🧠 `qwen2.5-md-finetuned`

Model Overview

qwen2.5-md-finetuned is a fine-tuned version of the Qwen2.5-Medium model, optimized for improved performance on domain-specific or task-specific data. This model leverages the powerful multilingual and multitask capabilities of the base Qwen2.5 architecture and is adapted further using Low-Rank Adaptation (LoRA) techniques for efficient fine-tuning.

✅ Base Model: Qwen2.5-Medium 🛠️ Fine-Tuned By: adi2606 📜 License: MIT 🧱 Adapter Format: adapter_model.safetensors (LoRA)

📌 Use Cases

This model is best suited for:

Custom conversational agents
Code or documentation assistants
Knowledge-based QA systems
Any application benefiting from Qwen2.5’s capabilities but requiring domain-specific fine-tuning

🔧 Fine-Tuning Details

Technique: Parameter-efficient fine-tuning using LoRA
Adapter Config: See adapter_config.json
Tokenizer: Includes full tokenizer configuration (tokenizer_config.json, vocab.json, merges.txt)
Additional Tokens: added_tokens.json and special_tokens_map.json for enhanced compatibility with downstream applications

💾 Files

Filename	Description
`adapter_model.safetensors`	LoRA adapter weights
`adapter_config.json`	Adapter configuration for inference
`tokenizer_config.json`	Tokenizer configuration
`tokenizer.json`	Pre-tokenized vocabulary
`vocab.json`	Vocabulary JSON
`merges.txt`	Merge rules for BPE tokenizer
`special_tokens_map.json`	Special tokens mapping
`added_tokens.json`	Custom added tokens
`chat_template.jinja`	Custom chat template (if applicable)

✅ How to Use

You can load this adapter with the base Qwen2.5-Medium model using peft:

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

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-Medium", device_map="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("adi2606/qwen2.5-md-finetuned", trust_remote_code=True)
model = PeftModel.from_pretrained(base_model, "adi2606/qwen2.5-md-finetuned")

📈 Performance

(Optional section) If you have evaluation metrics or benchmark results, they can be added here. Example:

Domain accuracy: 89.3%
BLEU/ROUGE/F1 scores if applicable

📚 Citation

If you use this model in your work, please consider citing it:

@misc{adi2606qwen25md,
  author = {adi2606},
  title = {qwen2.5-md-finetuned},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/adi2606/qwen2.5-md-finetuned}},
}

🤝 Contributions

If you find issues or would like to contribute improvements to the model or tokenizer, feel free to open a pull request or discussion on the model repository.

🧠 qwen2.5-md-finetuned