---
library_name: peft
base_model: Qwen/Qwen3-30B-A3B
tags:
  - bioreasoning
  - sft
  - lora
  - tinker
---

# bioreasoning-qwen3-30ba3b-sft-20260218-ckpt-1500

LoRA adapter checkpoint from supervised fine-tuning on the [bioreasoning dataset](https://huggingface.co/datasets/abugoot-primeintellect/bioreasoning_v0211_prime_sft).

## Training Details

- **Base model:** Qwen/Qwen3-30B-A3B
- **Method:** LoRA (rank 32)
- **Training platform:** [Tinker](https://tinker-docs.thinkingmachines.ai/) (ThinkingMachines)
- **Dataset:** abugoot-primeintellect/bioreasoning_v0211_prime_sft (~193k train examples)
- **Hyperparameters:** batch_size=128, lr=2e-5 (linear decay), weight_decay=0.01, seq_len=8192
- **Epochs:** 3
- **Checkpoint:** step 001500 (epoch 0, batch 1500)
- **Loss masking:** Last assistant message only

## All Checkpoints

| Step | Epoch | HF Repo |
|------|-------|---------|
| 1500 | ~1 | [bioreasoning-qwen3-30ba3b-sft-20260218-ckpt-1500](https://huggingface.co/abugoot-primeintellect/bioreasoning-qwen3-30ba3b-sft-20260218-ckpt-1500) |
| 3000 | ~2 | [bioreasoning-qwen3-30ba3b-sft-20260218-ckpt-3000](https://huggingface.co/abugoot-primeintellect/bioreasoning-qwen3-30ba3b-sft-20260218-ckpt-3000) |
| final | 3 | [bioreasoning-qwen3-30ba3b-sft-20260218-ckpt-final](https://huggingface.co/abugoot-primeintellect/bioreasoning-qwen3-30ba3b-sft-20260218-ckpt-final) |