Dataset / metadata /dataset_inventory.json
dsfasdfaserwer's picture
Add Dataset #2: Magicoder-OSS-Instruct-75K python subset (+673 records)
159bae2 verified
Raw History Blame Contribute Delete
18.5 kB
{
"generated": "2026-10-03 (updated after Dataset #1 and #2 processing)",
"method": "HF API + dataset cards + papers/blogs/pipeline repos + web search. Evidence classes: DOCUMENTED (by creators), VERIFIED_BY_US, INFERRED, UNKNOWN.",
"datasets": [
{
"name": "bigcode/self-oss-instruct-sc2-exec-filter-50k",
"url": "https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-50k",
"platform": "Hugging Face (BigCode)",
"license": "ODC-BY-1.0 (permissive, attribution)",
"rows": 50661,
"size_bytes": 285743988,
"python_relevance": "HIGH (Python only)",
"programming_relevance": "HIGH",
"reasoning_relevance": "MEDIUM-HIGH (responses include [Reasoning] + [Explanation] sections per generation template)",
"executable_code": true,
"tests_provided": "PARTIAL - generation template requires a [Test] section and responses were execution-filtered against tests; released response field retains tests/asserts for some rows but not all (VERIFIED_BY_US on samples)",
"execution_verified": "DOCUMENTED: pipeline executes each response in a Docker sandbox and keeps only passing ones (github.com/bigcode-project/starcoder2-self-align, execution_filter.py; NeurIPS'24 paper arXiv:2410.24198). We additionally re-execute during curation.",
"human_verified": "NO for this 50k file (a human-reviewed 15.5k subset existed as starcoder2-instruct, now not publicly accessible / 404)",
"expert_verification": "Peer-reviewed pipeline (NeurIPS 2024)",
"llm_generated": true,
"llm_generator": "StarCoder2-15B (self-generated, no proprietary distillation)",
"generated_answers_verified": "YES - execution filter documented",
"grounded_in_real_code": "YES - seeds are real Python functions from The Stack via MultiPL-T pipeline (arXiv:2308.09895)",
"realistic_instructions": "MOSTLY - instructions generated from seed-code concepts, function-level tasks, natural phrasing (sampled rows)",
"explanations_present": true,
"cot_present": "concise explicit reasoning (not private CoT)",
"synthetic": "synthetic instructions+responses grounded in real OSS seeds",
"contamination_possible": "LOW-MEDIUM: decontamination scripts (MBPP, DS-1000) documented in pipeline but application to the released file is UNKNOWN; we run our own screening",
"training_intended": "YES - built to SFT StarCoder2-Instruct; not a benchmark",
"limitations": "No released per-row test field for all rows; self-generated (single model); decontamination application unknown",
"tier": "1 (Exceptional among available candidates)",
"tier_reason": "Documented execution verification of every kept row; permissive license; real-code grounding; training intent; Python-only",
"decision": "PROCESS (rank #1)"
},
{
"name": "ise-uiuc/Magicoder-OSS-Instruct-75K",
"url": "https://huggingface.co/datasets/ise-uiuc/Magicoder-OSS-Instruct-75K",
"platform": "Hugging Face (UIUC)",
"license": "MIT",
"rows": 75197,
"size_bytes": 605849053,
"python_relevance": "MIXED - multilingual (cpp/python/...); filter lang==python",
"programming_relevance": "HIGH",
"reasoning_relevance": "MEDIUM",
"executable_code": true,
"tests_provided": "NO test field in released rows; acceptance required our own derived/inline/released tests",
"execution_verified": "DOCUMENTED (Magicoder paper arXiv:2312.02120): solutions self-generated with tests, execution-filtered; released file is named 'decontaminated'",
"human_verified": false,
"expert_verification": "Peer-reviewed (ICML 2024)",
"llm_generated": true,
"llm_generator": "gpt-3.5-turbo-1106",
"generated_answers_verified": "DOCUMENTED yes (paper); per-row test artifacts not released",
"grounded_in_real_code": "YES - OSS code seeds",
"realistic_instructions": "MOSTLY",
"explanations_present": false,
"cot_present": false,
"synthetic": true,
"contamination_possible": "LOW - filename says decontaminated; paper reports negligible benchmark overlap",
"training_intended": "YES - training Magicoder models",
"limitations": "No released tests -> our acceptance requires us to construct tests; multilingual; GPT-3.5-era quality",
"tier": "2 (Very strong)",
"tier_reason": "Strong documented pipeline, MIT license, decontaminated, but released rows lack tests/explanations",
"decision": "PROCESSED (rank #2) - python subset; 38,284 rows inspected; 673 accepted (verifiable subset only)"
},
{
"name": "ai2-adapt-dev/personahub_code_v2_34999",
"url": "https://huggingface.co/datasets/ai2-adapt-dev/personahub_code_v2_34999",
"platform": "Hugging Face (AI2 mirror of Tulu-3 data)",
"license": "UNKNOWN on card (parent Tulu-3 data is ODC-BY; mirror has no license tag) (VERIFIED_BY_US)",
"rows": 34999,
"size_bytes": 48449834,
"python_relevance": "HIGH (mostly Python; 'Write a python function ...')",
"programming_relevance": "HIGH",
"reasoning_relevance": "MEDIUM",
"executable_code": true,
"tests_provided": "NOT consistently in released messages (sample row: code only)",
"execution_verified": "UNKNOWN/weakly documented for this exact subset",
"human_verified": false,
"expert_verification": "Tulu-3 paper documents dataset curation (arXiv:2411.15124)",
"llm_generated": true,
"llm_generator": "Llama-3.1-405B-Instruct (Tulu 3 pipeline)",
"generated_answers_verified": "UNKNOWN",
"grounded_in_real_code": "NO - persona-driven synthetic prompts",
"realistic_instructions": "MIXED - persona flavoring can be artificial (sample: 'journalist who reviews mime performances')",
"explanations_present": "variable",
"cot_present": "no",
"synthetic": true,
"contamination_possible": "MEDIUM-UNKNOWN",
"training_intended": "YES - Tulu-3 SFT mix",
"limitations": "Sampled first row is internally inconsistent (example claims 7 unique words while listing 8) -> per-row verification mandatory; license unclear on mirror",
"tier": "3 (Usable after substantial filtering)",
"tier_reason": "Training intent and scale, but no released tests, unknown license tag on mirror, quality issues found in sampling",
"decision": "MAYBE PROCESS LATER (rank #4) - only rows we can fully verify"
},
{
"name": "m-a-p/Code-Feedback",
"url": "https://huggingface.co/datasets/m-a-p/Code-Feedback",
"platform": "Hugging Face (M-A-P)",
"license": "Apache-2.0",
"rows": 68000,
"size_bytes": 5200531638,
"python_relevance": "PARTIAL - multilingual (first sampled row is Ruby)",
"programming_relevance": "HIGH",
"reasoning_relevance": "MEDIUM (feedback/critique text)",
"executable_code": true,
"tests_provided": false,
"execution_verified": "NO evidence",
"human_verified": false,
"expert_verification": "paper arXiv:2402.14658 describes construction (from Magicoder-Evol-Instruct + TACO-Python subset + feedback generation)",
"llm_generated": true,
"llm_generator": "Mixtral-8x22B-Instruct feedback over existing data",
"generated_answers_verified": false,
"grounded_in_real_code": "partial (TACO subset)",
"realistic_instructions": "MIXED",
"explanations_present": true,
"cot_present": false,
"synthetic": true,
"contamination_possible": "MEDIUM (TACO/LeetCode-derived subset)",
"training_intended": true,
"limitations": "No tests, no execution evidence, multilingual, large files, derivative of other datasets",
"tier": "3 (low end)",
"tier_reason": "Permissive license and training intent but no verification evidence",
"decision": "DEFER (rank #5, likely skip unless higher-ranked sources exhaust)"
},
{
"name": "BAAI/TACO",
"url": "https://huggingface.co/datasets/BAAI/TACO",
"platform": "Hugging Face (BAAI)",
"license": "Apache-2.0",
"rows": "~1.5M (train ~911k)",
"size_bytes": 10046006960,
"python_relevance": "HIGH (multi-language, Python large)",
"programming_relevance": "HIGH (competitive programming)",
"reasoning_relevance": "HIGH (algorithmic)",
"executable_code": true,
"tests_provided": "YES (judge-style input/output tests)",
"execution_verified": "solutions come from online judges with accept status (DOCUMENTED)",
"human_verified": "judge-verified",
"expert_verification": "paper arXiv:2312.03006",
"llm_generated": false,
"generated_answers_verified": "n/a",
"grounded_in_real_code": "YES (real contest submissions)",
"realistic_instructions": "NO - competitive stdin/stdout problem statements, not typical human dev requests",
"explanations_present": false,
"cot_present": false,
"synthetic": false,
"contamination_possible": "HIGH for benchmark overlap - includes LeetCode problems (LiveCodeBench uses LeetCode/AtCoder/Codeforces); also used widely in eval studies",
"training_intended": "mixed - has train and test splits; test split is used as benchmark",
"limitations": "Contamination risk with LiveCodeBench-style evals; unrealistic instruction style; huge",
"tier": "REJECT for this project",
"tier_reason": "Benchmark-overlap risk (LeetCode-derived rows vs LiveCodeBench) + instruction style mismatch; 'when in doubt, exclude'",
"decision": "REJECT (documented)"
},
{
"name": "KodCode/KodCode-V1 (+ KodCode-V1-SFT-R1 / SFT-4o)",
"url": "https://huggingface.co/datasets/KodCode/KodCode-V1",
"platform": "Hugging Face (Kuaishou/KodCode)",
"license": "CC-BY-NC-4.0 (NON-COMMERCIAL)",
"rows": "~190k problems; SFT variants add solutions",
"size_bytes": 7544000000,
"python_relevance": "HIGH",
"programming_relevance": "HIGH",
"reasoning_relevance": "HIGH",
"executable_code": true,
"tests_provided": "YES (paired pytest-style tests in released rows)",
"execution_verified": "YES (pass-sequence metadata per row)",
"human_verified": false,
"expert_verification": "paper arXiv:2503.02951",
"llm_generated": true,
"generated_answers_verified": true,
"grounded_in_real_code": "NO (knowledge-map synthesized)",
"realistic_instructions": "MIXED - many are LeetCode-style",
"explanations_present": "variable (R1 traces are long CoT)",
"cot_present": "YES (R1 <think> traces - conflicts with our concise-rationale policy)",
"synthetic": true,
"contamination_possible": "MEDIUM - rows carry benchmark_similarity scores; subsets literally named 'Leetcode'",
"training_intended": "YES (SFT/RL)",
"limitations": "CC-BY-NC license incompatible with our Apache-2.0 repo redistribution; LeetCode-style subsets create benchmark-overlap risk",
"tier": "REJECT for this project",
"tier_reason": "Licensing (non-commercial) + contamination risk; would otherwise be Tier 1-2",
"decision": "REJECT (documented)"
},
{
"name": "Benchmarks / evaluation suites (HumanEval, MBPP(+), EvalPlus, DS-1000, APPS, CodeContests, LiveCodeBench, SWE-bench*, BigCodeBench, ClassEval, MultiPL-E, PLB, RepoEval, CRUXEval, SWE-Bench ProMax, Terminal-Bench, StructEditBench, APPS+)",
"url": "various",
"platform": "various",
"license": "various",
"rows": "n/a",
"size_bytes": 0,
"python_relevance": "HIGH but irrelevant",
"programming_relevance": "HIGH",
"reasoning_relevance": "HIGH",
"executable_code": true,
"tests_provided": true,
"execution_verified": true,
"human_verified": true,
"expert_verification": true,
"llm_generated": false,
"generated_answers_verified": "n/a",
"grounded_in_real_code": "variable",
"realistic_instructions": "variable",
"explanations_present": false,
"cot_present": false,
"synthetic": false,
"contamination_possible": "IS the contamination",
"training_intended": "NO - evaluation/benchmark purpose",
"limitations": "Training on these causes false positives during benchmarking",
"tier": "REJECT",
"tier_reason": "Primary purpose is benchmarking/evaluation (user's hard exclusion rule)",
"decision": "REJECT (excluded by policy)"
},
{
"name": "RajMaheshwari/Exercism-Python (+ DCAgent/a1_exercism_python)",
"url": "https://huggingface.co/datasets/RajMaheshwari/Exercism-Python",
"platform": "Hugging Face (scrape of Exercism)",
"license": "MIT claimed on card; underlying Exercism track content varies",
"rows": "small (~hundreds)",
"size_bytes": 1426346,
"python_relevance": "HIGH",
"programming_relevance": "HIGH",
"reasoning_relevance": "HIGH",
"executable_code": true,
"tests_provided": "YES (human-written pytest suites)",
"execution_verified": "human-maintained exercises",
"human_verified": "YES (Exercism community)",
"expert_verification": "community review",
"llm_generated": false,
"generated_answers_verified": "n/a",
"grounded_in_real_code": "YES",
"realistic_instructions": "exercise-style",
"explanations_present": "sometimes",
"cot_present": false,
"synthetic": false,
"contamination_possible": "HIGH - Exercism Python tasks are used as Terminal-Bench 2 benchmark tasks (observed benchmark repos built from 'exercism_python' on HF)",
"training_intended": "unclear scrape",
"limitations": "Benchmark overlap (Terminal-Bench); scrape provenance/license unclear",
"tier": "REJECT for this project",
"tier_reason": "Contamination with an active benchmark + unclear scrape licensing; 'when in doubt, exclude'",
"decision": "REJECT (documented)"
},
{
"name": "Synthetic unverified instruction sets: sahil2801/CodeAlpaca-20k, iamtarun/python_code_instructions_18k_alpaca, glaiveai/glaive-code-assistant-v2, nickrosh/Evol-Instruct-Code-80k-v1, codefuse-ai/Evol-instruction-66k, QuixiAI/dolphin-coder, theblackcat102/evol-codealpaca-v1",
"url": "various",
"platform": "Hugging Face",
"license": "cc-by-4.0 / none / cc-by-nc-sa-4.0 / apache-2.0",
"rows": "20k-110k each",
"size_bytes": 0,
"python_relevance": "HIGH",
"programming_relevance": "HIGH",
"reasoning_relevance": "LOW-MEDIUM",
"executable_code": true,
"tests_provided": false,
"execution_verified": false,
"human_verified": false,
"expert_verification": false,
"llm_generated": true,
"generated_answers_verified": false,
"grounded_in_real_code": false,
"realistic_instructions": "variable, often template-evolved",
"explanations_present": "variable",
"cot_present": false,
"synthetic": true,
"contamination_possible": "dolphin-coder: LeetCode-derived -> HIGH; others MEDIUM-UNKNOWN",
"training_intended": true,
"limitations": "No execution verification; several non-commercial licenses; known quality issues (duplicated/trivial/incoherent rows reported by community)",
"tier": "4 (Low confidence) / REJECT",
"tier_reason": "Fails quality bar; NC licenses fail licensing bar; dolphin-coder fails contamination bar",
"decision": "REJECT (documented)"
},
{
"name": "Pretraining corpora: codeparrot/github-code-clean, code-search-net/code_search_net, nvidia/Nemotron-Post-Training-Dataset-v1 (code chat part), open-thoughts/OpenThoughts-114k (code part)",
"url": "various",
"platform": "Hugging Face",
"license": "apache-2.0 / other / cc-by-4.0 / apache-2.0",
"rows": "millions",
"size_bytes": 0,
"python_relevance": "HIGH",
"programming_relevance": "HIGH",
"reasoning_relevance": "LOW",
"executable_code": true,
"tests_provided": false,
"execution_verified": false,
"human_verified": false,
"expert_verification": "partial (filtering pipelines documented)",
"llm_generated": "variable",
"generated_answers_verified": false,
"grounded_in_real_code": "YES (github-code, CSN)",
"realistic_instructions": "NO - no natural instructions (docstrings only)",
"explanations_present": false,
"cot_present": "OpenThoughts has long CoT (policy conflict)",
"synthetic": "variable",
"contamination_possible": "OpenThoughts code from TACO/LeetCodeGym -> HIGH; CSN test split used in evals -> MEDIUM",
"training_intended": "pretraining/SFT-mix",
"limitations": "Wrong shape for verified instruction dataset; would require full test synthesis; OpenThoughts contamination + CoT policy conflict",
"tier": "REJECT / DEFER for this project",
"tier_reason": "github-code/CSN: no instructions/tests -> not suitable without massive synthesis. Nemotron: mixed provenance, 238GB, code subset provenance heterogeneous -> defer. OpenThoughts: contamination + fake-CoT policy -> reject.",
"decision": "REJECT/DEFER (documented)"
},
{
"name": "Salesforce/xlam-function-calling-60k",
"url": "https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k",
"platform": "Hugging Face",
"license": "CC-BY-4.0",
"rows": 60000,
"size_bytes": 472495664,
"python_relevance": "LOW-MEDIUM (function-call JSON, some Python helpers)",
"programming_relevance": "MEDIUM (API/tool calling)",
"reasoning_relevance": "MEDIUM",
"executable_code": "partial",
"tests_provided": false,
"execution_verified": "documented unification+formatting pipeline, not execution",
"human_verified": false,
"expert_verification": "paper arXiv:2409.03215",
"llm_generated": "partially",
"generated_answers_verified": false,
"grounded_in_real_code": "real API specs (GlaiveAPI etc.)",
"realistic_instructions": "chat queries",
"explanations_present": false,
"cot_present": false,
"synthetic": "partial",
"contamination_possible": "BFCL eval overlap concerns exist (Glaive/Berkeley sources are used in function-calling evals)",
"training_intended": true,
"limitations": "Off-target domain (tool calling, not Python programming/reasoning); eval-overlap risk (BFCL)",
"tier": "out of scope",
"tier_reason": "Not Python programming/reasoning training per objective; contamination concerns",
"decision": "REJECT (out of scope, documented)"
}
],
"ranking_for_processing": [
{
"rank": 1,
"dataset": "bigcode/self-oss-instruct-sc2-exec-filter-50k",
"tier": 1
},
{
"rank": 2,
"dataset": "ise-uiuc/Magicoder-OSS-Instruct-75K (lang==python subset)",
"tier": 2,
"note": "PROCESSED 2026-10-03: 38,284 python rows inspected; 1,338 execution-verifiable (derived/inline/released tests); 746 passed; 71 benchmark-overlap + 2 flaky dropped; 673 accepted"
},
{
"rank": 3,
"dataset": "ai2-adapt-dev/personahub_code_v2_34999",
"tier": 3,
"note": "only if license clarified and rows individually verifiable with our own tests"
},
{
"rank": 4,
"dataset": "m-a-p/Code-Feedback (python subset)",
"tier": 3,
"note": "only rows we can fully verify"
}
]
}