[DNM][AMD] agentx-v0.4 rebased from commit chore/agentx-v0.4 commit 823189e#1709

Closed

seungrokj wants to merge 147 commits into

chore/agentx-v0.4from

amd/agentx-v0.4_rebase0611

Commits on Jun 25, 2026

[AMD] agentic: add hicache/lmcache configs, update agentic scripts for mi355x models

seungrokj
and
claude
committed
Add GLM5.1 & Qwen3.5 MI300 Agentic Scripts

ajith-sirra-amd
authored and
seungrokj
committed
[AMD] add DSV4-FP4-MI355x atom agentic benchmark and master yaml config

seungrokj
and
claude
committed
[AMD] update DSV4-FP4-MI355x atom agentic benchmark and master yaml config

seungrokj
and
claude
committed
Add DSV4 MI355X Agentic Scripts

ajith-sirra-amd
authored and
seungrokj
committed
[AMD] dsv4_fp4_mi355x_atom.sh: update agentic benchmark script

seungrokj
and
claude
committed
Add DSV4 MI355X Agentic Scripts

ajith-sirra-amd
authored and
seungrokj
committed
[AMD] update DSV4-FP4-MI355X SGLang agentic benchmark and master yaml config

seungrokj
and
claude
committed
[AMD] update DSV4-FP4-MI355X SGLang agentic/fixed-seq-len benchmark scripts and master yaml

seungrokj
and
claude
committed
[AMD] remove unused CACHE_ARGS from dsv4_fp4_mi355x_sglang agentic script

seungrokj
and
claude
committed
[AMD] tune hicache ratio and disable none-offloading in agentic config

seungrokj
and
claude
committed
[AMD] remove --disable-radix-cache from dsv4_fp4_mi355x_sglang agentic script

seungrokj
and
claude
committed
[AMD] add dsv4 sglang disagg

billishyahao
authored and
seungrokj
committed
recipe for agentic dsr1 fp4 and dsv4 fp4

authored and
seungrokj
committed
fix image

billishyahao
authored and
seungrokj
committed
fix the image

billishyahao
authored and
seungrokj
committed
fix

billishyahao
authored and
seungrokj
committed
fix

billishyahao
authored and
seungrokj
committed
bump image

billishyahao
authored and
seungrokj
committed
remove numactl

ichbinblau
authored and
seungrokj
committed
update dsv4 recipe

ichbinblau
authored and
seungrokj
committed
update dsv4 agentic config and benchmark script

seungrokj
and
claude
committed
bump sglang image for dsv4-fp4-mi355x-agentic-hicache

seungrokj
and
claude
committed
revert sglang image, comment out blk size 1 variant

seungrokj
and
claude
committed
switch to sglang v0.5.13 image with page-size 1

seungrokj
and
claude
committed
test high con only in SA

ichbinblau
authored and
seungrokj
committed
revert to v0.5.12 image with page-size 256

seungrokj
and
claude
committed
update dsr1 agentic con=1

ichbinblau
authored and
seungrokj
committed
add hicache L3 support

ichbinblau
authored and
seungrokj
committed
dump commands in log files

ichbinblau
authored and
seungrokj
committed
bump to latest image and fix

ichbinblau
authored and
seungrokj
committed
fix

ichbinblau
authored and
seungrokj
committed
Adding MINIMAX-M3 MI355X MXFP8 VLLM Agentic support

ajith-sirra-amd
authored and
seungrokj
committed
dsv4: fix prefill deadlock, kernel chunk cap, and router circuit-breaker for agentic disagg

authored and
seungrokj
committed
dsv4: move prefill/memory-pool patches into setup_deps and tune RDMA/HiCache

authored and
seungrokj
committed
agentic: add 061326/061526 weka trace datasets and switch default to 061526

authored and
seungrokj
committed
add more con list to dsv4 disagg agentic

ichbinblau
authored and
seungrokj
committed
dsv4: bump sglang-rocm image to mi35x-20260618 for disagg-agentic-hicache

authored and
seungrokj
committed
aiperf: bump submodule to 313c717

authored and
seungrokj
committed
run higher con only

ichbinblau
authored and
seungrokj
committed
fix runner bug

ichbinblau
authored and
seungrokj
committed
lower the max running req

ichbinblau
authored and
seungrokj
committed
Honor SCENARIO_SUBDIR so agentic-codin recipe is picked

ichbinblau
authored and
seungrokj
committed
enable offloading:none

ichbinblau
authored and
seungrokj
committed
relax the router timeout

ichbinblau
authored and
seungrokj
committed
add a patch to solve decode hang issue

ichbinblau
authored and
seungrokj
committed
[AMD] add dsv4_fp4_mi355x_vllm agentic script and update kimik2.5/amd-master

seungrokj
and
claude
committed
[AMD] fix dsv4_fp4_mi355x_vllm agentic script

seungrokj
and
claude
committed
[AMD] update amd-master.yaml

seungrokj
and
claude
committed
[AMD] fix missing total-cpu-dram-gb on hicache/lmcache agentic entries

seungrokj
and
claude
committed
[AMD] update amd-master.yaml

seungrokj
and
claude
committed
[AMD] set max-model-len 180000 for dsv4_fp4_mi355x_vllm agentic

seungrokj
and
claude
committed
[AMD] remove VLLM_PREFIX_CACHE_RETENTION_INTERVAL from dsv4_fp4_mi355x_vllm agentic

seungrokj
and
claude
committed
agentic: flatten trace replay output and stage artifacts for CI upload

ichbinblau
and
cursoragent
committed
[AMD] reduce gpu-memory-utilization and max-model-len for dsv4_fp4_mi355x_vllm agentic

seungrokj
and
claude
committed
[AMD] lower gpu-memory-utilization to 0.35 for lmcache offloading path

seungrokj
and
claude
committed
Revert "[AMD] lower gpu-memory-utilization to 0.35 for lmcache offloading path"
seungrokj
committed
[AMD] fix lmcache hybrid kv cache manager flag and bump aiperf submodule

seungrokj
and
claude
committed
[AMD] restore --disable-hybrid-kv-cache-manager for lmcache path

seungrokj
and
claude
committed
[AMD] reduce max-model-len to 100000 for dsv4_fp4_mi355x_vllm agentic

seungrokj
and
claude
committed
[AMD] add no-offloading baseline and expand conc-list for dsv4-fp4-mi355x-vllm-agentic-lmcache

seungrokj
and
claude
committed
[AMD] enable fp4 indexer cache and hybrid KV cache manager for mi355x vllm agentic

seungrokj
and
claude
committed
[AMD] remove --attention_config.use_fp4_indexer_cache from dsv4_fp4_mi355x_vllm (not yet)

seungrokj
and
claude
committed
[AMD] narrow dsv4-fp4-mi355x-vllm-agentic-lmcache conc-list to [16]

seungrokj
and
claude
committed
[AMD] swap dsv4-fp4-mi355x-vllm-agentic-lmcache search-space to lmcache offloading

seungrokj
and
claude
committed

Commits on Jun 26, 2026

Commits on Jun 27, 2026

Commits on Jun 28, 2026

Commits on Jun 29, 2026

Commits on Jun 30, 2026

[AMD] update dsv4-fp4-mi355x-vllm-agentic: merge lmcache config, set warmup to 1200s, switch to lmcache conc42

seungrokj
and
claude
committed