---
library_name: peft
base_model: openlm-research/open_llama_7b_v2
tags:
  - lora
  - qlora
  - reasoning
  - think
  - openthoughts
---

# open-llama-7b-openthought-mid-lora

LoRA adapter for OpenLLaMA 7B v2, fine-tuned on OpenThoughts-114k (DeepSeek-R1 reasoning traces).

## Training

- **Base**: open_llama_7b_v2 (4-bit NF4)
- **Method**: QLoRA (r=32, alpha=32)
- **Data**: open-thoughts/OpenThoughts-114k (filtered <= 2024 tokens, 10,582 samples)
- **Epochs**: 3
- **Batch size**: 64
- **LR**: 2e-4 (cosine, warmup 24 steps)
- **Loss**: full sequence (pad masked)
- **Packing**: greedy bin packing sorted by length
- **Context**: 2024 (native, no RoPE scaling)
- **Special tokens**: Qwen3.5 chat template
- **modules_to_save**: embed_tokens, lm_head

## Format

```
<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
{question}<|im_end|>
<|im_start|>assistant
<think>
{reasoning}
</think>

{answer}<|im_end|>
```

## Merged model

See [ping98k/open-llama-7b-openthought-mid-4bit](https://huggingface.co/ping98k/open-llama-7b-openthought-mid-4bit)