DeepSeek-R1-Distill-Qwen-1.5B-Alpaca-Instruct

Fine-tuned DeepSeek-R1-Distill-Qwen-1.5B for instruction-following tasks using LoRA on the Alpaca dataset.

Overview

• Base Model: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B (1.5B parameters)
• Fine-tuning Method: LoRA (4-bit quantization)
• Dataset: Alpaca instruction dataset (52K samples)
• Training: 3 epochs with optimized hyperparameters

Key Features

• Improved instruction following capabilities
• Conversational AI for question answering
• Memory efficient training with LoRA
• Production-ready merged model

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("sweatSmile/DeepSeek-R1-Distill-Qwen-1.5B-Alpaca-Instruct")
tokenizer = AutoTokenizer.from_pretrained("sweatSmile/DeepSeek-R1-Distill-Qwen-1.5B-Alpaca-Instruct")

# Example
prompt = "Human: What is machine learning?\n\nAssistant:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Details

• LoRA rank: 8, alpha: 16
• 4-bit NF4 quantization with bfloat16
• Learning rate: 1e-4 with cosine scheduling
• Batch size: 8, Max length: 512 tokens

Trained for efficient deployment in production environments.