---
language:
- en
license: apache-2.0
tags:
- text-generation
- technical-documentation
- readme
- qwen
- qlora
pipeline_tag: text-generation
base_model: Qwen/Qwen2.5-Coder-7B-Instruct
model-index:
- name: Tech-Scribe-v1
  results:
    - task:
        type: text-generation
        name: Text Generation
      dataset:
        name: collected_data_external
        type: tech-docs
      metrics:
        - type: loss
          value: 1.1258
---

# Tech Scribe (Qwen 2.5 7B Fine-tune)

**Tech Scribe** is a specialized language model fine-tuned to generate high-quality, structured technical documentation (READMEs, Model Cards) from simple project descriptions. It is built on top of `Qwen/Qwen2.5-Coder-7B-Instruct` using QLoRA.

## Usage

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

# Config for 4-bit loading
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_quant_type="nf4"
)

# Load Base Model
base_model_name = "Qwen/Qwen2.5-Coder-7B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
    base_model_name,
    quantization_config=bnb_config,
    device_map="auto"
)

# Load Tech Scribe Adapter
adapter_name = "Darmm/tech-scribe-v1" # Example path
model = PeftModel.from_pretrained(model, adapter_name)
tokenizer = AutoTokenizer.from_pretrained(base_model_name)

# Generate
project_idea = "A Python library for real-time sentiment analysis using websockets"
prompt = f"### Instruction:\nWrite a high-quality technical README or Model Card for the project \"{project_idea}\".\n\n### Response:\n"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=1024, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True).split("### Response:")[1])
```

## Model Description

- **Developed by:** Darmm Lab
- **Base Model:** `Qwen/Qwen2.5-Coder-7B-Instruct`
- **Fine-tuning Method:** QLoRA (4-bit quantization with LoRA adapters)
- **Task:** Technical Documentation Generation
- **Language:** English

## Training (summary)

The model was fine-tuned on a curated dataset of high-quality READMEs from top open-source repositories (e.g., PyTorch, FastAPI, React, HuggingFace Transformers).

- **Epochs:** 1 (Prototype run)
- **Batch size:** 1 (Gradient Accumulation: 8)
- **Learning rate:** 2e-4
- **Optimizer:** AdamW
- **Hardware:** NVIDIA A100 80GB

## Metrics

```json
{
  "eval_loss": 1.1258,
  "train_loss": 1.2937,
  "epoch": 0.73
}
```

## Intended Use

- Rapidly generating boilerplate documentation for new software projects.
- converting rough notes into structured Markdown documentation.
- Learning best practices for technical writing structure.

## Limitations

- **Prototype Status:** This model was trained on a small subset of data for demonstration purposes.
- **Hallucination:** Like all LLMs, it may generate plausible-sounding but incorrect installation instructions or API calls. Always verify the generated code.

## Citation

```bibtex
@misc{techscribe2026,
  author = {Darmm Lab},
  title = {Tech Scribe: Automated Technical Documentation Generator},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face Repository},
  howpublished = {\url{https://huggingface.co/Darmm/tech-scribe-v1}}
}
```