Upload BioMamba-MedQA-SFT-1.3b weights

README.md

@@ -7,28 +7,35 @@ tags:
 - mamba2
 - biomedical
 - pubmed
-- medqa
-- bioasq
 - pubmedqa
+- bioasq
+- medqa
 - sft
 library_name: mamba-ssm
 base_model: zmzfpc/biomamba-1.3b
 datasets:
 - qiaojin/PubMedQA
-- GBaker/MedQA-USMLE-4-options
-- nanyy1025/bioasq_7b_yesno
 ---
 
-# BioMamba-MedQA-SFT-1.3b
+# BioMamba-BioMedQA-SFT-1.3b
+
+BioMamba-1.3b supervised-finetuned for biomedical Yes/No/Maybe question
+answering. The base model is the CPT checkpoint
+[`zmzfpc/biomamba-1.3b`](https://huggingface.co/zmzfpc/biomamba-1.3b).
+
+## Training data (SFT)
+
+- **PubMedQA** (`qiaojin/PubMedQA`, `pqa_labeled` split) — MIT
+- **BioASQ** (Yes/No subset, BioASQ 7b + 13b combined, locally mixed) — BioASQ DUA, non-commercial research use only
+
+Instruction format: biomedical research question → answer **yes / no / maybe**.
 
-BioMamba-1.3b supervised-finetuned for biomedical multiple-choice QA
-(USMLE-style / MedQA). The base model is the CPT checkpoint
-[`zmzfpc/biomamba-1.3b`](https://huggingface.co/zmzfpc/biomamba-1.3b),
-further trained on a mixture of:
+## Evaluation
 
-- **PubMedQA** (`pqa_labeled` split) — MIT
-- **BioASQ 7b Yes/No** (`nanyy1025/bioasq_7b_yesno`) — non-commercial research use only
-- **MedQA USMLE 4-option** — MIT
+We evaluate on **PubMedQA**, **BioASQ (Yes/No)**, and **MedQA (USMLE 4-option)**.
+**MedQA is not included in the SFT data** — the model is run on MedQA in a
+transfer (non-in-distribution) setting, which is why this family is named
+`biomedqa` rather than `medqa`.
 
 ## Loading
 
@@ -38,16 +45,17 @@ The checkpoint is saved in **mamba-ssm native format**, load with `mamba-ssm`:
 from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
 from transformers import AutoTokenizer
 
-model = MambaLMHeadModel.from_pretrained("zmzfpc/biomamba-medQA-sft-1.3b", device="cuda", dtype="bfloat16")
-tokenizer = AutoTokenizer.from_pretrained("zmzfpc/biomamba-medQA-sft-1.3b")
+model = MambaLMHeadModel.from_pretrained("zmzfpc/biomamba-biomedqa-sft-1.3b", device="cuda", dtype="bfloat16")
+tokenizer = AutoTokenizer.from_pretrained("zmzfpc/biomamba-biomedqa-sft-1.3b")
 ```
 
 `AutoModelForCausalLM.from_pretrained` will **not** work on this config.
 
 ## Intended use
 
-Biomedical QA (MedQA, PubMedQA, BioASQ Yes/No). Not intended for clinical
-decision-making and not validated on any patient-facing task.
+Biomedical Yes/No/Maybe QA and transfer to MedQA-style multiple-choice.
+Not intended for clinical decision-making and not validated on any
+patient-facing task.
 
 ## Citation