Skip to content

docs: add flashcomm feature guide.#34

Merged
XuZhang99 merged 1 commit into
xLLM-AI:mainfrom
Cooofish:docs/flashcomm-feature-guide
Jul 7, 2026
Merged

docs: add flashcomm feature guide.#34
XuZhang99 merged 1 commit into
xLLM-AI:mainfrom
Cooofish:docs/flashcomm-feature-guide

Conversation

@Cooofish

@Cooofish Cooofish commented Jul 7, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Add Chinese and English FlashComm feature documentation.
  • Explain the FC1 sequence-sharding design and MMRS fused Matmul + ReduceScatter path.
  • Document suitable workloads, default-off flags, recommended configuration, and validation guidance.

Validation

  • npm run check
  • npm run build

- Document the FlashComm sequence-sharding and MMRS fused-operator design in both Chinese and English.

- Explain the suitable long-prefill NPU workloads, default-off behavior, recommended flags, and validation checklist.

- Include practical configuration guidance for TP>=8, Graph Mode, chunked prefill, and MMRS comm_mode fallback.
@Cooofish Cooofish changed the title docs: add flashcomm feature guide docs: add flashcomm feature guide. Jul 7, 2026
@XuZhang99 XuZhang99 merged commit 2da9b09 into xLLM-AI:main Jul 7, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants