phyen / README.md
parani01's picture
Create README.md
988909d verified
# 🧠 Phyen — Fine-Tuned Qwen Model for Physics and Engineering Reasoning
**Model ID:** [parani01/phyen](https://huggingface.co/parani01/phyen)
**Base Model:** Qwen (7B or equivalent)
**Type:** Fully fine-tuned and merged model (not LoRA)
**Framework:** PyTorch + Transformers
**Files:** 4 × `.safetensors` shards + tokenizer
---
## 🧩 Model Summary
**Phyen** is a specialized variant of the Qwen model, fine-tuned for **physics**, **engineering**, and **technical scientific reasoning**.
It has been trained and merged to perform better on domain-specific text such as:
- Thermodynamics
- Fluid mechanics
- Structural analysis
- General physics conceptual reasoning
It retains general Qwen language ability but prioritizes scientific precision.
---
## 🚀 Usage Example
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "parani01/phyen"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
prompt = "Explain the laws of thermodynamics in simple words."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=120)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
---
## 📘 Intended Use
**Intended for:**
- Physics and engineering question answering
- Scientific writing and conceptual reasoning
- Educational or research assistants
**Not intended for:**
- General conversation
- Legal or medical advice
- Sensitive or factual decision-making outside the training domain
---
## ⚙️ Technical Details
| Field | Description |
|-------|--------------|
| **Architecture** | Qwen-style Transformer |
| **Parameters** | ~7 Billion |
| **Precision** | bfloat16 / float16 (auto-detect) |
| **Framework** | PyTorch + safetensors |
| **Tokenizer** | Qwen tokenizer |
---
## ⚠️ Limitations
- May hallucinate answers outside scientific domains
- Requires GPU (≥16GB VRAM recommended) for efficient inference
- Does not guarantee factual correctness in all contexts
---
## 🧩 Training Information
- **Base Model:** Qwen-7B (open-source base)
- **Fine-tuned with:** domain-specific corpus of physics and engineering text
- **Merged into:** full model weights (`merged_vlm_physics`)
---
## 🏷️ License
Add your license here (for example, Apache-2.0 or MIT).
Ensure you comply with the original Qwen base model’s license.
---
## 👨‍💻 Author
Developed and fine-tuned by **Parani Dharan**
Published at: [Hugging Face — parani01](https://huggingface.co/parani01)
---
## 💬 Citation
If you use this model, please cite it as:
```
@model{parani2025phyen,
title={Phyen: Fine-tuned Qwen Model for Physics and Engineering Reasoning},
author={Parani Dharan},
year={2025},
publisher={Hugging Face}
}
```