Alkahest 0.8B Q4 WebGPU

Browser-oriented ONNX package for the Heretic-modified Alkahest 0.8B checkpoint.

This is a test package published before replacing the older thomasjvu/alkahest-0.8b browser target.

Runtime Contract

  • Base processor/model family: Qwen/Qwen3.5-0.8B
  • Source checkpoint: thomasjvu/alkahest-0.8b-heretic-merged
  • Target runtime: Transformers.js / WebGPU
  • Dtype map:
    • embed_tokens: q4
    • decoder_model_merged: q4
    • vision_encoder: fp16

Expected ONNX Sessions

  • onnx/vision_encoder_fp16.onnx
  • onnx/embed_tokens_q4.onnx
  • onnx/decoder_model_merged_q4.onnx

The package passed the repo validation gate on Kaggle version 19, including CPU onnxruntime session creation for all three packaged ONNX sessions.

Downloads last month
67
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support