sylvester-francis/typescript-slm-7b-reasoning

TypeScript SLM 7B - Reasoning Variant

7B TypeScript model with reasoning capabilities for TypeScript code generation, optimized for React, Next.js, Angular, and Node.js.

Model Details

• Base Model: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
• Model Size: 7B parameters
• Training Method: LoRA (Low-Rank Adaptation)
• Context Length: 2048 tokens
• LoRA Rank: 64
• Training Dataset: 5,000 high-quality TypeScript samples

Reasoning Capabilities

This variant includes chain-of-thought reasoning for better code understanding and generation.

Training Configuration

• Batch Size: 2
• Gradient Accumulation: 16
• Effective Batch Size: 32
• Learning Rate: 0.0001
• Epochs: 3
• Hardware: Google Colab A100 40GB

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

base_model = "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B"
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    device_map="auto",
    torch_dtype="auto"
)
tokenizer = AutoTokenizer.from_pretrained(base_model)

# Load LoRA adapter
model = PeftModel.from_pretrained(model, "sylvester-francis/typescript-slm-7b-reasoning")

# Generate code
prompt = "Write a React component with TypeScript:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0]))

Repository

https://github.com/sylvester-francis/slm-typescript-model