Agasthya0/colabmind-coder-6.7b-ml-qlora

Model Card: ColabMind-Coder-6.7B-LoRA

Model Details

• Base Model: deepseek-ai/deepseek-coder-6.7b-base
• Technique: LoRA fine-tuning with PEFT
• Language: English, programming languages (Python, Machine Learning)
• Type: Causal LM for code generation

Intended Uses

• Direct Use: Code completion, code explanation, small script generation
• Downstream Use: Can be fine-tuned for domain-specific code tasks
• Out of Scope: Malicious code generation, production-grade critical systems without human review

Training

• Data: Filtered samples from The-Stack-v2 & curated coding datasets
• Procedure: LoRA fine-tuning on Google Colab (T4 GPU, 8GB VRAM)
• Precision: Mixed fp16

Limitations

• May produce incorrect or insecure code
• Bias from training data may persist
• Not optimized for very large-scale projects

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

model = AutoModelForCausalLM.from_pretrained("Agasthya0/colabmind-coder-6.7b-lora")
tokenizer = AutoTokenizer.from_pretrained("Agasthya0/colabmind-coder-6.7b-lora")
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
print(pipe("def fibonacci(n):")[0]["generated_text"])