Note: This is a comprehensive fine-tune with extensive parameter modifications, resulting in a larger adapter size compared to typical LoRA adapters. This enables more significant model adaptations while maintaining compatibility with PEFT infrastructure.

🚀 Quick Start

Installation

pip install transformers peft torch

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2-7B-Instruct",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Load adapter
model = PeftModel.from_pretrained(
    base_model,
    "InedxsAI/Inedxs.AI"
)

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(
    "Qwen/Qwen2-7B-Instruct",
    trust_remote_code=True
)

# Generate
prompt = "Bonjour, qui es-tu ?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📊 Training Details

Training Configuration

Framework: PEFT 0.17.1
LoRA Rank: Optimized for conversational tasks
Target Modules: Attention and feedforward layers
Training Precision: Mixed precision (FP16/BF16)

Training Data

Fine-tuned on conversational datasets focusing on:

Instruction following
Dialogue and conversation
Question answering
Task completion

🎯 Intended Uses

Primary Use Cases

Conversational AI assistants
Instruction-following chatbots
Interactive question-answering systems
General-purpose dialogue agents

Out-of-Scope Uses

Medical diagnosis or advice
Legal counseling
Financial advice without proper disclaimers
Generating harmful or malicious content

⚡ Performance

This adapter enhances Qwen2-7B-Instruct's capabilities in:

Conversational coherence
Instruction understanding
Task-specific responses
Multi-turn dialogue

🔧 Integration Options

Option 1: Merge with Base Model

python

# Merge adapter for inference
merged_model = model.merge_and_unload()
merged_model.save_pretrained("./merged_model")

Option 2: Convert to GGUF

For efficient inference with llama.cpp or Ollama, see the quantized version: 👉 InedxsAI/Inedxs.AI-GGUF

🛡️ Limitations & Biases

Inherits limitations from base Qwen2-7B-Instruct model
May exhibit biases present in training data
Performance varies with prompt quality and context
Not suitable for critical decision-making without human oversight

📜 Citation

bibtex

@misc{inedxsai2025,
  title={InedxsAI: PEFT Adapter for Qwen2-7B-Instruct},
  author={InedxsAI},
  year={2025},
  howpublished={\url{https://huggingface.co/InedxsAI/Inedxs.AI}}
}

🔗 Related Resources

Base Model: Qwen/Qwen2-7B-Instruct
GGUF Version: InedxsAI/Inedxs.AI-GGUF
Framework: PEFT by Hugging Face

📧 Contact

For questions or issues, please open an issue on the model repository.

Made with ❤️ by InedxsAI