CoolFace
Modelpublic

dominant-strategies/quai-qwen3.5-27b-igemm-int8

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face
1likes7downloads
Model Card

Quai Inference Qwen3.5-27B IGEMM Int8

Built for Quai Inference.

This repository contains InferenceGemm/Tensor Work Proof W8A8 quantized checkpoint artifacts derived from Qwen/Qwen3.5-27B. It is not an official Qwen release.

Contents

  • —igemm-quantized-*.safetensors: quantized tensor payloads.
  • —manifest.json: verifier-facing tensor manifest and Merkle roots.
  • —index.json: checkpoint index for the quantized tensor shards.
  • —runtime-alias-audit.json, when present: runtime-to-manifest alias audit.
  • —tokenizer/config files copied from the base model directory.

Benchmark Status

Strict SGLang on a Vast.ai NVIDIA A100-SXM4-80GB: 105.26 baseline tok/s, 96.97 receipt-mode tok/s, 7.87% overhead, 1 accepted strict Tensor Work Receipt, 30,096-byte canonical receipt, and Go strict dense policy verification.

The paper benchmark rows were produced with strict SGLang Tensor Work Receipt emission and Go verification. The raw benchmark logs, receipts, and local evidence packet are intentionally not uploaded to this model repository.

Loading

These files are intended for the Quai InferenceGemm harness in this repository, not vanilla transformers weight loading:

text
checkpoints/<this-checkpoint>/

Use the base model tokenizer/config with the InferenceGemm quantized payloads and verifier manifest.

Upstream

  • —Base model: Qwen/Qwen3.5-27B
  • —Upstream license: apache-2.0

Redistribution must comply with the upstream model license and applicable export control restrictions.