Downloads · 30 days
0
bio-posttrain-qwen3-4b-protein-sft-lora
mims-harvard/bio-posttrain-qwen3-4b-protein-sft-lora
bio-posttrain-qwen3-4b-protein-sft-lora is a machine learning model from mims-harvard. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as apache-2.0.
Protein supervised fine-tuning (SFT) LoRA adapter from How Post-Training Shapes Biological Reasoning Models.
Access
Public
Updated Jun 22, 2026
Repo size
1.1 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors1.1 GB · 100%
At a glance
- Library
- transformers
- License
- apache-2.0
- Access
- Public
- Created
- Jun 22, 2026
- Updated
- Jun 22, 2026
- SHA
- daebaaed
Base models
Model card
From the Hugging Face model README
Bio-posttrain Qwen3-4B Protein SFT (LoRA)
Protein supervised fine-tuning (SFT) LoRA adapter from How Post-Training Shapes Biological Reasoning Models.
Model details
- Base model:
Qwen/Qwen3-4B-Thinking-2507 - Protein encoder: ESM3
esm3_sm_open_v1(not included) - LoRA: rank 128, alpha 256
- Validation loss: 0.9308
Note: This release contains the text LoRA adapter only (
sft_lora_adapter/). Load the base Qwen model, ESM3, and GO graph encoder separately using BioReason-Pro.
Loading
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B-Thinking-2507", trust_remote_code=True)
model = PeftModel.from_pretrained(base, "mims-harvard/bio-posttrain-qwen3-4b-protein-sft-lora")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B-Thinking-2507")
Collection
Part of the Bio-posttrain collection.
- Library
- transformers
- License
- apache-2.0
- Created
- Jun 22, 2026
- Updated
- Jun 22, 2026