README.md

---
license: apache-2.0
base_model: mistralai/Mistral-7B-v0.1
tags:
  - mistral
  - text-generation
  - copywriting
  - social-media
  - twitter
  - french
  - english
inference: true
---

# X-Bot LLM

A fine-tuned Mistral-7B model specialized for copywriting and social media content generation, particularly optimized for Twitter/X posts.

## Model Description

This model is based on Mistral-7B-v0.1 and has been fine-tuned for generating engaging social media content with flexible length, tone, and goal options.

### Key Features

- **Flexible Length Control**: Generate content from short tweets to longer threads
- **Tone Adaptation**: Supports professional, neutral, promotional, friendly, and moody tones
- **Goal-Oriented**: Optimized for reach, discussion, out-of-context, or neutral content
- **4-bit Quantization**: Optimized for efficient inference with BitsAndBytes

## Model Details

- **Base Model**: mistralai/Mistral-7B-v0.1
- **Architecture**: MistralForCausalLM
- **Quantization**: 4-bit (NF4) with BitsAndBytes
- **Context Length**: 32,768 tokens
- **Languages**: French and English

## Usage

### Using Transformers

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
from transformers import BitsAndBytesConfig
import torch

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("hugov/x-bot-llm", trust_remote_code=True)

# Configure quantization for GPU
quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
)

# Load model
model = AutoModelForCausalLM.from_pretrained(
    "hugov/x-bot-llm",
    quantization_config=quantization_config,
    device_map="auto",
    trust_remote_code=True,
    dtype=torch.float16
)

# Generate text
prompt = "Write a tweet about artificial intelligence"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=100,
        temperature=0.8,
        top_p=0.9,
        do_sample=True
    )

generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_text)
```

### Using Hugging Face Inference API

```python
import requests

API_URL = "https://router.huggingface.co/hf-inference/v1/models/hugov/x-bot-llm"
headers = {"Authorization": f"Bearer {YOUR_HF_TOKEN}"}

def generate(prompt, max_new_tokens=100, temperature=0.8, top_p=0.9):
    payload = {
        "inputs": prompt,
        "parameters": {
            "max_new_tokens": max_new_tokens,
            "temperature": temperature,
            "top_p": top_p,
            "return_full_text": False
        }
    }
    response = requests.post(API_URL, headers=headers, json=payload)
    return response.json()

# Example
result = generate("Write a tweet about AI")
print(result)
```

## Generation Parameters

### Length Options
- `really_short`: 40 tokens (~80-150 chars, 1-2 phrases)
- `short`: 60 tokens (~150-220 chars, single idea)
- `normal`: 100 tokens (~220-280 chars, complete thought)
- `developed`: 200 tokens (thread starter or multiple tweets)

### Goal Options
- `reach`: Maximum creativity for virality (temperature: 0.95, top_p: 0.95)
- `discussion`: Moderate for conversations (temperature: 0.85, top_p: 0.9)
- `out_of_context`: Balanced for standalone content (temperature: 0.8, top_p: 0.85)
- `neutral`: Default balanced settings (temperature: 0.8, top_p: 0.9)

### Tone Options
- `pro`: Professional (slightly more conservative)
- `neutral`: No adjustment
- `promo`: Promotional (more creative)
- `friendly`: Friendly (slightly more creative)
- `moody`: Emotional (more creative)

## Requirements

- Python 3.8+
- PyTorch 2.0+
- Transformers 4.57.1+
- BitsAndBytes (for 4-bit quantization)
- CUDA-capable GPU (recommended)

## Limitations

- This model requires CUDA/GPU support for local inference due to 4-bit quantization
- CPU inference is not supported
- For CPU usage, consider using the Hugging Face Inference API

## Citation

```bibtex
@model{hugov/x-bot-llm,
  title={X-Bot LLM: A Fine-tuned Mistral-7B for Social Media Copywriting},
  author={hugov},
  year={2024},
  base_model={mistralai/Mistral-7B-v0.1}
}
```

## License

Apache 2.0