Switch Qwen3.5-0.8B int4 quant to k_quant_mixed (block_size=32)

[xiaofeihan1]

Summary

Update the int4 quantization configuration for Qwen3.5-0.8B in both the CUDA and WebGPU recipes:

• int4_algo_config: k_quant_linear → k_quant_mixed
• int4_block_size: add 32

Files changed

• Qwen-Qwen3.5-0.8B/builtin/cuda/text.json
• Qwen-Qwen3.5-0.8B/builtin/webgpu/text.json

Motivation

k_quant_mixed with a 32-block size produces better accuracy/quality for the Qwen3.5-0.8B text model than the previous k_quant_linear setting, while keeping the int4 precision and accuracy level unchanged.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Updates the int4 quantization configuration for the Qwen3.5-0.8B text model to use a mixed quantization algorithm and explicitly sets a quantization block size.

Changes:

• Switch int4_algo_config from k_quant_linear to k_quant_mixed for CUDA and WebGPU builders
• Add int4_block_size: 32 to both backend configs

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
Qwen-Qwen3.5-0.8B/builtin/webgpu/text.json	Switches int4 algo config to mixed and adds explicit block size for WebGPU
Qwen-Qwen3.5-0.8B/builtin/cuda/text.json	Switches int4 algo config to mixed and adds explicit block size for CUDA

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[xiaofeihan1]

@apsonawane Added 8bit quantized in lm_head for Qwen3.5 0.8B. PTAL, thanks!