Skip to content

(recipe)update GLM5.2 agentic recipe - #2117

Draft
zhuyuhua-v wants to merge 1 commit into
mainfrom
yuhua/update_glm5.2_recipe
Draft

(recipe)update GLM5.2 agentic recipe#2117
zhuyuhua-v wants to merge 1 commit into
mainfrom
yuhua/update_glm5.2_recipe

Conversation

@zhuyuhua-v

@zhuyuhua-v zhuyuhua-v commented Sep 2, 2026

Copy link
Copy Markdown
Collaborator

Motivation

Technical Details

Test Plan

Test Result

image

Submission Checklist

Signed-off-by: zhuyuhua-v <yuhzhu@amd.com>
@github-actions

github-actions Bot commented Sep 2, 2026

Copy link
Copy Markdown
Contributor

🏷️ CI Guide

Runs automatically on every eligible PR before approval:

  • ✅ Pre Checkin: Black, Ruff, catalog schema validation, non-GPU unit tests

Heavy model tests:

  • ✅ Run after the PR is approved and Pre Checkin passes
  • ✅ Run immediately when an approval review is submitted
  • ✅ Can be requested before approval with labels
Label Tests
ci:full Run all heavy PR model tests: native ATOM, vLLM, and SGLang
ci:atom Run native ATOM model accuracy tests
ci:vllm Run ATOM vLLM OOT model accuracy tests
ci:sglang Run ATOM SGLang model accuracy tests

Heavy jobs are skipped when the PR is not approved and no matching ci:* label is present.
Add labels via the sidebar or gh pr edit 2117 --add-label <label>

@zjin-lcf

zjin-lcf commented Sep 2, 2026

Copy link
Copy Markdown

@zhuyuhua-v

Can you please explain the changes in details ?

@zhuyuhua-v zhuyuhua-v changed the title (recipe)upodate GLM5.2 agentic recipe (recipe)update GLM5.2 agentic recipe Sep 10, 2026
giovanniguastiamd added a commit to SemiAnalysisAI/InferenceX that referenced this pull request Sep 10, 2026
…age)

- Bump Docker image: glm52_agentic_0813 → glm52_agentic_0902
- MTP: 3 → 4 speculative tokens; switch acceptance from rate (0.6633)
  to forced length (3.33) via --spec-decode-acceptance-length
- Add LMCACHE_NUMA_MODE=auto
- Add DCP arm (tp=4, dcp-size=4, conc=16) replacing the old TP8 arm
- Extend MTP conc-list: [2,4,8,10] → [2,4,8,10,12]
- Script: branch on DCP_SIZE>1 to pass --decode-context-parallel-size
  and suppress spec-decoding args for large-concurrency runs

Recipe reference: ROCm/ATOM#2117

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
giovanniguastiamd added a commit to giovanniguastiamd/InferenceX that referenced this pull request Sep 10, 2026
…, DCP, 0902 image)

- Bump Docker image: glm52_agentic_0813 → glm52_agentic_0902
- MTP: 3 → 4 speculative tokens; switch acceptance from rate (0.6633)
  to forced length (3.33) via --spec-decode-acceptance-length
- Add LMCACHE_NUMA_MODE=auto
- Add DCP arm (tp=4, dcp-size=4, conc=16) replacing the old TP8 arm
- Extend MTP conc-list: [2,4,8,10] → [2,4,8,10,12]
- Script: branch on DCP_SIZE>1 to pass --decode-context-parallel-size
  and suppress spec-decoding args for large-concurrency runs

Recipe reference: ROCm/ATOM#2117

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
giovanniguastiamd added a commit to giovanniguastiamd/InferenceX that referenced this pull request Sep 10, 2026
…, DCP, 0902 image)

- Bump Docker image: glm52_agentic_0813 → glm52_agentic_0902
- MTP: 3 → 4 speculative tokens; switch acceptance from rate (0.6633)
  to forced length (3.33) via --spec-decode-acceptance-length
- Add LMCACHE_NUMA_MODE=auto
- Add DCP arm (tp=4, dcp-size=4, conc=16) replacing the old TP8 arm
- Extend MTP conc-list: [2,4,8,10] → [2,4,8,10,12]
- Script: branch on DCP_SIZE>1 to pass --decode-context-parallel-size
  and suppress spec-decoding args for large-concurrency runs

Recipe reference: ROCm/ATOM#2117

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@zhuyuhua-v

Copy link
Copy Markdown
Collaborator Author

@zhuyuhua-v

Can you please explain the changes in details ?

We use TP + MTP for low conc and TP + DCP for high conc to achieve higher performance in agentic scenarios

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