Upload README.md with huggingface_hub

README.md

@@ -10,159 +10,6 @@ tags:
 - code
 - swe
 - rl
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-dataset_info:
-  features:
-  - name: base_commit
-    dtype: string
-  - name: created_at
-    dtype: string
-  - name: image_name
-    dtype: string
-  - name: instance_id
-    dtype: string
-  - name: interface
-    dtype: string
-  - name: language
-    dtype: string
-  - name: license
-    dtype: string
-  - name: patch
-    dtype: string
-  - name: pr_description
-    dtype: string
-  - name: problem_statement
-    dtype: string
-  - name: repo
-    dtype: string
-  - name: test_patch
-    dtype: string
-  - name: FAIL_TO_PASS
-    list: string
-  - name: PASS_TO_PASS
-    list: string
-  - name: install_config
-    struct:
-    - name: base_image_name
-      dtype: string
-    - name: docker_specs
-      struct:
-      - name: _variant
-        dtype: string
-      - name: bazel_version
-        dtype: string
-      - name: bun_version
-        dtype: string
-      - name: cargo_version
-        dtype: string
-      - name: deno_version
-        dtype: string
-      - name: docker_version
-        dtype: string
-      - name: erlang_version
-        dtype: string
-      - name: gcc_version
-        dtype: string
-      - name: go_version
-        dtype: string
-      - name: helm_version
-        dtype: string
-      - name: java_version
-        dtype: string
-      - name: jdk_version
-        dtype: string
-      - name: llvm_version
-        dtype: string
-      - name: lua_version
-        dtype: string
-      - name: luajit_version
-        dtype: string
-      - name: neovim_version
-        dtype: string
-      - name: node_version
-        dtype: string
-      - name: npm_version
-        dtype: string
-      - name: nvim_version
-        dtype: string
-      - name: pnpm_version
-        dtype: string
-      - name: python_image
-        dtype: string
-      - name: python_version
-        dtype: string
-      - name: redis_version
-        dtype: string
-      - name: ruby_version
-        dtype: string
-      - name: rust_version
-        dtype: string
-      - name: rustc_version
-        dtype: string
-      - name: solana_version
-        dtype: string
-      - name: sqlite_version
-        dtype: string
-    - name: install
-      list: string
-    - name: log_parser
-      dtype: string
-    - name: test_cmd
-      dtype: string
-  - name: meta
-    struct:
-    - name: llm_metadata
-      struct:
-      - name: code
-        dtype: string
-      - name: confidence
-        dtype: float64
-      - name: detected_issues
-        struct:
-        - name: B1
-          dtype: bool
-        - name: B2
-          dtype: bool
-        - name: B3
-          dtype: bool
-        - name: B4
-          dtype: bool
-        - name: B5
-          dtype: bool
-        - name: B6
-          dtype: bool
-      - name: difficulty
-        dtype: string
-      - name: external_urls
-        list: string
-      - name: intent_completeness
-        dtype: string
-      - name: pr_categories
-        list: string
-      - name: reasoning
-        dtype: string
-      - name: test_alignment_issues
-        list: string
-    - name: num_modified_files
-      dtype: int64
-    - name: num_modified_lines
-      dtype: int64
-    - name: pr_author
-      dtype: string
-    - name: pr_labels
-      list: string
-    - name: pr_url
-      dtype: string
-  splits:
-  - name: train
-    num_bytes: 276334043
-    num_examples: 6275
-  download_size: 226440922
-  dataset_size: 276334043
 ---
 
 # SWE-rebench-V2-Filtered-Verified
@@ -171,7 +18,7 @@ dataset_info:
 
 Filtered and gold-patch-verified subset of Nebius's
 [SWE-rebench-V2](https://huggingface.co/datasets/nebius/SWE-rebench-V2)
-([paper](https://arxiv.org/abs/2602.23866)): **6,275 / 32,079** freshly-mined GitHub PR tasks
+([paper](https://arxiv.org/abs/2602.23866)): **6,272 / 32,079** freshly-mined GitHub PR tasks
 across 17 languages. Default dataset of the `swerebench_v2_v1` taskset.
 
 ## Changes vs upstream
@@ -193,9 +40,13 @@ across 17 languages. Default dataset of the `swerebench_v2_v1` taskset.
 * Gold-patch validation pass 1 via SolveEnv — reward 1.0 required.
 * Independent pass 2 — flaky rows removed.
 * No-edit pass — rows solvable with zero edits removed.
+* RL always-fail audit (2026-07) — instances that scored 0.0 in every 16-rollout group across
+  independent GLM-5.2 RL samplings were re-gold-validated twice; 2 deterministically
+  unresolvable + 1 flaky removed.
 
 Per-row outcomes ship with this repo (`swe-rebench-v2-validation.jsonl`,
-`swe-rebench-v2-flaky-instances.json`, `swe-rebench-v2-no-edit-pass-exclusions.jsonl`).
+`swe-rebench-v2-flaky-instances.json`, `swe-rebench-v2-no-edit-pass-exclusions.jsonl`,
+`swe-rebench-v2-rl-alwaysfail-exclusions.json`).
 `image_name` is rewritten to the public Prime image registry (`prime/primeintellect/...`) so
 sandbox rollouts avoid Docker Hub rate limits and get fast pulls. Schema is otherwise unchanged.
 
@@ -205,7 +56,7 @@ License mirrors upstream: CC-BY-4.0.
 
 | Split | Rows |
 |---|---:|
-| `train` | 6,275 |
+| `train` | 6,272 |
 
 ## How to use
 
@@ -295,6 +146,16 @@ Plus a no-edit pass blocklist at
 ``swe-rebench-v2-no-edit-pass-exclusions.jsonl``: instances where the no-edit
 debugger got reward 1.0, meaning the tests pass without applying any agent edit.
 
+Plus an RL always-fail audit blocklist at
+``swe-rebench-v2-rl-alwaysfail-exclusions.json``: instances flagged during
+GLM-5.2 RL training (every 16-rollout group scored 0.0 across independent
+samplings) and then gold-validated twice via the ``swerebench_v2_v1`` taskset's
+``validate`` hook (2026-07-17). Two are deterministically unresolvable under
+the current verifier (graded FAIL_TO_PASS tests the row's ``test_cmd`` never
+runs, or tests needing JDK toolchains absent from the image) and one is flaky
+(2,422 all-or-nothing tests spawning servers/ports). Per-row verdicts and
+reasons live in the file.
+
 After filtering, ``image_name`` is rewritten from
 ``docker.io/swerebenchv2/<name>:<tag>`` to the public Prime image registry at
 ``prime/primeintellect/<name>:<tag>`` (the ``swerebenchv2/`` namespace is
@@ -366,6 +227,12 @@ _NO_EDIT_PASSING = frozenset(
     json.loads(line)["instance_id"] for line in _NO_EDIT_PASS_PATH.read_text(encoding="utf-8").splitlines() if line
 )
 
+# RL always-fail audit exclusions: instances whose reward was 0.0 in every
+# 16-rollout group across independent GLM-5.2 RL samplings, then gold-validated
+# twice (2026-07-17) — two deterministically unresolvable, one flaky.
+_RL_ALWAYSFAIL_PATH = Path(__file__).parent / "swe-rebench-v2-rl-alwaysfail-exclusions.json"
+_RL_ALWAYSFAIL = frozenset(entry["instance_id"] for entry in json.loads(_RL_ALWAYSFAIL_PATH.read_text()))
+
 # statamic/cms rows add PHP tests but run unrelated JS suites, so base +
 # test_patch can score reward 1.0 without the issue fix.
 _DROP_REPOS = frozenset({"statamic/cms"})
@@ -427,6 +294,8 @@ def _passes_filter(example: dict, min_confidence: float) -> bool:
         return False
     if example.get("instance_id") in _NO_EDIT_PASSING:
         return False
+    if example.get("instance_id") in _RL_ALWAYSFAIL:
+        return False
     if (example.get("repo") or "").lower() in _DROP_REPOS:
         return False
     return True
@@ -485,12 +354,13 @@ def main(repo_name: str, push_to_hub: bool, private: bool, min_confidence: float
         create_repo(repo_name, private=private, repo_type="dataset", exist_ok=True)
         push_card_to_hub(repo_name, push_to_hub)
         dataset.push_to_hub(repo_name, private=private)
-        upload_file(
-            path_or_fileobj=str(_NO_EDIT_PASS_PATH),
-            path_in_repo=_NO_EDIT_PASS_PATH.name,
-            repo_id=repo_name,
-            repo_type="dataset",
-        )
+        for sidecar in (_NO_EDIT_PASS_PATH, _RL_ALWAYSFAIL_PATH):
+            upload_file(
+                path_or_fileobj=str(sidecar),
+                path_in_repo=sidecar.name,
+                repo_id=repo_name,
+                repo_type="dataset",
+            )
         print(f"✅ Pushed dataset to https://huggingface.co/datasets/{repo_name}")