Instructions to use Bobkao/BobIDE_AI_2B_MLC with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLC-LLM
How to use Bobkao/BobIDE_AI_2B_MLC with MLC-LLM:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
BobIDE_AI_2B_MLC
本模型為專為 Bob IDE 瀏覽器端 WebGPU 推論(WebLLM)打造的輕量化繁體中文 STEM 與講義摘要模型(q4f16_1 MLC 量化格式)。
模型來源與授權聲明 (Attribution & License)
- **基礎模型 (Base Model)**:Qwen/Qwen3.5-2B(Apache License 2.0,Alibaba Qwen 團隊)
- **蒸餾教師模型 (Teacher Model for SFT Data)**:Qwen/Qwen3.8-27B(Apache License 2.0;使用
unsloth/Qwen3.8-27B-GGUFUD-Q4_K_M生成並清洗為2,185筆繁體中文 STEM/講義摘要高純度訓練語料) - **衍生處理 (Derivative Work)**:基於
Qwen/Qwen3.5-2B以上述開源蒸餾語料進行 QLoRA SFT 微調(<think>/<answer>結構與OFF/LOW/MEDIUM分級控制及 PDF 講義重點摘要),合併權重後以 MLC LLM 匯出為q4f16_1WebGPU 格式。 - 授權條款:承襲原基礎模型採 Apache License 2.0 釋出(完整條款見同目錄 LICENSE 檔案)。
模型規格與執行建議
- 架構 / 量化:
qwen3_5/q4f16_1(權重總計 ~1.03 GB,含本機實測精簡編譯之Qwen3.5-2B-q4f16_1_cs128_b1-webgpu.wasm,即prefill_chunk_size=128, max_batch_size=1) - 建議視窗與分塊:
context_window_size: 1024、prefill_chunk_size: 128 - 推薦快取後端:
cacheBackend: "opfs"(或"indexeddb"),不建議使用預設CacheStorage以避免瀏覽器記憶體異常飆高。
- Downloads last month
- 35
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support