# DISC-Med-SFT

License: restricted (NOASSERTION); Mdexa hosting: link-only. Declared Apache-2.0 is broader than its upstream; the strictest upstream applies (cMedQA2 non-commercial research only; MedDialog copyright belongs to haodf.com); link only.

- Commercial use: No · Derivatives: Unclear · Redistribution: Unclear · File on Mdexa: No
- License status: Restricted (NOASSERTION)
- Mdexa hosting: Link only — License is restricted: the declared license is broader than upstream; the strictest upstream applies (cMedQA2 non-commercial; MedDialog copyright haodf.com); link only.
- Code repository license: Apache-2.0 (license-file, verified)
- Source risk: Derived, Scraped, Real patient text, Model-generated
- Last verified: 2026-11-12
- Card: https://mdexa.com/datasets/disc-med-sft/

## Key facts

- Domain: Clinical medicine
- Content type: QA, dialogue and instruction data
- Size: ≈470,000 items (GitHub); 464,898 rows (HF subset) — GitHub 称近47万条；HF 子集 464,898行
- Language: Chinese
- Modality: Text
- Publisher: 复旦大学 DISC 实验室（FudanDISC）
- Year: 2023（arXiv 2308.14346）

## Commercial use of DISC-Med-SFT

- Commercial use: No.
- License status: Restricted.
Verdicts record what sources state and are not legal advice.

## DISC-Med-SFT license evidence

Declares Apache-2.0, but upstream cMedQA2 is limited to non-commercial research and a MedDialog mirror says copyright belongs to haodf.com; the strictest upstream applies: restricted

- **Hugging Face card metadata** — Apache-2.0: HF dataset card: Apache-2.0 — "license: apache-2.0" — https://huggingface.co/datasets/Flmc/DISC-Med-SFT (2026-11-12; Page opened)
- **Repository license file** — Apache-2.0: GitHub repository Apache-2.0; no separate data license — "Apache License Version 2.0, January 2004" — https://github.com/FudanDISC/DISC-MedLLM (2026-11-12; Page opened)
- **Upstream terms**: Upstream cMedQA2 README (cmedqa2 card, snapshot s01): non-commercial research only — "available for non-commercial research" — https://github.com/zhangsheng93/cMedQA2 (2026-11-12; Page opened)
- **Upstream terms**: Upstream MedDialog third-party mirror (meddialog-cn card); secondhand — "All copyrights of the data belong to haodf.com." — https://huggingface.co/datasets/UCSD26/medical_dialog (2026-11-12; Second-hand only; second-hand, not verified)

## Source and lineage

Restructured from existing datasets: MedDialog 400k, cMedQA2 20k, CMeKG 50k, human preference data 2k, MedMCQA 8k, MOSS-SFT 33k, Alpaca-GPT4-zh 1k
- Includes: MedDialog (https://mdexa.com/datasets/meddialog-cn/)
- Includes: cMedQA2 (https://mdexa.com/datasets/cmedqa2/)
- Includes: CMeKG
- Includes: MedMCQA
- Includes: MOSS-SFT
- Includes: Alpaca-GPT4-zh

## Known issues with DISC-Med-SFT

- Upstream includes MedDialog (copyright belongs to haodf.com) and cMedQA2 (non-commercial research statement), inconsistent with the Apache-2.0 label
- HF has only a subset (without the behavior preference data)
- Part of the data was reconstructed or generated by an LLM (snapshot s03, quoted in the zh text).

## Preflight

Not run (data not ingested).

## Downloading DISC-Med-SFT and API access

Mdexa does not serve files. Original source:
- https://huggingface.co/datasets/Flmc/DISC-Med-SFT
- https://github.com/FudanDISC/DISC-MedLLM

### Access recipe

- .gitattributes · 2 KB · https://huggingface.co/datasets/Flmc/DISC-Med-SFT/resolve/b10eedfc23530ebe257a149b00db73c4162fbfc3/.gitattributes
- DISC-Med-SFT_released.jsonl · 820.4 MB · sha256 34beda9ccf9a7a59399d853c47b378b8212b41df6f80f67b58502074daa9d545 · https://huggingface.co/datasets/Flmc/DISC-Med-SFT/resolve/b10eedfc23530ebe257a149b00db73c4162fbfc3/DISC-Med-SFT_released.jsonl
- README.md · 1 KB · https://huggingface.co/datasets/Flmc/DISC-Med-SFT/resolve/b10eedfc23530ebe257a149b00db73c4162fbfc3/README.md

## Citing DISC-Med-SFT

No recommended citation text.
- arXiv:2308.14346

## Machine-readable

- JSON: https://mdexa.com/datasets/disc-med-sft.json
- 2026-11-13 License status changed from conflict to restricted and hosting from pending to link-only: the declared license is broader than its upstream sources, so the strictest upstream applies.
- 2026-11-13 Added specialty tags: multi-specialty(inferred).
- 2026-11-13 Added the model-generated risk flag (the dialogues were rebuilt with a language model). Status and hosting unchanged.
- 2026-11-13 Removed the stored size-tier tag; the tier is now computed from the size field when pages are built.
- 2026-11-13 Added dataset-level tags: 4 (2 stated by the authors with excerpts, 2 inferred from the content). No conclusion changed.
- 2026-11-12 Recorded the code repository license separately; it is not treated as a data license statement.
- 2026-11-12 Added verbatim license excerpts and snapshot ids from captured page snapshots; no conclusion changed.
- 2026-11-12 License status split into license_status (what the license says) and hosting (what Mdexa does); the earlier conclusion is kept for traceability.
- 2026-11-12 Card converted from an earlier manual catalog; no new facts added; reclassified under taxonomy v2.

## Continue in this database

This record is one of 109 processed and verified records (datasets, data resources and open models). To find others, call `https://mcp.mdexa.com/v1/datasets?q=<words>` or connect the MCP server at `https://mcp.mdexa.com/mcp`; no key is needed for 20 calls per minute. Rows are under 1 KB; an empty result means "not listed". For a list of items, one batch check (free account) returns a report. Guide: /agents.md. Similar records: CMtMedQA (https://mdexa.com/datasets/cmtmedqa/); Huatuo-Llama-Med-Chinese 知识库问答与文献数据 (https://mdexa.com/datasets/huatuo-llama-med-chinese-data/); BianQueCorpus（扁鹊健康大数据） (https://mdexa.com/datasets/bianque-corpus/).