Skip to content

[AMD/ROCM] Update gptoss-fp4-mi355x-atom config #1195

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
seungrokj merged 6 commits into from
May 2, 2026
Merged

[AMD/ROCM] Update gptoss-fp4-mi355x-atom config #1195

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
867de5c
Update gptoss-fp4-mi355x-atom config and add perf-changelog entry
seungrokj Apr 27, 2026
ea2f92b
Fill in perf-changelog description for gptoss-fp4-mi355x-atom
seungrokj Apr 27, 2026
54daae0
Merge branch 'main' into srok/atom_gptoss_fp4
chunfangamd Apr 30, 2026
636b505
Merge branch 'main' into srok/atom_gptoss_fp4
seungrokj May 1, 2026
b1852b8
Merge branch 'main' into srok/atom_gptoss_fp4
chunfangamd May 1, 2026
bdd39c7
Merge branch 'main' into srok/atom_gptoss_fp4
seungrokj May 2, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions .github/configs/amd-master.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -688,7 +688,7 @@ gptoss-fp4-mi355x-vllm:
- { tp: 8, conc-start: 4, conc-end: 8 }

gptoss-fp4-mi355x-atom:
image: rocm/atom:rocm7.1.1-ubuntu24.04-pytorch2.9-atom0.1.1-MI350x
image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post
model: openai/gpt-oss-120b
model-prefix: gptoss
runner: mi355x
Expand All @@ -699,12 +699,12 @@ gptoss-fp4-mi355x-atom:
- isl: 1024
osl: 1024
search-space:
- { tp: 1, conc-start: 16, conc-end: 128 }
- { tp: 1, conc-start: 16, conc-end: 256 }
- { tp: 8, ep: 1, conc-start: 4, conc-end: 32 }
- isl: 8192
osl: 1024
search-space:
- { tp: 1, conc-start: 4, conc-end: 128 }
- { tp: 1, conc-start: 4, conc-end: 256 }
- { tp: 8, ep: 1, conc-start: 4, conc-end: 16 }

dsr1-fp8-mi355x-atom:
Expand Down
2 changes: 2 additions & 0 deletions benchmarks/single_node/gptoss_fp4_mi355x_atom.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,7 @@ fi

# Start GPU monitoring (power, temperature, clocks every second)
start_gpu_monitor
MEM_FRAC_STATIC=0.9

set -x

Expand All @@ -54,6 +55,7 @@ python3 -m atom.entrypoints.openai_server \
--server-port $PORT \
-tp $TP \
--kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN $EP \
--gpu-memory-utilization $MEM_FRAC_STATIC \
--block-size $BLOCK_SIZE > $SERVER_LOG 2>&1 &

SERVER_PID=$!
Expand Down
6 changes: 6 additions & 0 deletions perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1892,3 +1892,9 @@
description:
- "Pass --dsv4 to run_benchmark_serving so MTP benchmarks use the DSv4 chat template (PR #1153)"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1182

- config-keys:
- gptoss-fp4-mi355x-atom
description:
- ""
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/

Check failure on line 1900 in perf-changelog.yaml

View check run for this annotation

Claude / Claude Code Review

perf-changelog entry has empty description and incomplete pr-link 

The new perf-changelog.yaml entry for `gptoss-fp4-mi355x-atom` (lines 1895-1900) is a forgotten template: `description` is an empty string `- ""` and `pr-link` is `.../pull/` with no PR number. As written, the entry documents nothing and links nowhere — please fill in a real description (atom image bump rocm7.1.1/atom0.1.1 → rocm7.2.2/atom0.1.2.post, conc-end 128 → 256 for tp1 configs, and the new `--gpu-memory-utilization 0.9` flag) and set `pr-link` to `https://github.com/SemiAnalysisAI/Infere
Comment thread
seungrokj marked this conversation as resolved.
Outdated
Loading