ep1 step 2120/
|
| 215 |
-
ep1 step 2130/
|
| 216 |
-
ep1 step 2140/
|
| 217 |
-
ep1 step 2150/
|
| 218 |
-
ep1 step 2160/
|
| 219 |
-
ep1 step 2170/
|
| 220 |
-
ep1 step 2180/
|
| 221 |
-
ep1 step 2190/
|
| 222 |
-
ep1 step 2200/
|
| 223 |
-
ep1 step 2210/
|
| 224 |
-
ep1 step 2220/
|
| 225 |
-
ep1 step 2230/
|
| 226 |
-
ep1 step 2240/
|
| 227 |
-
ep1 step 2250/
|
| 228 |
-
ep1 step 2260/
|
| 229 |
-
ep1 step 2270/
|
| 230 |
-
ep1 step 2280/
|
| 231 |
-
ep1 step 2290/
|
| 232 |
-
ep1 step 2300/
|
| 233 |
-
ep1 step 2310/
|
| 234 |
-
ep1 step 2320/
|
| 235 |
-
ep1 step 2330/
|
| 236 |
-
ep1 step 2340/
|
| 237 |
-
ep1 step 2350/
|
| 238 |
-
ep1 step 2360/
|
| 239 |
-
ep1 step 2370/
|
| 240 |
-
ep1 step 2380/
|
| 241 |
-
ep1 step 2390/
|
| 242 |
-
ep1 step 2400/
|
| 243 |
-
ep1 step 2410/
|
| 244 |
-
ep1 step 2420/
|
| 245 |
-
ep1 step 2430/
|
| 246 |
-
ep1 step 2440/
|
| 247 |
-
ep1 step 2450/
|
| 248 |
-
ep1 step 2460/
|
| 249 |
-
ep1 step 2470/
|
| 250 |
-
ep1 step 2480/
|
| 251 |
-
ep1 step 2490/
|
| 252 |
-
ep1 step 2500/
|
| 253 |
-
ep1 step 2510/
|
| 254 |
-
ep1 step 2520/
|
| 255 |
-
ep1 step 2530/
|
| 256 |
-
ep1 step 2540/
|
| 257 |
-
ep1 step 2550/
|
| 258 |
-
ep1 step 2560/
|
| 259 |
-
ep1 step 2570/
|
| 260 |
-
ep1 step 2580/
|
| 261 |
-
ep1 step 2590/
|
| 262 |
-
ep1 step 2600/
|
| 263 |
-
ep1 step 2610/
|
| 264 |
-
ep1 step 2620/
|
| 265 |
-
ep1 step 2630/
|
| 266 |
-
ep1 step 2640/
|
| 267 |
-
ep1 step 2650/
|
| 268 |
-
ep1 step 2660/
|
| 269 |
-
ep1 step 2670/
|
| 270 |
-
ep1 step 2680/
|
| 271 |
-
ep1 step 2690/
|
| 272 |
-
ep1 step 2700/
|
| 273 |
-
ep1 step 2710/
|
| 274 |
-
ep1 step 2720/
|
| 275 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
device=cuda trainable params=10.6M
|
| 2 |
+
12576 training requests (holdout=[]), questions by type {'score': 3448, 'noul': 5224, 'choice': 6904}
|
| 3 |
+
ep0 step 10/3144 loss 1.795 kl 0.000 anchor 0.000 0.034s/rec
|
| 4 |
+
ep0 step 20/3144 loss 1.875 kl 0.000 anchor 0.000 0.027s/rec
|
| 5 |
+
ep0 step 30/3144 loss 1.948 kl 0.000 anchor 0.000 0.025s/rec
|
| 6 |
+
ep0 step 40/3144 loss 1.790 kl 0.000 anchor 0.000 0.024s/rec
|
| 7 |
+
ep0 step 50/3144 loss 1.686 kl 0.000 anchor 0.000 0.024s/rec
|
| 8 |
+
ep0 step 60/3144 loss 1.551 kl 0.000 anchor 0.000 0.023s/rec
|
| 9 |
+
ep0 step 70/3144 loss 1.935 kl 0.000 anchor 0.000 0.023s/rec
|
| 10 |
+
ep0 step 80/3144 loss 1.590 kl 0.000 anchor 0.000 0.023s/rec
|
| 11 |
+
ep0 step 90/3144 loss 1.468 kl 0.000 anchor 0.000 0.022s/rec
|
| 12 |
+
ep0 step 100/3144 loss 1.362 kl 0.000 anchor 0.000 0.022s/rec
|
| 13 |
+
ep0 step 110/3144 loss 1.683 kl 0.000 anchor 0.000 0.021s/rec
|
| 14 |
+
ep0 step 120/3144 loss 1.762 kl 0.000 anchor 0.000 0.022s/rec
|
| 15 |
+
ep0 step 130/3144 loss 1.560 kl 0.000 anchor 0.000 0.021s/rec
|
| 16 |
+
ep0 step 140/3144 loss 1.430 kl 0.000 anchor 0.000 0.021s/rec
|
| 17 |
+
ep0 step 150/3144 loss 1.604 kl 0.000 anchor 0.000 0.021s/rec
|
| 18 |
+
ep0 step 160/3144 loss 1.428 kl 0.000 anchor 0.000 0.022s/rec
|
| 19 |
+
ep0 step 170/3144 loss 1.325 kl 0.000 anchor 0.000 0.022s/rec
|
| 20 |
+
ep0 step 180/3144 loss 1.089 kl 0.000 anchor 0.000 0.022s/rec
|
| 21 |
+
ep0 step 190/3144 loss 1.364 kl 0.000 anchor 0.000 0.022s/rec
|
| 22 |
+
ep0 step 200/3144 loss 1.199 kl 0.000 anchor 0.000 0.022s/rec
|
| 23 |
+
ep0 step 210/3144 loss 1.304 kl 0.000 anchor 0.000 0.022s/rec
|
| 24 |
+
ep0 step 220/3144 loss 1.004 kl 0.000 anchor 0.000 0.022s/rec
|
| 25 |
+
ep0 step 230/3144 loss 0.926 kl 0.000 anchor 0.000 0.022s/rec
|
| 26 |
+
ep0 step 240/3144 loss 1.061 kl 0.000 anchor 0.000 0.022s/rec
|
| 27 |
+
ep0 step 250/3144 loss 1.028 kl 0.000 anchor 0.000 0.022s/rec
|
| 28 |
+
ep0 step 260/3144 loss 1.001 kl 0.000 anchor 0.000 0.022s/rec
|
| 29 |
+
ep0 step 270/3144 loss 0.932 kl 0.000 anchor 0.000 0.022s/rec
|
| 30 |
+
ep0 step 280/3144 loss 0.861 kl 0.000 anchor 0.000 0.022s/rec
|
| 31 |
+
ep0 step 290/3144 loss 0.984 kl 0.000 anchor 0.000 0.023s/rec
|
| 32 |
+
ep0 step 300/3144 loss 0.998 kl 0.000 anchor 0.000 0.023s/rec
|
| 33 |
+
ep0 step 310/3144 loss 0.840 kl 0.000 anchor 0.000 0.023s/rec
|
| 34 |
+
ep0 step 320/3144 loss 0.777 kl 0.000 anchor 0.000 0.023s/rec
|
| 35 |
+
ep0 step 330/3144 loss 0.940 kl 0.000 anchor 0.000 0.023s/rec
|
| 36 |
+
ep0 step 340/3144 loss 0.931 kl 0.000 anchor 0.000 0.023s/rec
|
| 37 |
+
ep0 step 350/3144 loss 0.879 kl 0.000 anchor 0.000 0.023s/rec
|
| 38 |
+
ep0 step 360/3144 loss 0.738 kl 0.000 anchor 0.000 0.023s/rec
|
| 39 |
+
ep0 step 370/3144 loss 0.805 kl 0.000 anchor 0.000 0.023s/rec
|
| 40 |
+
ep0 step 380/3144 loss 0.730 kl 0.000 anchor 0.000 0.023s/rec
|
| 41 |
+
ep0 step 390/3144 loss 1.098 kl 0.000 anchor 0.000 0.023s/rec
|
| 42 |
+
ep0 step 400/3144 loss 1.247 kl 0.000 anchor 0.000 0.023s/rec
|
| 43 |
+
ep0 step 410/3144 loss 0.931 kl 0.000 anchor 0.000 0.023s/rec
|
| 44 |
+
ep0 step 420/3144 loss 0.782 kl 0.000 anchor 0.000 0.023s/rec
|
| 45 |
+
ep0 step 430/3144 loss 0.757 kl 0.000 anchor 0.000 0.023s/rec
|
| 46 |
+
ep0 step 440/3144 loss 0.847 kl 0.000 anchor 0.000 0.023s/rec
|
| 47 |
+
ep0 step 450/3144 loss 0.649 kl 0.000 anchor 0.000 0.023s/rec
|
| 48 |
+
ep0 step 460/3144 loss 0.729 kl 0.000 anchor 0.000 0.023s/rec
|
| 49 |
+
ep0 step 470/3144 loss 0.784 kl 0.000 anchor 0.000 0.023s/rec
|
| 50 |
+
ep0 step 480/3144 loss 0.626 kl 0.000 anchor 0.000 0.023s/rec
|
| 51 |
+
ep0 step 490/3144 loss 0.696 kl 0.000 anchor 0.000 0.023s/rec
|
| 52 |
+
ep0 step 500/3144 loss 0.848 kl 0.000 anchor 0.000 0.023s/rec
|
| 53 |
+
ep0 step 510/3144 loss 0.977 kl 0.000 anchor 0.000 0.023s/rec
|
| 54 |
+
ep0 step 520/3144 loss 0.742 kl 0.000 anchor 0.000 0.023s/rec
|
| 55 |
+
ep0 step 530/3144 loss 1.122 kl 0.000 anchor 0.000 0.023s/rec
|
| 56 |
+
ep0 step 540/3144 loss 0.631 kl 0.000 anchor 0.000 0.023s/rec
|
| 57 |
+
ep0 step 550/3144 loss 0.627 kl 0.000 anchor 0.000 0.023s/rec
|
| 58 |
+
ep0 step 560/3144 loss 0.993 kl 0.000 anchor 0.000 0.023s/rec
|
| 59 |
+
ep0 step 570/3144 loss 0.778 kl 0.000 anchor 0.000 0.023s/rec
|
| 60 |
+
ep0 step 580/3144 loss 0.607 kl 0.000 anchor 0.000 0.023s/rec
|
| 61 |
+
ep0 step 590/3144 loss 0.908 kl 0.000 anchor 0.000 0.023s/rec
|
| 62 |
+
ep0 step 600/3144 loss 0.611 kl 0.000 anchor 0.000 0.023s/rec
|
| 63 |
+
ep0 step 610/3144 loss 0.503 kl 0.000 anchor 0.000 0.023s/rec
|
| 64 |
+
ep0 step 620/3144 loss 0.611 kl 0.000 anchor 0.000 0.023s/rec
|
| 65 |
+
ep0 step 630/3144 loss 0.755 kl 0.000 anchor 0.000 0.023s/rec
|
| 66 |
+
ep0 step 640/3144 loss 0.804 kl 0.000 anchor 0.000 0.023s/rec
|
| 67 |
+
ep0 step 650/3144 loss 0.795 kl 0.000 anchor 0.000 0.023s/rec
|
| 68 |
+
ep0 step 660/3144 loss 0.853 kl 0.000 anchor 0.000 0.023s/rec
|
| 69 |
+
ep0 step 670/3144 loss 0.919 kl 0.000 anchor 0.000 0.023s/rec
|
| 70 |
+
ep0 step 680/3144 loss 0.748 kl 0.000 anchor 0.000 0.023s/rec
|
| 71 |
+
ep0 step 690/3144 loss 0.715 kl 0.000 anchor 0.000 0.023s/rec
|
| 72 |
+
ep0 step 700/3144 loss 0.708 kl 0.000 anchor 0.000 0.023s/rec
|
| 73 |
+
ep0 step 710/3144 loss 0.534 kl 0.000 anchor 0.000 0.023s/rec
|
| 74 |
+
ep0 step 720/3144 loss 0.430 kl 0.000 anchor 0.000 0.022s/rec
|
| 75 |
+
ep0 step 730/3144 loss 0.902 kl 0.000 anchor 0.000 0.022s/rec
|
| 76 |
+
ep0 step 740/3144 loss 0.615 kl 0.000 anchor 0.000 0.022s/rec
|
| 77 |
+
ep0 step 750/3144 loss 0.746 kl 0.000 anchor 0.000 0.022s/rec
|
| 78 |
+
ep0 step 760/3144 loss 0.530 kl 0.000 anchor 0.000 0.022s/rec
|
| 79 |
+
ep0 step 770/3144 loss 0.571 kl 0.000 anchor 0.000 0.022s/rec
|
| 80 |
+
ep0 step 780/3144 loss 0.626 kl 0.000 anchor 0.000 0.022s/rec
|
| 81 |
+
ep0 step 790/3144 loss 0.675 kl 0.000 anchor 0.000 0.022s/rec
|
| 82 |
+
ep0 step 800/3144 loss 0.738 kl 0.000 anchor 0.000 0.022s/rec
|
| 83 |
+
ep0 step 810/3144 loss 0.567 kl 0.000 anchor 0.000 0.022s/rec
|
| 84 |
+
ep0 step 820/3144 loss 0.512 kl 0.000 anchor 0.000 0.022s/rec
|
| 85 |
+
ep0 step 830/3144 loss 0.559 kl 0.000 anchor 0.000 0.022s/rec
|
| 86 |
+
ep0 step 840/3144 loss 0.784 kl 0.000 anchor 0.000 0.022s/rec
|
| 87 |
+
ep0 step 850/3144 loss 0.738 kl 0.000 anchor 0.000 0.022s/rec
|
| 88 |
+
ep0 step 860/3144 loss 0.572 kl 0.000 anchor 0.000 0.022s/rec
|
| 89 |
+
ep0 step 870/3144 loss 0.867 kl 0.000 anchor 0.000 0.022s/rec
|
| 90 |
+
ep0 step 880/3144 loss 0.742 kl 0.000 anchor 0.000 0.022s/rec
|
| 91 |
+
ep0 step 890/3144 loss 0.541 kl 0.000 anchor 0.000 0.022s/rec
|
| 92 |
+
ep0 step 900/3144 loss 1.086 kl 0.000 anchor 0.000 0.022s/rec
|
| 93 |
+
ep0 step 910/3144 loss 0.744 kl 0.000 anchor 0.000 0.022s/rec
|
| 94 |
+
ep0 step 920/3144 loss 0.467 kl 0.000 anchor 0.000 0.022s/rec
|
| 95 |
+
ep0 step 930/3144 loss 0.477 kl 0.000 anchor 0.000 0.022s/rec
|
| 96 |
+
ep0 step 940/3144 loss 0.888 kl 0.000 anchor 0.000 0.022s/rec
|
| 97 |
+
ep0 step 950/3144 loss 0.436 kl 0.000 anchor 0.000 0.022s/rec
|
| 98 |
+
ep0 step 960/3144 loss 1.119 kl 0.000 anchor 0.000 0.022s/rec
|
| 99 |
+
ep0 step 970/3144 loss 0.702 kl 0.000 anchor 0.000 0.022s/rec
|
| 100 |
+
ep0 step 980/3144 loss 0.628 kl 0.000 anchor 0.000 0.022s/rec
|
| 101 |
+
ep0 step 990/3144 loss 0.583 kl 0.000 anchor 0.000 0.022s/rec
|
| 102 |
+
ep0 step 1000/3144 loss 0.710 kl 0.000 anchor 0.000 0.022s/rec
|
| 103 |
+
ep0 step 1010/3144 loss 0.692 kl 0.000 anchor 0.000 0.022s/rec
|
| 104 |
+
ep0 step 1020/3144 loss 0.662 kl 0.000 anchor 0.000 0.022s/rec
|
| 105 |
+
ep0 step 1030/3144 loss 0.611 kl 0.000 anchor 0.000 0.022s/rec
|
| 106 |
+
ep0 step 1040/3144 loss 0.772 kl 0.000 anchor 0.000 0.022s/rec
|
| 107 |
+
ep0 step 1050/3144 loss 0.444 kl 0.000 anchor 0.000 0.022s/rec
|
| 108 |
+
ep0 step 1060/3144 loss 0.563 kl 0.000 anchor 0.000 0.022s/rec
|
| 109 |
+
ep0 step 1070/3144 loss 0.568 kl 0.000 anchor 0.000 0.022s/rec
|
| 110 |
+
ep0 step 1080/3144 loss 0.770 kl 0.000 anchor 0.000 0.022s/rec
|
| 111 |
+
ep0 step 1090/3144 loss 0.601 kl 0.000 anchor 0.000 0.022s/rec
|
| 112 |
+
ep0 step 1100/3144 loss 0.577 kl 0.000 anchor 0.000 0.022s/rec
|
| 113 |
+
ep0 step 1110/3144 loss 0.634 kl 0.000 anchor 0.000 0.022s/rec
|
| 114 |
+
ep0 step 1120/3144 loss 0.738 kl 0.000 anchor 0.000 0.022s/rec
|
| 115 |
+
ep0 step 1130/3144 loss 0.566 kl 0.000 anchor 0.000 0.022s/rec
|
| 116 |
+
ep0 step 1140/3144 loss 0.494 kl 0.000 anchor 0.000 0.022s/rec
|
| 117 |
+
ep0 step 1150/3144 loss 0.548 kl 0.000 anchor 0.000 0.022s/rec
|
| 118 |
+
ep0 step 1160/3144 loss 1.086 kl 0.000 anchor 0.000 0.022s/rec
|
| 119 |
+
ep0 step 1170/3144 loss 0.643 kl 0.000 anchor 0.000 0.022s/rec
|
| 120 |
+
ep0 step 1180/3144 loss 0.578 kl 0.000 anchor 0.000 0.022s/rec
|
| 121 |
+
ep0 step 1190/3144 loss 0.587 kl 0.000 anchor 0.000 0.022s/rec
|
| 122 |
+
ep0 step 1200/3144 loss 0.602 kl 0.000 anchor 0.000 0.022s/rec
|
| 123 |
+
ep0 step 1210/3144 loss 0.559 kl 0.000 anchor 0.000 0.022s/rec
|
| 124 |
+
ep0 step 1220/3144 loss 0.718 kl 0.000 anchor 0.000 0.022s/rec
|
| 125 |
+
ep0 step 1230/3144 loss 0.625 kl 0.000 anchor 0.000 0.022s/rec
|
| 126 |
+
ep0 step 1240/3144 loss 0.493 kl 0.000 anchor 0.000 0.022s/rec
|
| 127 |
+
ep0 step 1250/3144 loss 0.717 kl 0.000 anchor 0.000 0.022s/rec
|
| 128 |
+
ep0 step 1260/3144 loss 0.512 kl 0.000 anchor 0.000 0.022s/rec
|
| 129 |
+
ep0 step 1270/3144 loss 0.643 kl 0.000 anchor 0.000 0.022s/rec
|
| 130 |
+
ep0 step 1280/3144 loss 0.673 kl 0.000 anchor 0.000 0.022s/rec
|
| 131 |
+
ep0 step 1290/3144 loss 0.789 kl 0.000 anchor 0.000 0.022s/rec
|
| 132 |
+
ep0 step 1300/3144 loss 0.838 kl 0.000 anchor 0.000 0.022s/rec
|
| 133 |
+
ep0 step 1310/3144 loss 0.614 kl 0.000 anchor 0.000 0.022s/rec
|
| 134 |
+
ep0 step 1320/3144 loss 0.578 kl 0.000 anchor 0.000 0.022s/rec
|
| 135 |
+
ep0 step 1330/3144 loss 0.826 kl 0.000 anchor 0.000 0.022s/rec
|
| 136 |
+
ep0 step 1340/3144 loss 0.763 kl 0.000 anchor 0.000 0.022s/rec
|
| 137 |
+
ep0 step 1350/3144 loss 0.327 kl 0.000 anchor 0.000 0.022s/rec
|
| 138 |
+
ep0 step 1360/3144 loss 0.850 kl 0.000 anchor 0.000 0.022s/rec
|
| 139 |
+
ep0 step 1370/3144 loss 0.667 kl 0.000 anchor 0.000 0.022s/rec
|
| 140 |
+
ep0 step 1380/3144 loss 0.667 kl 0.000 anchor 0.000 0.022s/rec
|
| 141 |
+
ep0 step 1390/3144 loss 0.494 kl 0.000 anchor 0.000 0.022s/rec
|
| 142 |
+
ep0 step 1400/3144 loss 0.710 kl 0.000 anchor 0.000 0.022s/rec
|
| 143 |
+
ep0 step 1410/3144 loss 0.414 kl 0.000 anchor 0.000 0.022s/rec
|
| 144 |
+
ep0 step 1420/3144 loss 0.586 kl 0.000 anchor 0.000 0.022s/rec
|
| 145 |
+
ep0 step 1430/3144 loss 0.527 kl 0.000 anchor 0.000 0.022s/rec
|
| 146 |
+
ep0 step 1440/3144 loss 0.608 kl 0.000 anchor 0.000 0.022s/rec
|
| 147 |
+
ep0 step 1450/3144 loss 0.831 kl 0.000 anchor 0.000 0.022s/rec
|
| 148 |
+
ep0 step 1460/3144 loss 0.619 kl 0.000 anchor 0.000 0.022s/rec
|
| 149 |
+
ep0 step 1470/3144 loss 0.670 kl 0.000 anchor 0.000 0.022s/rec
|
| 150 |
+
ep0 step 1480/3144 loss 0.535 kl 0.000 anchor 0.000 0.022s/rec
|
| 151 |
+
ep0 step 1490/3144 loss 0.619 kl 0.000 anchor 0.000 0.022s/rec
|
| 152 |
+
ep0 step 1500/3144 loss 0.625 kl 0.000 anchor 0.000 0.022s/rec
|
| 153 |
+
ep0 step 1510/3144 loss 0.563 kl 0.000 anchor 0.000 0.021s/rec
|
| 154 |
+
ep0 step 1520/3144 loss 0.677 kl 0.000 anchor 0.000 0.021s/rec
|
| 155 |
+
ep0 step 1530/3144 loss 0.457 kl 0.000 anchor 0.000 0.021s/rec
|
| 156 |
+
ep0 step 1540/3144 loss 0.568 kl 0.000 anchor 0.000 0.021s/rec
|
| 157 |
+
ep0 step 1550/3144 loss 0.532 kl 0.000 anchor 0.000 0.021s/rec
|
| 158 |
+
ep0 step 1560/3144 loss 0.592 kl 0.000 anchor 0.000 0.021s/rec
|
| 159 |
+
ep0 step 1570/3144 loss 0.807 kl 0.000 anchor 0.000 0.021s/rec
|
| 160 |
+
ep1 step 1580/3144 loss 0.668 kl 0.000 anchor 0.000 0.021s/rec
|
| 161 |
+
ep1 step 1590/3144 loss 0.312 kl 0.000 anchor 0.000 0.021s/rec
|
| 162 |
+
ep1 step 1600/3144 loss 0.450 kl 0.000 anchor 0.000 0.021s/rec
|
| 163 |
+
ep1 step 1610/3144 loss 0.367 kl 0.000 anchor 0.000 0.021s/rec
|
| 164 |
+
ep1 step 1620/3144 loss 0.667 kl 0.000 anchor 0.000 0.021s/rec
|
| 165 |
+
ep1 step 1630/3144 loss 0.465 kl 0.000 anchor 0.000 0.021s/rec
|
| 166 |
+
ep1 step 1640/3144 loss 0.376 kl 0.000 anchor 0.000 0.021s/rec
|
| 167 |
+
ep1 step 1650/3144 loss 0.464 kl 0.000 anchor 0.000 0.021s/rec
|
| 168 |
+
ep1 step 1660/3144 loss 0.354 kl 0.000 anchor 0.000 0.021s/rec
|
| 169 |
+
ep1 step 1670/3144 loss 0.652 kl 0.000 anchor 0.000 0.021s/rec
|
| 170 |
+
ep1 step 1680/3144 loss 0.432 kl 0.000 anchor 0.000 0.021s/rec
|
| 171 |
+
ep1 step 1690/3144 loss 0.394 kl 0.000 anchor 0.000 0.021s/rec
|
| 172 |
+
ep1 step 1700/3144 loss 0.524 kl 0.000 anchor 0.000 0.021s/rec
|
| 173 |
+
ep1 step 1710/3144 loss 0.358 kl 0.000 anchor 0.000 0.021s/rec
|
| 174 |
+
ep1 step 1720/3144 loss 1.152 kl 0.000 anchor 0.000 0.021s/rec
|
| 175 |
+
ep1 step 1730/3144 loss 0.465 kl 0.000 anchor 0.000 0.021s/rec
|
| 176 |
+
ep1 step 1740/3144 loss 0.358 kl 0.000 anchor 0.000 0.021s/rec
|
| 177 |
+
ep1 step 1750/3144 loss 0.351 kl 0.000 anchor 0.000 0.021s/rec
|
| 178 |
+
ep1 step 1760/3144 loss 0.398 kl 0.000 anchor 0.000 0.021s/rec
|
| 179 |
+
ep1 step 1770/3144 loss 0.676 kl 0.000 anchor 0.000 0.021s/rec
|
| 180 |
+
ep1 step 1780/3144 loss 0.438 kl 0.000 anchor 0.000 0.021s/rec
|
| 181 |
+
ep1 step 1790/3144 loss 0.355 kl 0.000 anchor 0.000 0.021s/rec
|
| 182 |
+
ep1 step 1800/3144 loss 0.536 kl 0.000 anchor 0.000 0.021s/rec
|
| 183 |
+
ep1 step 1810/3144 loss 0.432 kl 0.000 anchor 0.000 0.021s/rec
|
| 184 |
+
ep1 step 1820/3144 loss 0.291 kl 0.000 anchor 0.000 0.021s/rec
|
| 185 |
+
ep1 step 1830/3144 loss 0.505 kl 0.000 anchor 0.000 0.021s/rec
|
| 186 |
+
ep1 step 1840/3144 loss 0.439 kl 0.000 anchor 0.000 0.021s/rec
|
| 187 |
+
ep1 step 1850/3144 loss 0.621 kl 0.000 anchor 0.000 0.021s/rec
|
| 188 |
+
ep1 step 1860/3144 loss 0.644 kl 0.000 anchor 0.000 0.021s/rec
|
| 189 |
+
ep1 step 1870/3144 loss 0.407 kl 0.000 anchor 0.000 0.021s/rec
|
| 190 |
+
ep1 step 1880/3144 loss 0.360 kl 0.000 anchor 0.000 0.021s/rec
|
| 191 |
+
ep1 step 1890/3144 loss 0.394 kl 0.000 anchor 0.000 0.021s/rec
|
| 192 |
+
ep1 step 1900/3144 loss 0.547 kl 0.000 anchor 0.000 0.021s/rec
|
| 193 |
+
ep1 step 1910/3144 loss 0.518 kl 0.000 anchor 0.000 0.021s/rec
|
| 194 |
+
ep1 step 1920/3144 loss 0.461 kl 0.000 anchor 0.000 0.021s/rec
|
| 195 |
+
ep1 step 1930/3144 loss 0.333 kl 0.000 anchor 0.000 0.021s/rec
|
| 196 |
+
ep1 step 1940/3144 loss 0.486 kl 0.000 anchor 0.000 0.021s/rec
|
| 197 |
+
ep1 step 1950/3144 loss 0.934 kl 0.000 anchor 0.000 0.021s/rec
|
| 198 |
+
ep1 step 1960/3144 loss 0.412 kl 0.000 anchor 0.000 0.021s/rec
|
| 199 |
+
ep1 step 1970/3144 loss 0.464 kl 0.000 anchor 0.000 0.021s/rec
|
| 200 |
+
ep1 step 1980/3144 loss 0.627 kl 0.000 anchor 0.000 0.021s/rec
|
| 201 |
+
ep1 step 1990/3144 loss 0.482 kl 0.000 anchor 0.000 0.021s/rec
|
| 202 |
+
ep1 step 2000/3144 loss 0.326 kl 0.000 anchor 0.000 0.021s/rec
|
| 203 |
+
ep1 step 2010/3144 loss 0.796 kl 0.000 anchor 0.000 0.021s/rec
|
| 204 |
+
ep1 step 2020/3144 loss 0.554 kl 0.000 anchor 0.000 0.021s/rec
|
| 205 |
+
ep1 step 2030/3144 loss 0.697 kl 0.000 anchor 0.000 0.021s/rec
|
| 206 |
+
ep1 step 2040/3144 loss 0.319 kl 0.000 anchor 0.000 0.021s/rec
|
| 207 |
+
ep1 step 2050/3144 loss 0.298 kl 0.000 anchor 0.000 0.021s/rec
|
| 208 |
+
ep1 step 2060/3144 loss 0.464 kl 0.000 anchor 0.000 0.021s/rec
|
| 209 |
+
ep1 step 2070/3144 loss 0.308 kl 0.000 anchor 0.000 0.021s/rec
|
| 210 |
+
ep1 step 2080/3144 loss 0.439 kl 0.000 anchor 0.000 0.021s/rec
|
| 211 |
+
ep1 step 2090/3144 loss 0.243 kl 0.000 anchor 0.000 0.021s/rec
|
| 212 |
+
ep1 step 2100/3144 loss 0.654 kl 0.000 anchor 0.000 0.021s/rec
|
| 213 |
+
ep1 step 2110/3144 loss 0.357 kl 0.000 anchor 0.000 0.021s/rec
|
| 214 |
+
ep1 step 2120/3144 loss 0.567 kl 0.000 anchor 0.000 0.021s/rec
|
| 215 |
+
ep1 step 2130/3144 loss 0.389 kl 0.000 anchor 0.000 0.021s/rec
|
| 216 |
+
ep1 step 2140/3144 loss 0.499 kl 0.000 anchor 0.000 0.021s/rec
|
| 217 |
+
ep1 step 2150/3144 loss 0.503 kl 0.000 anchor 0.000 0.021s/rec
|
| 218 |
+
ep1 step 2160/3144 loss 0.405 kl 0.000 anchor 0.000 0.021s/rec
|
| 219 |
+
ep1 step 2170/3144 loss 0.202 kl 0.000 anchor 0.000 0.021s/rec
|
| 220 |
+
ep1 step 2180/3144 loss 0.357 kl 0.000 anchor 0.000 0.021s/rec
|
| 221 |
+
ep1 step 2190/3144 loss 0.451 kl 0.000 anchor 0.000 0.021s/rec
|
| 222 |
+
ep1 step 2200/3144 loss 0.435 kl 0.000 anchor 0.000 0.021s/rec
|
| 223 |
+
ep1 step 2210/3144 loss 0.441 kl 0.000 anchor 0.000 0.021s/rec
|
| 224 |
+
ep1 step 2220/3144 loss 0.590 kl 0.000 anchor 0.000 0.021s/rec
|
| 225 |
+
ep1 step 2230/3144 loss 0.418 kl 0.000 anchor 0.000 0.021s/rec
|
| 226 |
+
ep1 step 2240/3144 loss 0.648 kl 0.000 anchor 0.000 0.021s/rec
|
| 227 |
+
ep1 step 2250/3144 loss 0.303 kl 0.000 anchor 0.000 0.021s/rec
|
| 228 |
+
ep1 step 2260/3144 loss 0.317 kl 0.000 anchor 0.000 0.021s/rec
|
| 229 |
+
ep1 step 2270/3144 loss 0.477 kl 0.000 anchor 0.000 0.021s/rec
|
| 230 |
+
ep1 step 2280/3144 loss 0.575 kl 0.000 anchor 0.000 0.021s/rec
|
| 231 |
+
ep1 step 2290/3144 loss 0.244 kl 0.000 anchor 0.000 0.021s/rec
|
| 232 |
+
ep1 step 2300/3144 loss 0.382 kl 0.000 anchor 0.000 0.021s/rec
|
| 233 |
+
ep1 step 2310/3144 loss 0.503 kl 0.000 anchor 0.000 0.021s/rec
|
| 234 |
+
ep1 step 2320/3144 loss 0.308 kl 0.000 anchor 0.000 0.021s/rec
|
| 235 |
+
ep1 step 2330/3144 loss 0.576 kl 0.000 anchor 0.000 0.021s/rec
|
| 236 |
+
ep1 step 2340/3144 loss 0.345 kl 0.000 anchor 0.000 0.021s/rec
|
| 237 |
+
ep1 step 2350/3144 loss 0.423 kl 0.000 anchor 0.000 0.021s/rec
|
| 238 |
+
ep1 step 2360/3144 loss 0.441 kl 0.000 anchor 0.000 0.021s/rec
|
| 239 |
+
ep1 step 2370/3144 loss 0.598 kl 0.000 anchor 0.000 0.021s/rec
|
| 240 |
+
ep1 step 2380/3144 loss 0.504 kl 0.000 anchor 0.000 0.021s/rec
|
| 241 |
+
ep1 step 2390/3144 loss 0.297 kl 0.000 anchor 0.000 0.021s/rec
|
| 242 |
+
ep1 step 2400/3144 loss 0.246 kl 0.000 anchor 0.000 0.021s/rec
|
| 243 |
+
ep1 step 2410/3144 loss 0.345 kl 0.000 anchor 0.000 0.021s/rec
|
| 244 |
+
ep1 step 2420/3144 loss 0.462 kl 0.000 anchor 0.000 0.021s/rec
|
| 245 |
+
ep1 step 2430/3144 loss 0.407 kl 0.000 anchor 0.000 0.021s/rec
|
| 246 |
+
ep1 step 2440/3144 loss 0.534 kl 0.000 anchor 0.000 0.021s/rec
|
| 247 |
+
ep1 step 2450/3144 loss 0.344 kl 0.000 anchor 0.000 0.021s/rec
|
| 248 |
+
ep1 step 2460/3144 loss 0.377 kl 0.000 anchor 0.000 0.021s/rec
|
| 249 |
+
ep1 step 2470/3144 loss 0.428 kl 0.000 anchor 0.000 0.021s/rec
|
| 250 |
+
ep1 step 2480/3144 loss 0.417 kl 0.000 anchor 0.000 0.021s/rec
|
| 251 |
+
ep1 step 2490/3144 loss 0.686 kl 0.000 anchor 0.000 0.021s/rec
|
| 252 |
+
ep1 step 2500/3144 loss 0.384 kl 0.000 anchor 0.000 0.021s/rec
|
| 253 |
+
ep1 step 2510/3144 loss 0.576 kl 0.000 anchor 0.000 0.021s/rec
|
| 254 |
+
ep1 step 2520/3144 loss 0.377 kl 0.000 anchor 0.000 0.021s/rec
|
| 255 |
+
ep1 step 2530/3144 loss 0.392 kl 0.000 anchor 0.000 0.021s/rec
|
| 256 |
+
ep1 step 2540/3144 loss 0.341 kl 0.000 anchor 0.000 0.021s/rec
|
| 257 |
+
ep1 step 2550/3144 loss 0.439 kl 0.000 anchor 0.000 0.021s/rec
|
| 258 |
+
ep1 step 2560/3144 loss 0.467 kl 0.000 anchor 0.000 0.021s/rec
|
| 259 |
+
ep1 step 2570/3144 loss 0.527 kl 0.000 anchor 0.000 0.021s/rec
|
| 260 |
+
ep1 step 2580/3144 loss 0.337 kl 0.000 anchor 0.000 0.021s/rec
|
| 261 |
+
ep1 step 2590/3144 loss 0.822 kl 0.000 anchor 0.000 0.021s/rec
|
| 262 |
+
ep1 step 2600/3144 loss 0.363 kl 0.000 anchor 0.000 0.021s/rec
|
| 263 |
+
ep1 step 2610/3144 loss 0.670 kl 0.000 anchor 0.000 0.021s/rec
|
| 264 |
+
ep1 step 2620/3144 loss 0.254 kl 0.000 anchor 0.000 0.021s/rec
|
| 265 |
+
ep1 step 2630/3144 loss 0.373 kl 0.000 anchor 0.000 0.021s/rec
|
| 266 |
+
ep1 step 2640/3144 loss 0.635 kl 0.000 anchor 0.000 0.021s/rec
|
| 267 |
+
ep1 step 2650/3144 loss 0.474 kl 0.000 anchor 0.000 0.021s/rec
|
| 268 |
+
ep1 step 2660/3144 loss 0.331 kl 0.000 anchor 0.000 0.021s/rec
|
| 269 |
+
ep1 step 2670/3144 loss 0.423 kl 0.000 anchor 0.000 0.021s/rec
|
| 270 |
+
ep1 step 2680/3144 loss 0.484 kl 0.000 anchor 0.000 0.021s/rec
|
| 271 |
+
ep1 step 2690/3144 loss 0.420 kl 0.000 anchor 0.000 0.021s/rec
|
| 272 |
+
ep1 step 2700/3144 loss 0.479 kl 0.000 anchor 0.000 0.021s/rec
|
| 273 |
+
ep1 step 2710/3144 loss 0.364 kl 0.000 anchor 0.000 0.021s/rec
|
| 274 |
+
ep1 step 2720/3144 loss 0.341 kl 0.000 anchor 0.000 0.021s/rec
|
| 275 |
+
ep1 step 2730/3144 loss 0.251 kl 0.000 anchor 0.000 0.021s/rec
|
| 276 |
+
ep1 step 2740/3144 loss 0.485 kl 0.000 anchor 0.000 0.021s/rec
|
| 277 |
+
ep1 step 2750/3144 loss 0.343 kl 0.000 anchor 0.000 0.021s/rec
|
| 278 |
+
ep1 step 2760/3144 loss 0.441 kl 0.000 anchor 0.000 0.021s/rec
|
| 279 |
+
ep1 step 2770/3144 loss 0.324 kl 0.000 anchor 0.000 0.021s/rec
|
| 280 |
+
ep1 step 2780/3144 loss 0.330 kl 0.000 anchor 0.000 0.021s/rec
|
| 281 |
+
ep1 step 2790/3144 loss 0.232 kl 0.000 anchor 0.000 0.021s/rec
|
| 282 |
+
ep1 step 2800/3144 loss 0.286 kl 0.000 anchor 0.000 0.021s/rec
|
| 283 |
+
ep1 step 2810/3144 loss 0.214 kl 0.000 anchor 0.000 0.021s/rec
|
| 284 |
+
ep1 step 2820/3144 loss 0.424 kl 0.000 anchor 0.000 0.021s/rec
|
| 285 |
+
ep1 step 2830/3144 loss 0.538 kl 0.000 anchor 0.000 0.021s/rec
|
| 286 |
+
ep1 step 2840/3144 loss 0.397 kl 0.000 anchor 0.000 0.021s/rec
|
| 287 |
+
ep1 step 2850/3144 loss 0.231 kl 0.000 anchor 0.000 0.021s/rec
|
| 288 |
+
ep1 step 2860/3144 loss 0.383 kl 0.000 anchor 0.000 0.021s/rec
|
| 289 |
+
ep1 step 2870/3144 loss 0.388 kl 0.000 anchor 0.000 0.021s/rec
|
| 290 |
+
ep1 step 2880/3144 loss 0.581 kl 0.000 anchor 0.000 0.021s/rec
|
| 291 |
+
ep1 step 2890/3144 loss 0.770 kl 0.000 anchor 0.000 0.021s/rec
|
| 292 |
+
ep1 step 2900/3144 loss 0.503 kl 0.000 anchor 0.000 0.021s/rec
|
| 293 |
+
ep1 step 2910/3144 loss 0.142 kl 0.000 anchor 0.000 0.021s/rec
|
| 294 |
+
ep1 step 2920/3144 loss 0.269 kl 0.000 anchor 0.000 0.021s/rec
|
| 295 |
+
ep1 step 2930/3144 loss 0.502 kl 0.000 anchor 0.000 0.021s/rec
|
| 296 |
+
ep1 step 2940/3144 loss 0.706 kl 0.000 anchor 0.000 0.021s/rec
|
| 297 |
+
ep1 step 2950/3144 loss 0.495 kl 0.000 anchor 0.000 0.021s/rec
|
| 298 |
+
ep1 step 2960/3144 loss 0.493 kl 0.000 anchor 0.000 0.021s/rec
|
| 299 |
+
ep1 step 2970/3144 loss 0.379 kl 0.000 anchor 0.000 0.021s/rec
|
| 300 |
+
ep1 step 2980/3144 loss 0.239 kl 0.000 anchor 0.000 0.021s/rec
|
| 301 |
+
ep1 step 2990/3144 loss 0.511 kl 0.000 anchor 0.000 0.021s/rec
|
| 302 |
+
ep1 step 3000/3144 loss 0.584 kl 0.000 anchor 0.000 0.021s/rec
|
| 303 |
+
ep1 step 3010/3144 loss 0.341 kl 0.000 anchor 0.000 0.021s/rec
|
| 304 |
+
ep1 step 3020/3144 loss 0.396 kl 0.000 anchor 0.000 0.021s/rec
|
| 305 |
+
ep1 step 3030/3144 loss 0.490 kl 0.000 anchor 0.000 0.021s/rec
|
| 306 |
+
ep1 step 3040/3144 loss 0.269 kl 0.000 anchor 0.000 0.021s/rec
|
| 307 |
+
ep1 step 3050/3144 loss 0.250 kl 0.000 anchor 0.000 0.021s/rec
|
| 308 |
+
ep1 step 3060/3144 loss 0.407 kl 0.000 anchor 0.000 0.021s/rec
|
| 309 |
+
ep1 step 3070/3144 loss 0.289 kl 0.000 anchor 0.000 0.021s/rec
|
| 310 |
+
ep1 step 3080/3144 loss 0.499 kl 0.000 anchor 0.000 0.021s/rec
|
| 311 |
+
ep1 step 3090/3144 loss 0.354 kl 0.000 anchor 0.000 0.021s/rec
|
| 312 |
+
ep1 step 3100/3144 loss 0.414 kl 0.000 anchor 0.000 0.021s/rec
|
| 313 |
+
ep1 step 3110/3144 loss 0.701 kl 0.000 anchor 0.000 0.021s/rec
|
| 314 |
+
ep1 step 3120/3144 loss 0.499 kl 0.000 anchor 0.000 0.021s/rec
|
| 315 |
+
ep1 step 3130/3144 loss 0.413 kl 0.000 anchor 0.000 0.021s/rec
|
| 316 |
+
ep1 step 3140/3144 loss 0.319 kl 0.000 anchor 0.000 0.021s/rec
|
| 317 |
+
saved /runs/v7-06b/02-trial-2/checkpoint
|
training_config.json
CHANGED
|
@@ -3,28 +3,39 @@
|
|
| 3 |
"base": "Qwen/Qwen3-0.6B-Base",
|
| 4 |
"n_per_source": 1000,
|
| 5 |
"epochs": 2,
|
| 6 |
-
"lr": 0.
|
|
|
|
|
|
|
| 7 |
"lora": 16,
|
| 8 |
"accum": 1,
|
| 9 |
"holdout": "",
|
| 10 |
"perm_kl": 0.0,
|
| 11 |
"perm_frac": 0.3,
|
| 12 |
"ord_w": 0.0,
|
| 13 |
-
"suite": "/root/evals/
|
| 14 |
"train_sources": "",
|
| 15 |
"device": "cuda",
|
| 16 |
"batch": 8,
|
| 17 |
"dtype": "bf16",
|
| 18 |
"checkpointing": 0,
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
"base_revision": "",
|
| 20 |
"p_none": 0.1,
|
| 21 |
"p_none_distract": 0.12,
|
| 22 |
"p_distract": 0.15,
|
| 23 |
"p_none_pair": 0.25,
|
| 24 |
-
"
|
| 25 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
},
|
| 27 |
-
"suite_sha256": "
|
| 28 |
"base_revision": "da87bfb608c14b7cf20ba1ce41287e8de496c0cd",
|
| 29 |
"ordinal_objective": "ranked_probability_score",
|
| 30 |
"holdout": []
|
|
|
|
| 3 |
"base": "Qwen/Qwen3-0.6B-Base",
|
| 4 |
"n_per_source": 1000,
|
| 5 |
"epochs": 2,
|
| 6 |
+
"lr": 0.0001,
|
| 7 |
+
"head_lr": 0.0,
|
| 8 |
+
"weight_decay": 0.01,
|
| 9 |
"lora": 16,
|
| 10 |
"accum": 1,
|
| 11 |
"holdout": "",
|
| 12 |
"perm_kl": 0.0,
|
| 13 |
"perm_frac": 0.3,
|
| 14 |
"ord_w": 0.0,
|
| 15 |
+
"suite": "/root/evals/v7/decision-v7",
|
| 16 |
"train_sources": "",
|
| 17 |
"device": "cuda",
|
| 18 |
"batch": 8,
|
| 19 |
"dtype": "bf16",
|
| 20 |
"checkpointing": 0,
|
| 21 |
+
"option_isolation": 0,
|
| 22 |
+
"special_embeddings": 0,
|
| 23 |
+
"head_dim": 256,
|
| 24 |
+
"lora_targets": "all",
|
| 25 |
"base_revision": "",
|
| 26 |
"p_none": 0.1,
|
| 27 |
"p_none_distract": 0.12,
|
| 28 |
"p_distract": 0.15,
|
| 29 |
"p_none_pair": 0.25,
|
| 30 |
+
"synthetic_repeat": 1,
|
| 31 |
+
"public_frac": 1.0,
|
| 32 |
+
"anchor": "",
|
| 33 |
+
"anchor_w": 0.0,
|
| 34 |
+
"anchor_sources": "",
|
| 35 |
+
"out": "/runs/v7-06b/02-trial-2/checkpoint",
|
| 36 |
+
"seed": 2
|
| 37 |
},
|
| 38 |
+
"suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2",
|
| 39 |
"base_revision": "da87bfb608c14b7cf20ba1ce41287e8de496c0cd",
|
| 40 |
"ordinal_objective": "ranked_probability_score",
|
| 41 |
"holdout": []
|
training_metrics.json
CHANGED
|
@@ -1,14 +1,14 @@
|
|
| 1 |
{
|
| 2 |
-
"wall_seconds":
|
| 3 |
-
"records_seen":
|
| 4 |
-
"requested_records":
|
| 5 |
"truncated_records": 0,
|
| 6 |
"rejected_records": 0,
|
| 7 |
-
"optimizer_steps":
|
| 8 |
-
"forward_tokens":
|
| 9 |
-
"peak_device_bytes":
|
| 10 |
"device": "cuda",
|
| 11 |
"dtype": "bf16",
|
| 12 |
"batch": 8,
|
| 13 |
-
"peak_rss_bytes":
|
| 14 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"wall_seconds": 638.1045656204224,
|
| 3 |
+
"records_seen": 30370,
|
| 4 |
+
"requested_records": 25152,
|
| 5 |
"truncated_records": 0,
|
| 6 |
"rejected_records": 0,
|
| 7 |
+
"optimizer_steps": 3144,
|
| 8 |
+
"forward_tokens": 5829284,
|
| 9 |
+
"peak_device_bytes": 37059323392,
|
| 10 |
"device": "cuda",
|
| 11 |
"dtype": "bf16",
|
| 12 |
"batch": 8,
|
| 13 |
+
"peak_rss_bytes": 8731115520
|
| 14 |
}
|