metadata
license: mit
datasets:
- iamtarun/python_code_instructions_18k_alpaca
language:
- en
- fr
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
pipeline_tag: text-generation
library_name: mlx
tags:
- code
- python
- deepseek
- fine-tuned
- lora
DeepSeek-R1-Distill-Qwen-7B — Python Code Fine-tune
A LoRA fine-tuned version of DeepSeek-R1-Distill-Qwen-7B specialized for Python code generation.
Model Details
Model Description
- Developed by: Armand (@ArmanS11)
- Model type: Large Language Model — LoRA fine-tune
- Language(s): English
- License: MIT
- Finetuned from: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
Model Sources
- Base model: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
- Training dataset: https://huggingface.co/datasets/iamtarun/python_code_instructions_18k_alpaca
Uses
Direct Use
Generate Python code from natural language instructions. Examples:
- Writing functions, classes, algorithms
- Async/await patterns
- Data structures and error handling
Out-of-Scope Use
- Not intended for other programming languages
- Not suitable for production security-critical code without review
Bias, Risks, and Limitations
Generated code should always be reviewed before use in production. The model may occasionally produce syntactically incorrect code, particularly for complex async patterns.
Training Details
Training Data
iamtarun/python_code_instructions_18k_alpaca — 18,612 Python code instruction/response pairs.
- Train split: 17,681 examples
- Validation split: 931 examples
Training Hyperparameters
| Parameter | Value |
|---|---|
| Method | LoRA |
| LoRA Rank | 8 |
| LoRA Layers | 8 |
| Learning Rate | 5e-6 |
| Batch Size | 2 |
| Iterations | 2000 |
| Quantization | 4-bit |
Technical Specifications
Compute Infrastructure
Hardware
- Apple MacBook Pro M4 — 16 GB unified memory
Software
- MLX (Apple Silicon optimized)
- M-Courtyard fine-tuning app
Model Card Authors
Armand — @ArmandS11