Fox-AI-by-teolm30/Ult1-coding

1---2language: en3library_name: transformers4base_model: Qwen/Qwen2.5-3B-Instruct5pipeline_tag: text-generation6tags:7- qwen8- qwen2.59- 3b10- lora11- coding12- code13- software-engineering14license: apache-2.015---16 17# Ult1-Coding18 19A 3-billion-parameter coding specialist -- master-level software engineer.20 21Based on Qwen2.5-3B-Instruct with an embedded master programmer system prompt containing few-shot coding demonstrations and a coding-focused LoRA adapter (rank 16, 8 target module types).22 23Usage:24```python25from transformers import AutoModelForCausalLM, AutoTokenizer26 27model = AutoModelForCausalLM.from_pretrained("teolm30/Ult1-coding")28tokenizer = AutoTokenizer.from_pretrained("teolm30/Ult1-coding")29 30messages = [{"role": "user", "content": "Write a Python async web scraper with retry logic"}]31text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)32inputs = tokenizer(text, return_tensors="pt")33outputs = model.generate(**inputs, max_new_tokens=512)34print(tokenizer.decode(outputs[0], skip_special_tokens=True))35```36 37The system prompt with few-shot examples is auto-injected by the chat template - no manual system prompt needed.38 39GGUF: Download Ult1-Coding-Q8_0.gguf for CPU inference with llama.cpp.40 41Training Data: training_data.json contains 10 coding Q&A pairs (Python, JavaScript, Rust, SQL, TypeScript, Go). Use with train.py on a GPU.42 43Details:44- Base: Qwen2.5-3B-Instruct (3B params)45- LoRA: Rank 16, targets q/k/v/o + gate/up/down projections46- Context: 32,768 tokens47- Focus: Code generation, algorithms, system design, debugging48