---
language:
- fr
- en
license: apache-2.0
library_name: transformers
tags:
- qwen2
- merged
- lora
base_model: Qwen/Qwen3-1.7B
---

# Fablia-Qwen3-Format2 Merged Model

This repository contains the merged version of the Fablia-Qwen3-Format2 model, where the LoRA adapter has been merged with the base model.

## Model Details

- **Base Model**: Qwen/Qwen3-1.7B
- **LoRA Adapter**: Volko76/Fablia-Qwen3-1.7B-LoRa-Format1-WithName
- **Merged Model**: Full precision merged model ready for inference

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the merged model
tokenizer = AutoTokenizer.from_pretrained("Volko76/Fablia-Qwen3-1.7B-Format1-WithName", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    "Volko76/Fablia-Qwen3-1.7B-Format1-WithName",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Generate text
prompt = "P: Bonjour cher ami, que fais-tu ?\nL:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
```

## Original Components

- **Base Model**: [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B)
- **LoRA Adapter**: [Volko76/Fablia-Qwen3-1.7B-LoRa-Format1-WithName](https://huggingface.co/Volko76/Fablia-Qwen3-1.7B-LoRa-Format1-WithName)

## Model Architecture

This model is based on the Qwen3 architecture with the following modifications applied through LoRA fine-tuning and subsequent merging.