---
license: apache-2.0
base_model: Qwen/Qwen2.5-Coder-3B-Instruct
tags:
  - n8n
  - workflow
  - json
  - code
  - fine-tuned
  - merged
language:
  - fr
  - en
pipeline_tag: text-generation
---

# Qwen Coder 3B - N8N Workflow Generator (Merged)

This is the **merged** version of the LoRA fine-tuned model for generating N8N workflow JSON.

## Model Details

- **Base Model**: Qwen/Qwen2.5-Coder-3B-Instruct
- **Fine-tuned adapter**: eclaude/qwen-coder-3b-n8n-sft
- **Training**: SFT on 8,782 N8N workflow examples
- **Task**: Generate valid N8N workflow JSON from natural language prompts (French)

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("eclaude/qwen-coder-3b-n8n-merged")
tokenizer = AutoTokenizer.from_pretrained("eclaude/qwen-coder-3b-n8n-merged")

prompt = "Crée un workflow qui récupère des données d'une API et les envoie sur Slack"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=2048)
print(tokenizer.decode(outputs[0]))
```

## Training Data

- Dataset: [eclaude/n8n-workflows-sft](https://huggingface.co/datasets/eclaude/n8n-workflows-sft)
- 8,782 training samples
- 1,197 evaluation samples

## Metrics

- Training loss: 1.04
- Eval loss: 1.02
- Token accuracy: 73%