[AMD][MI355X] DSv4 SGLang agentic: v0.5.15 image, router, HiCache + MTP #2199

Closed

seungrokj wants to merge 16 commits into

mainfrom

amd/agentx_dsv4_sgl_mtp

Commits on Jul 14, 2026

[AMD][MI355X] Update DSv4 SGLang agentic launcher to v0.5.14 image with router and HiCache support

seungrokj
and
claude
committed
[AMD][MI355X] Add dram-utilization 0.80 to DSv4 SGLang agentic sweep config

seungrokj
and
claude
committed
Update perf-changelog with PR link for DSv4 SGLang agentic entry

seungrokj
and
claude
committed
[AMD][MI355X] Remove --context-length and comment out extra sweep configs for initial validation

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang agentic: use compressed attention, 2x max-running-requests, re-enable all sweep configs

seungrokj
and
claude
committed
[AMD][MI355X] Fix cudaHostRegister OOM: lower hicache-ratio to 4, adjust sweep concurrency

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang agentic: focus sweep on DPA configs, scale concurrency to 96

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang agentic: add non-DPA baseline config (tp=8, conc 1-8)

seungrokj
and
claude
committed
[AMD][MI355X] exclude mia1-p01-g12 from agentic salloc

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang: narrow DPA sweep ranges, add disagg agentic hicache recipe

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang: refactor launcher, rename config key, bump image

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang: fix kv-offload-backend to map syntax, add mori p2p, revert image

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang: remove disagg agentic hicache recipe

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang: add vllm-router and mooncake backend to agentic sweep

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang: bump agentic-hicache image to v0.5.15

seungrokj
and
claude
committed
[AMD][MI355X] DSv4 SGLang: update perf-changelog for image bump and drop disagg entry

seungrokj
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[AMD][MI355X] DSv4 SGLang agentic: v0.5.15 image, router, HiCache + MTP #2199

[AMD][MI355X] DSv4 SGLang agentic: v0.5.15 image, router, HiCache + MTP #2199
seungrokj wants to merge 16 commits into
mainfrom
amd/agentx_dsv4_sgl_mtp

Commits on Jul 14, 2026

[AMD][MI355X] Update DSv4 SGLang agentic launcher to v0.5.14 image with router and HiCache support

[AMD][MI355X] Add dram-utilization 0.80 to DSv4 SGLang agentic sweep config

Update perf-changelog with PR link for DSv4 SGLang agentic entry

[AMD][MI355X] Remove --context-length and comment out extra sweep configs for initial validation

[AMD][MI355X] DSv4 SGLang agentic: use compressed attention, 2x max-running-requests, re-enable all sweep configs

[AMD][MI355X] Fix cudaHostRegister OOM: lower hicache-ratio to 4, adjust sweep concurrency

[AMD][MI355X] DSv4 SGLang agentic: focus sweep on DPA configs, scale concurrency to 96

[AMD][MI355X] DSv4 SGLang agentic: add non-DPA baseline config (tp=8, conc 1-8)

[AMD][MI355X] exclude mia1-p01-g12 from agentic salloc

[AMD][MI355X] DSv4 SGLang: narrow DPA sweep ranges, add disagg agentic hicache recipe

[AMD][MI355X] DSv4 SGLang: refactor launcher, rename config key, bump image

[AMD][MI355X] DSv4 SGLang: fix kv-offload-backend to map syntax, add mori p2p, revert image

[AMD][MI355X] DSv4 SGLang: remove disagg agentic hicache recipe

[AMD][MI355X] DSv4 SGLang: add vllm-router and mooncake backend to agentic sweep

[AMD][MI355X] DSv4 SGLang: bump agentic-hicache image to v0.5.15

[AMD][MI355X] DSv4 SGLang: update perf-changelog for image bump and drop disagg entry