---
type: "post"
id: "humming-block-fp8-path-fails-deepseek-v4-1-ccbde0"
slug: "humming-block-fp8-path-fails-deepseek-v4-1-ccbde0"
title: "Humming block-FP8 path fails DeepSeek-V4.1-Flash on Hopper at tensor parallel 4 and 8"
description: "vLLM maintainer Michael Goin reported an open bug affecting DeepSeek-V4.1-Flash on NVIDIA Hopper GPUs when Humming block-FP8 MoE activations are used with pure tensor parallelism of 4 or 8."
retrieval_nugget: "vLLM maintainer Michael Goin reported an open bug affecting DeepSeek-V4.1-Flash on NVIDIA Hopper GPUs when Humming block-FP8 MoE activations are used with pure tensor parallelism of 4 or 8."
published_at: "2026-10-27"
updated_at: "2026-10-27"
record_date: "2026-10-27"
date_kind: "published_at"
topics: ["inference","models"]
entities: ["vLLM","Humming","DeepSeek-V4.1-Flash","NVIDIA Hopper","NVIDIA H200","DeepSeek"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "A new issue reports that Humming’s block-FP8 MoE path fails to start DeepSeek-V4.1-Flash on Hopper at tensor parallel sizes 4 and 8, exposing a compatibility limitation not stated in the Humming announcement."
prior_refs: ["memory:feb9e401-6837-8ff3-adf1-2e4ddc6c644b"]
source_url: "https://github.com/vllm-project/vllm/issues/63793"
source_urls: ["https://github.com/vllm-project/vllm/issues/63793"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:humming-block-fp8-path-fails-deepseek-v4-1-ccbde0"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/humming-block-fp8-path-fails-deepseek-v4-1-ccbde0/","markdown":"https://newruntime.com/posts/humming-block-fp8-path-fails-deepseek-v4-1-ccbde0.md","json":"https://newruntime.com/posts/humming-block-fp8-path-fails-deepseek-v4-1-ccbde0.json"}
---

# Humming block-FP8 path fails DeepSeek-V4.1-Flash on Hopper at tensor parallel 4 and 8

## Retrieval answer

vLLM maintainer Michael Goin reported an open bug affecting DeepSeek-V4.1-Flash on NVIDIA Hopper GPUs when Humming block-FP8 MoE activations are used with pure tensor parallelism of 4 or 8.

vLLM maintainer Michael Goin reported an open bug affecting DeepSeek-V4.1-Flash on NVIDIA Hopper GPUs when Humming block-FP8 MoE activations are used with pure tensor parallelism of 4 or 8.

The report used eight NVIDIA H200 GPUs, CUDA 13.0, PyTorch 2.14.0+cu130, humming-kernels 0.1.18, and a vLLM build from main. Startup fails during profiling with an AssertionError because the model's MoE intermediate size of 2304 becomes a per-rank K dimension of 576 at TP=4 or 288 at TP=8; neither is divisible by the 128-element input-scale group. TP=1 and TP=2 are listed as working, as is expert parallelism.

The issue says vLLM already pads expert K to a multiple of 128, but Humming must support the remaining partial scale group. A Humming GEMM fix is merged and a quantization fix remains draft, with a Humming release and vLLM dependency bump still planned. Main-branch Hopper serving defaults to BF16 activations, so that default path is described as unaffected. The report does not establish a release date for the fix or confirm the behavior across other models, GPU architectures, or deployments.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Inference](https://newruntime.com/topics/inference.md): Explore the inference topic hub.
- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)