alkahest-4b-v2

Browser-oriented ONNX export of a Qwen3.5 Heretic checkpoint packaged for WebGPU / Transformers.js.

Capabilities

  • Supported inputs: text, image, video

Version Notes

  • This is the enhanced browser v2 package for this model family.
  • Compared to the v1 package at thomasjvu/alkahest-4b, it adds support for video.
  • The lighter v1 package remains available at thomasjvu/alkahest-4b if you only need text, image.

Provenance

  • Source model: /home/jovyan/work/heretic-to-onnx/build/phala_gpu_tee/alkahest-4b-direct/inputs/source
  • Base model for inherited processor assets: Qwen/Qwen3.5-4B
  • Architecture family: qwen3_5_conditional_generation
  • Expected architecture: Qwen3_5ForConditionalGeneration
  • Target dtype: q4f16
  • Target device: webgpu

Expected ONNX Sessions

  • vision_encoder_q4f16.onnx
  • embed_tokens_q4f16.onnx
  • decoder_model_merged_q4f16.onnx

Usage

Load this repo with Transformers.js using the model's transformers.js_config metadata and WebGPU backend.

Downloads last month
13
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support