Base model: Qwen/Qwen3.8-27B (Apache-2.0)
Quantization: AutoRound, INT4, group size 128 (w4g128), symmetric, with 16-bit exceptions for linear_attn.in_proj_a/b and mtp.fc
Size: ~18 GB (vs ~52 GB for BF16)
Context: 256K
Architecture: qwen3_5 (Qwen3_5ForConditionalGeneration), 64 layers, 3:1 linear:full attention

Why

Runs a 27B dense model on 4x Intel Arc Pro B70 (32 GB each) with vLLM tensor parallelism. INT4 fits comfortably and runs faster than BF16 on this hardware.

Load with vLLM

bash

vllm serve devan-carlin/Qwen3.8-27B-int4-AutoRound \
  --quantization auto-round \
  --tensor-parallel-size 4 \
  --max-model-len 262144 \
  --kv-cache-dtype fp8 \
  --enable-prefix-caching \
  --gpu-memory-utilization 0.85 \
  --trust-remote-code

Note: loading symmetric INT4 checkpoints with empty-qzeros layers on the XPU/ARK path requires the qzeros guard fix (vllm PR #52428). Without it, symmetric layers crash with a copy_() shape mismatch during weight loading.

Benchmark (4x Intel Arc Pro B70, TP=4, max_tokens=16384)

Model	Gen speed	Notes
INT4 (this repo)	47.8 tok/s	18 GB
Qwen3.8-27B BF16	30.2 tok/s	~52 GB

58% faster than BF16
Quality parity on math, logic, algorithm, and code prompts (identical final answers)
INT4 reasons more verbosely (more completion tokens), so on very long outputs it can hit a fixed max_tokens cap earlier — raise the cap if needed

Quantization details

Method: AutoRound (Intel)
Bits: 4, group size: 128, symmetric
Only model.language_model.layers is quantized to INT4; embeddings, lm_head, the MTP head, and the vision tower stay in their original precision (model stays MTP-capable)
Within the quantized layers, linear_attn.in_proj_a/b are kept 16-bit

License

Apache-2.0 (same as the base model).

Notes

Multimodal. The repo ships the Qwen3-VL vision tower (model.visual.*) plus processor configs, and the vLLM Qwen3_5ForConditionalGeneration port loads it. Send images as image_url content parts (base64 or URL); text-only prompts work unchanged.

Quantized and benchmarked by devan-carlin on 4x Intel Arc Pro B70 (XPU).