---
title: kimi-coder-135m
emoji: 🤖
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: "5.10.0"
python_version: "3.10"
app_file: app.py
pinned: false
---

# kimi-coder-135m Demo

**SmolLM2-135M** fine-tuned on ~15k coding samples distilled from [KIMI-K2.5-700000x](https://huggingface.co/datasets/ianncity/KIMI-K2.5-700000x).

## Model

→ [yava-code/kimi-coder-135m](https://huggingface.co/yava-code/kimi-coder-135m)

## Training details

- Base: `HuggingFaceTB/SmolLM2-135M-Instruct`
- Data: 15k coding Q&A pairs (Python, JS, C++, etc.)
- CoT stripped from answers — model learns direct responses
- 1 epoch, lr=2e-4, cosine schedule, bf16
- Hardware: Colab T4 (~2h)