Skip to content

[atom CI/Nightly/Benchmark] Add MiniMax-M3 and Eagle#1356

Merged
valarLip merged 2 commits into
mainfrom
zejun/add_minimax_m3_into_atom_ci_benchmark_0625
Jun 25, 2026
Merged

[atom CI/Nightly/Benchmark] Add MiniMax-M3 and Eagle#1356
valarLip merged 2 commits into
mainfrom
zejun/add_minimax_m3_into_atom_ci_benchmark_0625

Conversation

@zejunchen-zejun

Copy link
Copy Markdown
Collaborator
image image

into atom infra

Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
@valarLip valarLip marked this pull request as ready for review June 25, 2026 14:53
Copilot AI review requested due to automatic review settings June 25, 2026 14:53
@valarLip valarLip merged commit 2dc5589 into main Jun 25, 2026
27 of 29 checks passed
@valarLip valarLip deleted the zejun/add_minimax_m3_into_atom_ci_benchmark_0625 branch June 25, 2026 14:53

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR updates ATOM’s CI benchmark/accuracy configuration to introduce MiniMax-M3 (MXFP8/MXFP4) and an EAGLE3 speculative decoding variant, wiring them into the nightly/manual benchmark catalog and adding GSM8K accuracy baselines/thresholds for M3 MXFP4 (and its Eagle3 mode).

Changes:

  • Replaced MiniMax-M2.7 benchmark dispatch toggles with new m3-mxfp8 / m3-mxfp4 toggles in the benchmark workflow.
  • Updated the benchmark model catalog to add MiniMax-M3 MXFP8/MXFP4 entries, each with an EAGLE3 variant.
  • Added accuracy entries (baseline/threshold + custom lm_eval command) for MiniMax-M3 MXFP4 and MiniMax-M3 MXFP4 Eagle3.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File Description
.github/workflows/atom-benchmark.yaml Updates workflow_dispatch model toggles to select the new M3 benchmark prefixes.
.github/benchmark/models.json Replaces MiniMax-M2.7 entries with MiniMax-M3 MXFP8/MXFP4 and adds EAGLE3 benchmark variants.
.github/benchmark/models_accuracy.json Adds GSM8K accuracy baselines/thresholds and commands for M3 MXFP4 (+ Eagle3).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

{
"label": "EAGLE3",
"suffix": "-eagle3",
"extra_args": "--max-num-seqs 256 --method eagle3 --draft-model Inferact/MiniMax-M3-EAGLE3 --num-speculative-tokens 3",
{
"label": "EAGLE3",
"suffix": "-eagle3",
"extra_args": "--max-num-seqs 256 --method eagle3 --draft-model Inferact/MiniMax-M3-EAGLE3 --num-speculative-tokens 3",
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.

3 participants