---
license: apache-2.0
base_model: Qwen/Qwen3-8B-Instruct
tags:
  - finance
  - multilingual
  - mapfinben
  - qwen3
  - lora
language:
  - en
  - zh
  - id
  - es
  - el
  - ja
---

# mapfinben-qwen3-merged-unified-v2

Unified LoRA fine-tuned Qwen3-8B-Instruct for **CCL26-Eval-MapFinBen** (v2 continued fine-tune on v1 adapter).

## Model Details

- **Base model:** [Qwen3-8B-Instruct](https://huggingface.co/Qwen/Qwen3-8B-Instruct)
- **Method:** LoRA SFT v1 (rank=16, 1 epoch) + v2 continue (0.5 epoch, lr=5e-5)
- **Training data:** MapFinBen train split, unified 51,064 samples
- **Framework:** LLaMA-Factory
- **Parameters:** ~8B (merged full weights)

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "Ljy2004/mapfinben-qwen3-merged-unified-v2"
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_path, trust_remote_code=True, torch_dtype="auto", device_map="auto"
)

messages = [{"role": "user", "content": "YOUR_PROMPT"}]
text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True, enable_thinking=False
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
```

**Important:** Use Qwen3 chat template with `enable_thinking=False` to match training.

## Citation

MapFinBen benchmark: https://github.com/HgITSE/MapFinBen