feat: validate ATOM and Infera on MI300X and MI355X - #2558

Open

cquil11 wants to merge 70 commits into

agent/srt-slurm-amd-integrationfrom

agent/atom-infera-validation

Commits on Aug 10, 2026

docs(amd): define srt-slurm bring-up contract
cquil11
committed
0641575
feat(amd): add MI300X srt-slurm aggregate scaffold
cquil11
committed
3a3e723
fix(amd): align MI300X recipe with vLLM 0.26
cquil11
committed
fc07a3e
fix(amd): honor MI300X Pyxis home policy
cquil11
committed
b95b4b8
fix(amd): align MI300X worker launch policy
cquil11
committed
aa681ff
feat(amd): use InferenceX fixed-sequence benchmark
cquil11
committed
1b256b9
fix(amd): exclude unsuitable MI300X nodes
cquil11
committed
16c5efc
feat(amd): add MI300X disaggregated validation target
cquil11
committed
b1aa29a
fix(amd): preserve custom benchmark arguments
cquil11
committed
181459d
Stage MI300X runtime config on compute nodes
cquil11
committed
9b728ab
Reuse stable ROCm container artifact
cquil11
committed
542ddda
fix(amd): align disagg runtime dependencies
cquil11
committed
2df8508
docs(amd): pin runtime source override head
cquil11
committed
b0376b3
fix: route MI300X orchestration over private fabric
cquil11
committed
9afc9a1
fix: discover MI300X private fabric addresses
cquil11
committed
8e139bd
feat: validate MI300X MoRI-IO routing
cquil11
committed
6f25810
fix: pin published vllm router image
cquil11
committed
5b45946
docs: update AMD runtime pin
cquil11
committed
568327b
fix(amd): use contiguous AITER cache for MoRI
cquil11
committed
1a6c28e
docs(amd): pin dynamic readiness runtime
cquil11
committed
5f0e320
docs(amd): pin CI-compatible runtime
cquil11
committed
8a59c1a
docs(amd): pin formatted runtime head
cquil11
committed
7476a92
docs: update AMD srt-slurm pin
cquil11
committed
738471e
ci(amd): validate srt-slurm disaggregation
cquil11
committed
64652fe
ci(amd): bootstrap srt-slurm compute runtime
cquil11
committed
d548a4f
ci(amd): validate node-local runtime images
cquil11
committed
8a1f8cc
ci(amd): stage srt-slurm on compute nodes
cquil11
committed
03c480d
ci(amd): validate aggregate srt-slurm serving
cquil11
committed
ba5e7c7
ci(amd): route aggregate through srt-slurm
cquil11
committed
d795641
fix(amd): select aggregate srt recipe explicitly
cquil11
committed
524b1a2
ci(amd): add MI355X srt-slurm validation
cquil11
committed
71e908a
Merge remote-tracking branch 'origin/main' into agent/srt-slurm-amd-integration
cquil11
committed
89eb32e
fix(amd): use writable MI355X shared root
cquil11
committed
33f2cfc
add production MI355X srt-slurm validation
cquil11
committed
9a7d233
fix MI355X model staging input
cquil11
committed
cc0b9f7
bump AMD runtime cache fix
cquil11
committed
1d60fb1
fix(amd): select Qwen3.5 text loader
cquil11
committed
87933a9
fix(amd): allocate full MI355X CPU topology
cquil11
committed
a55bed7
fix(amd): align Qwen3.5 Hugging Face cache root
cquil11
committed
d3ed345
fix(amd): use large-MR-capable MI355X runtime
cquil11
committed
3b3acff
fix(amd): make runtime staging reproducible
cquil11
committed
c5ce0a4
fix(amd): pin staged hub cache path
cquil11
committed
e22bf36

Commits on Aug 11, 2026

chore(amd): pin rebased srt-slurm runtime
cquil11
committed
147b2ec
fix(amd): pin native router infra correction
cquil11
committed
dd49816
Merge remote-tracking branch 'origin/main' into agent/srt-slurm-amd-integration
cquil11
committed
accf806
fix(amd): select the native SGLang Router frontend
cquil11
committed
7b3ee00
test(amd): enforce the native SGLang frontend contract
cquil11
committed
9403f0b
test: drop redundant AMD integration contract file
cquil11
committed
72ee996
feat: add ATOM and Infera validation lanes
cquil11
committed
3105070
docs: link ATOM validation pull request
cquil11
committed
226a875
fix: use writable Enroot staging runtime
cquil11
committed
dd597d5
fix: retry transient container imports
cquil11
committed
98ac36d
fix: install orchestration binaries on MI300X
cquil11
committed
c16663b
fix: stage srt-slurm infrastructure binaries
cquil11
committed
73d5e8e
fix(srt): pin Infera worker-registry readiness
cquil11
committed
e27c76f
fix(atom): validate upstream Infera KV event decoder
cquil11
committed
5021e0e
fix(mi300x): stage every eligible srt node
cquil11
committed
f307b55
fix(mi300x): cover the complete eligible node set
cquil11
committed
e5066e6
fix(mi300x): recover incomplete runtime checkouts
cquil11
committed
f6b200f
test(mi355x): add ATOM and Infera validation lanes
cquil11
committed
bb97fd7
fix(atom): keep Mooncake transfer paths reusable
cquil11
committed
85f55d6
fix(atom): validate Mooncake TCP transfer path
cquil11
committed
1a5b36f
fix(atom): verify installed framework versions
cquil11
committed
6c5311a
docs(atom): record validated TCP transport
cquil11
committed
d71afb3
Merge remote-tracking branch 'origin/main' into agent/atom-infera-validation
cquil11
committed
98b9b50
fix(atom): pin tcp-only Mooncake source overlay
cquil11
committed
2e2f375
fix: pin fetchable stable ATOM overlay
cquil11
committed
bdc0e20
fix: pin Mooncake TCP transport guard
cquil11
committed
f9bdbe1
fix(ci): restore e2e eval matrix output
cquil11
committed
eb814bb

Commits on Aug 17, 2026

fix(amd): serialize ATOM source staging
cquil11
committed
4a98aa9