| | --- |
| | language: |
| | - en |
| | license: apache-2.0 |
| | tags: |
| | - code |
| | - reasoning |
| | - coding |
| | - instruct |
| | - 8b |
| | - 1kz |
| | - lfm-inspiration |
| | library_name: transformers |
| | pipeline_tag: text-generation |
| | inference: true |
| | --- |
| | |
| | # bigcodemax |
| |
|
| | **Maximum coding + reasoning power in 8B parameters** |
| | Created by **[1kz](https://huggingface.co/1kz)** |
| |
|
| | An 8B model that punches way above its weight in code generation, software engineering, advanced reasoning, math, and long-context understanding. |
| |
|
| | ## Model Details |
| |
|
| | - **Developer**: [1kz](https://huggingface.co/1kz) |
| | - **Parameters**: 8.0B (dense) |
| | - **Context length**: 128K (RoPE scaled) |
| | - **Architecture**: Llama-3.1 style (same tokenizer & chat template as Meta-Llama-3.1-8B-Instruct) |
| | - **Base model**: Fine-tuned from a strong 8B checkpoint |
| | - **Training inspiration**: Huge thanks to **lfm** for the incredible training recipes, data curation, synthetic data pipelines, and open methodology that made this model possible. Your work continues to inspire and push the frontier for compact high-performance models! ❤️ |
| |
|
| | ## Strengths |
| |
|
| | - Best-in-class code generation, editing, and debugging |
| | - Strong mathematical & logical reasoning (CoT & ToT) |
| | - Excellent at understanding and refactoring large codebases |
| | - Agentic coding, tool use, and multi-step problem solving |
| | - Fast inference on consumer hardware (single 4090 / 24GB VRAM) |
| |
|
| | ## Quick Start |
| |
|
| | ```python |
| | from transformers import pipeline |
| | |
| | pipe = pipeline( |
| | "text-generation", |
| | model="1kz/bigcodemax", |
| | device_map="auto", |
| | torch_dtype="auto" |
| | ) |
| | |
| | messages = [ |
| | {"role": "system", "content": "You are bigcodemax, an expert coding and reasoning assistant."}, |
| | {"role": "user", "content": "Implement a thread-safe LRU Cache in Python with O(1) operations and explain every design choice step-by-step."} |
| | ] |
| | |
| | output = pipe(messages, max_new_tokens=2048, temperature=0.6, top_p=0.95, do_sample=True) |
| | print(output[0]["generated_text"][-1]["content"]) |
| | Benchmarks (internal eval) |
| | |
| | |
| | |
| | Massive thank you to lfm — without your public training logs, data mixing strategies, and relentless open-source experimentation, a model this capable at only 8B would not exist. You're building the future of accessible frontier intelligence. 🚀 |
