[AMD][MI35X]Update qwen3.5 perf by zhentaocc · Pull Request #1036 · SemiAnalysisAI/InferenceX

feat: enhance Qwen benchmark scripts with additional parameters

Chen, Todd

committed

Update perf-changelog.yaml to include new Qwen3.5 FP8 and BF16 SGLang benchmark configurations for MI355X, enhancing performance with updated CLI arguments.

Chen, Todd

committed

Update SGLang image versions for Qwen3.5 configurations in amd-master.yaml to v0.5.9, ensuring compatibility with recent changes.

Chen, Todd

committed

use 0327 build

Chen, Todd

committed

Update perf-changelog.yaml to reflect the new PR link for Qwen3.5 FP8 and BF16 SGLang benchmarks on MI355X, ensuring accurate tracking of performance enhancements.

Chen, Todd

committed

Update Qwen3.5 image tags in amd-master.yaml to v0.5.10rc0 for MI355X configurations and adjust perf-changelog.yaml to reflect the changes, ensuring accurate performance tracking and compatibility.

Chen, Todd

committed

Update Qwen3.5 FP8 and BF16 SGLang benchmark descriptions in perf-changelog.yaml to reflect improved CLI arguments for MI355X, ensuring better performance tracking.

Chen, Todd

committed

Enhance Qwen3.5 benchmark scripts for MI355X by adding EP_SIZE parameter and adjusting memory fraction. Updated launch_server command to include data-parallel-size and improved context length handl…

Chen, Todd

committed

Update search-space configurations in amd-master.yaml for Qwen3.5 benchmarks, increasing conc-end values and adding new entries for improved performance tuning on MI355X and MI300X.

Chen, Todd

committed

Remove context length parameter from Qwen3.5 BF16 and FP8 benchmark scripts for MI355X to streamline configuration and improve performance.

Chen, Todd

committed

update to 5.10 rocm for qwen35

zhentaocc

authored and

Chen, Todd

committed

Update Qwen3.5 benchmark configurations in amd-master.yaml to include EP_SIZE parameter for search-space entries, enhancing performance tuning for MI355X and MI300X. Adjusted perf-changelog.yaml to…

Chen, Todd

committed

Update context length calculations in Qwen3.5 benchmark scripts for BF16 and FP8 to improve performance tuning. Adjusted search-space configurations in amd-master.yaml to increase conc-end values f…

Chen, Todd

committed

Update image tags in amd-master.yaml for Qwen3.5 benchmarks to v0.5.10rc0-rocm700 for MI355X and MI300X configurations, ensuring compatibility and improved performance tracking.

Chen, Todd

committed

Update image tags in amd-master.yaml for Qwen3.5 benchmarks, changing MI355X to v0.5.10rc0-rocm700 and MI300X to v0.5.9-rocm720, ensuring compatibility and consistency across configurations.

Chen, Todd

committed

Remove data-parallel-size parameter and increase mem-fraction-static from 0.75 to 0.8 in Qwen3.5 BF16 and FP8 benchmark scripts to enhance performance tuning.

Chen, Todd

committed

Update sglang image for qwen3.5 mi355x configs to fix shared memory crash

chunfangamd

authored and

Chen, Todd

committed

Refine search-space configurations in amd-master.yaml for Qwen3.5 benchmarks, adjusting parameters to optimize performance for MI355X. Update perf-changelog.yaml to remove an outdated entry.

Chen, Todd

committed

Update image tags in amd-master.yaml and perf-changelog.yaml for Qwen3.5 benchmarks, replacing outdated sglang image references with the latest version to ensure consistency and improved performance.

Chen, Todd

committed

Remove aiter allreduce fusion option from Qwen3.5 FP8 MI355X benchmark script to simplify configuration and enhance performance tuning.

Chen, Todd

committed

optimize the search space

chunfangamd

authored and

Chen, Todd

committed

Upgrade image to 20260413

chunfangamd

authored and

Chen, Todd

committed

Update sglang image tags for Qwen3.5 benchmarks in amd-master.yaml to the latest version (20260414) for improved consistency and performance.

Chen, Todd

committed

Remove redundant search-space entry for qwen3.5-bf16-mi355x-sglang in amd-master.yaml to streamline configuration.

Chen, Todd

committed

Remove duplicate search-space entry for qwen3.5-bf16-mi355x-sglang in amd-master.yaml to enhance configuration clarity.

Chen, Todd

committed

Refine search-space configurations for qwen3.5-fp8-mi355x-sglang in amd-master.yaml by adjusting parameters to improve performance tuning.

Chen, Todd

committed

Update sglang image tags for qwen3.5 configurations in amd-master.yaml to the latest version (20260415) for improved consistency and performance.

Chen, Todd

committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[AMD][MI35X]Update qwen3.5 perf#1036

[AMD][MI35X]Update qwen3.5 perf#1036
chunfangamd merged 33 commits intomainfrom
todd/qwen35-mi35x

Commits on Apr 15, 2026

feat: enhance Qwen benchmark scripts with additional parameters

Update perf-changelog.yaml to include new Qwen3.5 FP8 and BF16 SGLang benchmark configurations for MI355X, enhancing performance with updated CLI arguments.

Update SGLang image versions for Qwen3.5 configurations in amd-master.yaml to v0.5.9, ensuring compatibility with recent changes.

use 0327 build

Update perf-changelog.yaml to reflect the new PR link for Qwen3.5 FP8 and BF16 SGLang benchmarks on MI355X, ensuring accurate tracking of performance enhancements.

Update Qwen3.5 image tags in amd-master.yaml to v0.5.10rc0 for MI355X configurations and adjust perf-changelog.yaml to reflect the changes, ensuring accurate performance tracking and compatibility.

Update Qwen3.5 FP8 and BF16 SGLang benchmark descriptions in perf-changelog.yaml to reflect improved CLI arguments for MI355X, ensuring better performance tracking.

Enhance Qwen3.5 benchmark scripts for MI355X by adding EP_SIZE parameter and adjusting memory fraction. Updated launch_server command to include data-parallel-size and improved context length handl…

Update search-space configurations in amd-master.yaml for Qwen3.5 benchmarks, increasing conc-end values and adding new entries for improved performance tuning on MI355X and MI300X.

Remove context length parameter from Qwen3.5 BF16 and FP8 benchmark scripts for MI355X to streamline configuration and improve performance.

update to 5.10 rocm for qwen35

Update Qwen3.5 benchmark configurations in amd-master.yaml to include EP_SIZE parameter for search-space entries, enhancing performance tuning for MI355X and MI300X. Adjusted perf-changelog.yaml to…

Update context length calculations in Qwen3.5 benchmark scripts for BF16 and FP8 to improve performance tuning. Adjusted search-space configurations in amd-master.yaml to increase conc-end values f…

Update image tags in amd-master.yaml for Qwen3.5 benchmarks to v0.5.10rc0-rocm700 for MI355X and MI300X configurations, ensuring compatibility and improved performance tracking.

Update image tags in amd-master.yaml for Qwen3.5 benchmarks, changing MI355X to v0.5.10rc0-rocm700 and MI300X to v0.5.9-rocm720, ensuring compatibility and consistency across configurations.

Remove data-parallel-size parameter and increase mem-fraction-static from 0.75 to 0.8 in Qwen3.5 BF16 and FP8 benchmark scripts to enhance performance tuning.

Update sglang image for qwen3.5 mi355x configs to fix shared memory crash

Refine search-space configurations in amd-master.yaml for Qwen3.5 benchmarks, adjusting parameters to optimize performance for MI355X. Update perf-changelog.yaml to remove an outdated entry.

Update image tags in amd-master.yaml and perf-changelog.yaml for Qwen3.5 benchmarks, replacing outdated sglang image references with the latest version to ensure consistency and improved performance.

Remove aiter allreduce fusion option from Qwen3.5 FP8 MI355X benchmark script to simplify configuration and enhance performance tuning.

optimize the search space

Upgrade image to 20260413

Update sglang image tags for Qwen3.5 benchmarks in amd-master.yaml to the latest version (20260414) for improved consistency and performance.

Remove redundant search-space entry for qwen3.5-bf16-mi355x-sglang in amd-master.yaml to streamline configuration.

Remove duplicate search-space entry for qwen3.5-bf16-mi355x-sglang in amd-master.yaml to enhance configuration clarity.

Refine search-space configurations for qwen3.5-fp8-mi355x-sglang in amd-master.yaml by adjusting parameters to improve performance tuning.

Update sglang image tags for qwen3.5 configurations in amd-master.yaml to the latest version (20260415) for improved consistency and performance.

Commits on Apr 16, 2026

Update sglang image tag for Qwen3.5 FP8 and BF16 benchmarks in perf-changelog.yaml to the latest version (v0.5.10rc0-rocm720-mi35x-20260415) for enhanced performance.

Add aiter allreduce fusion option to Qwen3.5 FP8 MI355X benchmark script for enhanced performance tuning.

Adjust CONTEXT_LENGTH in Qwen3.5 benchmark scripts for BF16 and FP8 configurations to improve performance tuning.

Remove duplicate entry for qwen3.5-fp8-mi355x-sglang in perf-changelog.yaml to enhance configuration clarity and maintain consistency.

Add upstream SGLang PR links to perf-changelog.yaml for qwen3.5 MI355X entry

Downgrade the image for Qwen3.5-FP8-MI355X-SGLang to 20260414