Draft in Parallel, Condition Through Depth: Adjacent Causal Injection for Speculative Decoding
Paper • 2609.36173 • Published • 2
Datasets:
id stringlengths 40 40 | source stringclasses 16
values | source_license stringclasses 4
values | domain stringclasses 9
values | prompt_messages listlengths 1 51 | response stringlengths 12 80k | output_token_ids listlengths 2 16.4k | finish_reason stringclasses 2
values | sampling dict | is_validation bool 2
classes | reasoning_effort stringclasses 3
values | enable_thinking bool 2
classes | chat_template_kwargs stringclasses 6
values | tools stringlengths 4 8.66k | provenance dict | prompt_messages_full_json stringlengths 55 162k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
fedcc091ffd8b4db65ebf8d7b21ba4e6f1e7c1a7 | Team-ACE/ToolACE | apache-2.0 | agentic | [
{
"role": "system",
"content": "You are a helpful assistant. Select appropriate tools for the user request, ask for missing required information, and answer directly when no tool applies."
},
{
"role": "user",
"content": "Can you check the current status and responsiveness of the servers hosting... | The user wants to check the status/responsiveness of the server. There's a "ping" tool that does exactly that. I'll call it.
</think>
<tool_call>
<function=ping>
</function>
</tool_call><|im_end|> | [
760,
1156,
6587,
310,
1716,
279,
2552,
31355,
2650,
12415,
314,
279,
3421,
13,
2532,
579,
264,
328,
9694,
1,
5224,
421,
1503,
6681,
421,
13,
353,
3172,
1562,
424,
13,
198,
248069,
271,
248058,
198,
27,
1628,
28,
9694,
29,
198,
510,
... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | xhigh | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "xhigh"} | [{"type": "function", "function": {"name": "ping", "description": "Check the availability and responsiveness of the server that hosts the stocks archive data.", "parameters": {"type": "object", "properties": {}, "required": []}, "required": null}}, {"type": "function", "function": {"name": "Moving_Average_Convergence_D... | {
"source_tag": "Team-ACE/ToolACE:train",
"source_revision": "6bda777c88d21e5a204703c1ee45597a8fa4f734",
"source_file": "data.json",
"source_row": "5896",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/Team-ACE/ToolACE",
"generation_model": "RedHatAI/Qwen3.8-Flash-Next-NVFP4",
"... | [{"role": "system", "content": "You are a helpful assistant. Select appropriate tools for the user request, ask for missing required information, and answer directly when no tool applies."}, {"role": "user", "content": "Can you check the current status and responsiveness of the servers hosting my stocks archive data?"}... |
fde3166a83ac9287a3bd6579e2fdf87da4954702 | OpenAssistant/oasst2 | apache-2.0 | multilingual | [
{
"role": "user",
"content": "Quels accordages de guitare utiliser pour jouer du métal ?"
}
] | The user is asking about guitar tunings used for playing metal music. This is a straightforward music knowledge question. I'll provide a comprehensive answer covering the most common alternative tunings used in metal, from standard down to very low tunings, with examples of bands and genres that use them. I'll organize... | [
760,
1156,
369,
9859,
883,
16488,
11131,
793,
1429,
364,
5442,
9041,
4477,
13,
1061,
369,
264,
29350,
4477,
6337,
3296,
13,
353,
3172,
3300,
264,
15352,
4087,
17661,
279,
1379,
4047,
10242,
11131,
793,
1429,
303,
9041,
11,
494,
5129,
... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | low | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "low"} | null | {
"source_tag": "OpenAssistant/oasst2:train",
"source_revision": "179dd21fc55192153d94adb0e0ce8f69e222bf75",
"source_file": "data/train-00000-of-00001-88ba0162028a73fc.parquet",
"source_row": "54760789-ea91-4c5d-b37f-5faefff9d121",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/Open... | [{"role": "user", "content": "Quels accordages de guitare utiliser pour jouer du métal ?"}] |
c8c4bbebea565832bb5e8fba058a9df713ca33f6 | bigcode/self-oss-instruct-sc2-exec-filter-50k | odc-by | code | [
{
"role": "user",
"content": "Construct a Python function called `get_random_three_persons`, which takes a JSON file path `fpath` and returns a dictionary containing three random persons from the JSON dictionary, with keys \"name\", \"age\", and \"country\". The JSON file contains a list of dictionaries, wh... | ```python
import json
import random
def get_random_three_persons(fpath):
"""
Reads a JSON file containing a list of dictionaries, each with keys "name", "age", and "country".
Returns a dictionary containing three random persons from the JSON list, with keys "name", "age", and "country".
The retur... | [
71093,
12305,
198,
464,
2857,
198,
464,
4055,
1358,
727,
615,
21941,
48332,
22689,
82,
925,
2267,
1590,
198,
262,
4071,
198,
262,
42272,
264,
4566,
999,
8222,
264,
1103,
314,
55588,
11,
1754,
440,
6679,
328,
591,
487,
328,
416,
487,
... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | low | false | {"enable_thinking": false, "preserve_thinking": true, "reasoning_effort": "low"} | null | {
"source_tag": "bigcode/self-oss-instruct-sc2-exec-filter-50k:train",
"source_revision": "356bb069eee815daa6e23e9a282eeefe1490ad44",
"source_file": "data/train-00000-of-00001.parquet",
"source_row": "689463",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/bigcode/self-oss-instruct-... | [{"role": "user", "content": "Construct a Python function called `get_random_three_persons`, which takes a JSON file path `fpath` and returns a dictionary containing three random persons from the JSON dictionary, with keys \"name\", \"age\", and \"country\". The JSON file contains a list of dictionaries, where each dic... |
c63591fa7490ec07a166e96cb94f841cc9249343 | HuggingFaceH4/ultrachat_200k | mit | chat | [
{
"role": "user",
"content": "Can you explain the difference between the biathlon and cross-country skiing events in the Olympic Games?"
},
{
"role": "assistant",
"content": "Biathlon and cross-country skiing are two different events in the Winter Olympics.\n\nThe biathlon is a combination of cr... | The user is asking about famous biathletes, continuing the conversation about biathlon. This is a straightforward factual question. I should mention several well-known biathletes from different countries and eras to give a good overview. Let me think of notable ones: Ole Einar Bjørndalen (Norway), Martin Fourcade (Fran... | [
760,
1156,
369,
9859,
883,
10919,
5843,
572,
13482,
11,
13942,
279,
10125,
883,
5843,
74288,
13,
1061,
369,
264,
29350,
57879,
3296,
13,
353,
1220,
6088,
3679,
1575,
20654,
5843,
572,
13482,
494,
2086,
5656,
321,
2627,
299,
310,
2873,
... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | xhigh | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "xhigh"} | null | {
"source_tag": "HuggingFaceH4/ultrachat_200k:train",
"source_revision": "8049631c405ae6576f93f445c6b8166f76f5505a",
"source_file": "data/train_sft-00002-of-00003-ee46ed25cfae92c6.parquet",
"source_row": "11607",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/HuggingFaceH4/ultrachat... | [{"role": "user", "content": "Can you explain the difference between the biathlon and cross-country skiing events in the Olympic Games?"}, {"role": "assistant", "content": "Biathlon and cross-country skiing are two different events in the Winter Olympics.\n\nThe biathlon is a combination of cross-country skiing and rif... |
0e628064927fdc8cc0fa896d1f8182cc4b1727fb | CohereForAI/aya_dataset | apache-2.0 | multilingual | [
{
"role": "user",
"content": "Waa immisa tirada lafaha ee ku jira jirka bini’aadamka?"
}
] | We need answer in Somali likely. User asks: "Waa immisa tirada lafaha ee ku jira jirka bini’aadamka?" Means "How many bones are in the human body?" Need answer. Need final in Somali. Could mention adults 206, newborns ~270/300 merge. Need maybe concise. Ensure language. Final: Jirka bini’aadamka ee qofka waa weyn waxaa... | [
1596,
1144,
4087,
303,
81771,
4222,
13,
2570,
16561,
25,
328,
54,
5137,
4735,
10445,
43133,
2499,
1147,
69,
13154,
35142,
35390,
492,
8565,
492,
396,
4405,
292,
6383,
38341,
49238,
4405,
7285,
57691,
328,
4199,
1599,
24066,
513,
303,
27... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | xhigh | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "xhigh"} | null | {
"source_tag": "CohereForAI/aya_dataset:train",
"source_revision": "f9ea04583f02a8f86404ff6c58bf75fe637df8a2",
"source_file": "data/train-00000-of-00001.parquet",
"source_row": "6636",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/CohereForAI/aya_dataset",
"generation_model": "R... | [{"role": "user", "content": "Waa immisa tirada lafaha ee ku jira jirka bini’aadamka?"}] |
b313293f481d3a2a3f5ab6e345ec224a2adb5800 | ise-uiuc/Magicoder-OSS-Instruct-75K | mit | code | [
{
"role": "user",
"content": "You are tasked with creating a Django model for a task management system. The model should include a field for the task name and a field for the author's name. The author's name should default to \"Anonymous\" if not provided. You need to write the Django model code to achieve ... | We need answer user's request. Need produce final answer with Django model code. Simple. Need maybe include class Task, models import. The request asks create Django model named Task with fields. We can provide code snippet.
Need ensure no extra too much maybe concise. Final should include import from django.db import ... | [
1596,
1144,
4087,
1156,
579,
1622,
13,
14235,
7936,
1534,
4087,
440,
50832,
1558,
1970,
13,
8722,
13,
14235,
6970,
2830,
523,
5261,
11,
3983,
1120,
13,
561,
1622,
16561,
1791,
50832,
1558,
6725,
5261,
440,
4881,
13,
1165,
628,
3300,
1... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | true | xhigh | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "xhigh"} | null | {
"source_tag": "ise-uiuc/Magicoder-OSS-Instruct-75K:train",
"source_revision": "5f839b1f368a76b161028bb9edff055db34022b2",
"source_file": "data-oss_instruct-decontaminated.jsonl",
"source_row": "7683",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/ise-uiuc/Magicoder-OSS-Instruct-7... | [{"role": "user", "content": "You are tasked with creating a Django model for a task management system. The model should include a field for the task name and a field for the author's name. The author's name should default to \"Anonymous\" if not provided. You need to write the Django model code to achieve this.\n\nCre... |
7fb7fcb3424c8e6ba94d961b33f3e20baa7eca03 | AI-MO/NuminaMath-CoT | apache-2.0 | math | [
{
"role": "user",
"content": "Is there a surjective function \\( f : \\mathbf{R} \\rightarrow \\mathbf{R} \\) such that for any real numbers \\( x \\) and \\( y \\), the value of \\( f(x+y) - f(x) - f(y) \\) is only one of two numbers: 0 and 1?"
}
] | We need answer math problem. Need solve. Problem: Does there exist surjective f:R->R such that for all real x,y, f(x+y)-f(x)-f(y) is only one of two numbers: 0 and 1. "is only one of two numbers: 0 and 1" means value belongs to {0,1}. Need existence of surjective function satisfying quasimorphism defect in {0,1}. We ne... | [
1596,
1144,
4087,
6673,
3377,
13,
14235,
11290,
13,
21397,
25,
12193,
1017,
2905,
1668,
48791,
281,
25,
49,
397,
49,
1680,
421,
364,
660,
1865,
830,
7127,
11,
281,
2007,
41590,
7060,
69,
2007,
7060,
69,
6803,
8,
369,
1132,
799,
314,... | length | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | xhigh | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "xhigh"} | null | {
"source_tag": "AI-MO/NuminaMath-CoT:train",
"source_revision": "9d8d210c9f6a36c8f3cd84045668c9b7800ef517",
"source_file": "data/train-00000-of-00005.parquet",
"source_row": "6655",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/AI-MO/NuminaMath-CoT",
"generation_model": "RedHatA... | [{"role": "user", "content": "Is there a surjective function \\( f : \\mathbf{R} \\rightarrow \\mathbf{R} \\) such that for any real numbers \\( x \\) and \\( y \\), the value of \\( f(x+y) - f(x) - f(y) \\) is only one of two numbers: 0 and 1?"}] |
a96673bdaf1f1e8c903180159d810ff68c74647a | bigcode/self-oss-instruct-sc2-exec-filter-50k | odc-by | code | [
{
"role": "user",
"content": "Implement a Python function `print_printables` that takes a string `s`, and return a string of all printable characters from `s`, without using the built-in `string.printable` constant. This function should not ignore characters that appear multiple times in `s`.\n\nYour code s... | Let me think about this problem:
1. I need to implement a function `print_printables` that filters a string to keep only printable characters.
2. I cannot use `string.printable`.
3. I should not deduplicate - if a character appears multiple times, it should appear multiple times in the result.
What counts as "printab... | [
9764,
728,
1683,
883,
411,
3377,
25,
271,
16,
13,
353,
1144,
310,
4072,
264,
709,
1510,
1302,
9767,
4635,
63,
421,
13019,
264,
886,
310,
2426,
1132,
40608,
5587,
13,
198,
17,
13,
353,
4021,
958,
1510,
889,
2145,
470,
27653,
198,
1... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | low | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "low"} | null | {
"source_tag": "bigcode/self-oss-instruct-sc2-exec-filter-50k:train",
"source_revision": "356bb069eee815daa6e23e9a282eeefe1490ad44",
"source_file": "data/train-00000-of-00001.parquet",
"source_row": "195906",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/bigcode/self-oss-instruct-... | [{"role": "user", "content": "Implement a Python function `print_printables` that takes a string `s`, and return a string of all printable characters from `s`, without using the built-in `string.printable` constant. This function should not ignore characters that appear multiple times in `s`.\n\nYour code should pass t... |
2c29a98d3a9cf2bcbee37a6b307d6396fc27a1d3 | gretelai/synthetic_text_to_sql | apache-2.0 | structured | [
{
"role": "user",
"content": "Write a SQL query for the following request.\n\nWhat is the distribution of fan demographics by favorite sports?\n\nDatabase schema and sample data:\nCREATE TABLE fan_demographics_sports (id INT, fan VARCHAR(255), age INT, gender VARCHAR(10), sport VARCHAR(255)); INSERT INTO fa... | The user wants a SQL query that shows the distribution of fan demographics by favorite sports. This means they want to see how fans are distributed across different sports, likely grouped by sport and possibly by gender to show the demographic breakdown.
Looking at the schema: `fan_demographics_sports` has columns: id... | [
760,
1156,
6587,
264,
7622,
3134,
421,
4774,
279,
7731,
314,
8145,
60138,
539,
6714,
9541,
13,
1061,
3254,
781,
1318,
310,
1436,
1204,
6985,
513,
4098,
3808,
2086,
9541,
11,
4222,
38710,
539,
10239,
321,
10451,
539,
9533,
310,
1420,
2... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | low | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "low"} | null | {
"source_tag": "gretelai/synthetic_text_to_sql:train",
"source_revision": "740ab236e64503fba51be1101df7a1be83bf455d",
"source_file": "synthetic_text_to_sql_train.snappy.parquet",
"source_row": "51644",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/gretelai/synthetic_text_to_sql",
... | [{"role": "user", "content": "Write a SQL query for the following request.\n\nWhat is the distribution of fan demographics by favorite sports?\n\nDatabase schema and sample data:\nCREATE TABLE fan_demographics_sports (id INT, fan VARCHAR(255), age INT, gender VARCHAR(10), sport VARCHAR(255)); INSERT INTO fan_demographi... |
3ccbfe030a2612fd14c936c64f648c314644f162 | HuggingFaceH4/ultrachat_200k | mit | chat | [
{
"role": "user",
"content": "Given the text: This dish is best made with small cucumbers, that have few seeds, if you can find them.\nCut some of the skin of the cucumbers away lengthways to create long thick or thin stripes. This can be done with a specially designed cutting tool or potato peeler. Cut the... | We need answer user asks: Given text, can you substitute small cucumbers with larger ones for this dish? Need answer based on text. We have to produce final answer. Need think: Text says best made with small cucumbers that have few seeds, if you can find them. It instructs cutting, removing seeds. Could larger cucumber... | [
1596,
1144,
4087,
1156,
16561,
25,
15767,
1414,
11,
628,
488,
26782,
2526,
71763,
367,
1837,
440,
7875,
5981,
364,
411,
11659,
30,
14235,
4087,
3018,
383,
1414,
13,
1165,
599,
310,
7936,
1534,
4087,
13,
14235,
1683,
25,
2824,
2640,
17... | stop | {
"temperature": 1,
"top_k": 20,
"top_p": 0.95
} | false | xhigh | true | {"enable_thinking": true, "preserve_thinking": true, "reasoning_effort": "xhigh"} | null | {
"source_tag": "HuggingFaceH4/ultrachat_200k:train",
"source_revision": "8049631c405ae6576f93f445c6b8166f76f5505a",
"source_file": "data/train_sft-00002-of-00003-ee46ed25cfae92c6.parquet",
"source_row": "11523",
"upstream_source": null,
"source_url": "https://huggingface.co/datasets/HuggingFaceH4/ultrachat... | [{"role": "user", "content": "Given the text: This dish is best made with small cucumbers, that have few seeds, if you can find them.\nCut some of the skin of the cucumbers away lengthways to create long thick or thin stripes. This can be done with a specially designed cutting tool or potato peeler. Cut the cucumbers i... |
End of preview. Expand in Data Studio
Qwen3.8-Flash-Next DSpine research generations
An attributed, conservative subset of on-policy target generations prepared for speculative-draft research. 73,157 records, 176,812,512 response tokens, 18 parquet shards. The full retained generation inventory was 100,318 records / 228,480,448 response tokens. Five sources were excluded in full after a per-source license audit and owner review.
There is no blanket Apache-2.0 license for these prompts. Each row has a
source_license and provenance. The prompt sources retain their individual
licenses and attribution obligations; see LICENSES.md and
sources.json. New model outputs do not erase prompt rights or
relicense copied underlying content. This collection imposes no additional
blanket license over third-party material. The separate draft-head/code license
is not a grant covering this dataset.
Included sources and attribution
Original final answers were removed; queries and earlier conversation/tool context were retained. Responses are new target-model continuations. Source publishers and contributors are credited below; pinned revision/file/row details are preserved in each row. Counts describe this exported subset, not upstream dataset sizes.
| Source dataset | Attribution | Prompt license | Records | Response tokens |
|---|---|---|---|---|
| AI-MO/NuminaMath-CoT | AI-MO and contributors | apache-2.0 | 4,134 | 12,277,826 |
| CohereForAI/aya_dataset | CohereForAI and contributors | apache-2.0 | 5,016 | 9,238,437 |
| HuggingFaceH4/ultrachat_200k | HuggingFaceH4 and contributors | mit | 9,957 | 27,393,024 |
| NousResearch/hermes-function-calling-v1 | NousResearch and contributors | apache-2.0 | 1,161 | 620,846 |
| OpenAssistant/oasst2 | OpenAssistant and contributors | apache-2.0 | 2,644 | 6,472,747 |
| THUDM/LongCite-45k | THUDM and contributors | apache-2.0 | 2,040 | 5,723,439 |
| Team-ACE/ToolACE | Team-ACE and contributors | apache-2.0 | 8,467 | 2,550,331 |
| allenai/WildChat-1M | allenai and contributors | odc-by | 9,538 | 27,009,296 |
| bigcode/self-oss-instruct-sc2-exec-filter-50k | bigcode and contributors | odc-by | 4,686 | 12,928,612 |
| glaiveai/glaive-function-calling-v2 | glaiveai and contributors | apache-2.0 | 4,752 | 589,494 |
| gretelai/synthetic_text_to_sql | gretelai and contributors | apache-2.0 | 3,832 | 2,598,768 |
| ise-uiuc/Magicoder-OSS-Instruct-75K | ise-uiuc and contributors | mit | 4,631 | 21,616,813 |
| meta-math/MetaMathQA | meta-math and contributors | mit | 3,764 | 2,077,831 |
| nvidia/Nemotron-Agentic-v1 | nvidia and contributors | cc-by-4.0 | 1 | 191 |
| nvidia/OpenCodeInstruct | nvidia and contributors | cc-by-4.0 | 4,547 | 11,820,246 |
| nvidia/OpenMathReasoning | nvidia and contributors | cc-by-4.0 | 3,987 | 33,894,611 |
WildChat uses ODC-BY, retroactively replacing its earlier ImpACT license. StarCoder2 Self-Alignment also uses ODC-BY. That license covers database rights, not independently copyrighted content. Do not assume it relicenses code copied inside a prompt. The export does not include unused raw seed-code fields. WildChat IP hashes, headers, locations, timestamps and source user IDs were not part of prompt extraction. Redacted/toxic conversations and known credential patterns were filtered in the original prompt build.
OpenMathReasoning contributes only its selected additional_problems split.
The selected Terminal-Corpus and neulab components have no generated records in
this retained inventory, so neither is included in this dataset.
Sources excluded after review
| Source | Removed records | Removed response tokens | Reason |
|---|---|---|---|
| SWE-bench/SWE-smith-trajectories | 8,124 | 2,406,806 | MIT dataset card does not settle licenses/attribution of embedded repository code in tool context. |
| nebius/SWE-agent-trajectories | 4,168 | 1,664,516 | Card requires repository-specific licenses and notes Llama-output obligations. |
| m-a-p/CodeFeedback-Filtered-Instruction | 4,632 | 29,199,546 | Named upstream branches include NC/ND and NC/SA terms despite the aggregator's Apache card. |
| mlabonne/open-perfectblend | 7,226 | 14,955,586 | Mixed upstream chain, including CodeAlpaca-derived material with NC data terms; the whole source was omitted. |
| nvidia/Llama-Nemotron-Post-Training-Dataset | 2,950 | 3,220,680 | Per-sample CC-BY/ODC-BY/CC-BY-SA and possible Llama terms were not retained as row licenses in extracted prompts. |
These exclusions remove 27,100 records and 51,447,134 response tokens. Later row-level upstream/repository-license joins could recover eligible portions; none was performed for this conservative version. LMSYS-Chat-1M, unknown-license, noncommercial and no-derivatives candidates are not admitted. The differently licensed Arena preference subset is not LMSYS-Chat-1M, but it is still omitted here as part of the whole Open-PerfectBlend exclusion. The complete review is in dataset-license-audit.md.
Generation model, tokenizer and settings
Target: RedHatAI/Qwen3.8-Flash-Next-NVFP4,
revision c8f2fb1b9869f686b214782036123b10ff96d14a. The tokenizer and chat
template are from the same snapshot; tokenizer file hashes are in manifest.json.
Temperature 1.0, top-k 20, top-p 0.95. Exact prompt token IDs were
sent to the generation server. Outputs were decoded using the pinned tokenizer,
skip_special_tokens=False, clean_up_tokenization_spaces=False. Output IDs
are authoritative; reasoning/tool/end markers remain in the response text.
Responses are outputs of Qwen3.8-Flash-Next (through the RedHatAI NVFP4
quantization), which is distributed under the Qwen Community License 1.0; a copy
is in licenses/Qwen-Community-License-1.0.txt. Its conditions, including those
for Model-as-a-Service and AI Work Assistant businesses, may apply to uses of
these outputs.
The actual thinking flags are 67,143 true and 6,014 false, with no unknown
flags. Reasoning effort: 36,271 xhigh, 22,026 medium, 14,860 low. Original
per-prompt chat_template_kwargs and tools are retained. Do not describe every
record as thinking enabled. Finish reasons: 67,733 stop and 5,424 length-capped.
Serving numerical differences across generation platforms remain possible;
the nominal T/k/p settings alone do not imply bit-identical output laws.
Schema and stable validation split
| Column | Meaning |
|---|---|
id |
Stable canonical prompt ID. |
source, source_license, domain |
Attributed source, prompt license and domain. |
prompt_messages |
Role/content text messages; content is nullable where tool calls carry the message. |
prompt_messages_full_json, tools |
Original full message dictionaries/tool schemas as JSON, preserving tool calls and IDs. |
response, output_token_ids |
Decoded continuation and original generated token IDs. |
finish_reason, sampling |
Original completion status and T/k/p. |
enable_thinking, reasoning_effort, chat_template_kwargs |
Actual prompt-specific thinking/template metadata. |
provenance |
Source tag/revision/file/row/upstream source/URL, generation model/revision/platform family/part. |
is_validation |
The trainer's existing deterministic 1% hash split. |
There are 72,452 training rows and 705 validation rows. The rule is the first
16 hex digits of SHA-256(id), divided by 2**64, less than 0.01. This is distinct
from the separate 2,000 prompt-family holdouts used to produce evaluation caches.
The parquet container is loaded as one HF train split; filter the flag before
training, rather than training on every loaded row:
from datasets import load_dataset
data = load_dataset("ProCreations/Qwen3.8-Flash-Next-DSpine-generations", split="train")
train = data.filter(lambda row: not row["is_validation"])
validation = data.filter(lambda row: row["is_validation"])
Filtering, intended use and limitations
Use for research on target imitation, speculative draft heads and sampling. Do not treat these synthetic continuations as verified answers. They can contain mistakes, harmful content and copied prompt text. This is a generated subset of a deliberately mixed prompt pool, not a representative sample of all use cases.
The export honors the two-ID EXCLUDE_IDS.json and exact/near prompt dedup.
Prompt decontamination used normalized 13-grams against GSM8K test, MATH-500,
HumanEval, MBPP test, LiveCodeBench, MT-Bench, Arena-Hard and local serving prompts;
a supplemental tool-argument audit produced the two excluded IDs. Neither ID
appears in this export. These procedures reduce known overlap and are not proof
of zero contamination. The exclusion file/hash and audit are included.
The full-row validation checks unique IDs, allowed sources/licenses, source provenance, target revision, sampling, token totals, deterministic split and shard hashes; sample decoding is checked against the local tokenizer. 61 rows were removed by a credential-pattern scan.
No target features, probability caches, optimizer states or target weights are
included. Files and checksums are listed in manifest.json; validation and
credential-scan scope are recorded in validation-receipt.json.
Credit the source dataset authors above, Qwen/Red Hat for the target, and the DSpine and DFlash authors. This independent collection is not endorsed by those projects.
- Downloads last month
- 80