feat:[AutoDeploy] E2E build example for llama4 VLM - #9

Closed

Fridah-nv wants to merge 32 commits into

feat/ad_coverage_week1from

user/fridah/vllama4

Commits on May 18, 2025

Refine doc (#4420)
juney-nvidia
authored
ac610b3
Refine doc (#4421)
juney-nvidia
authored
58e2d6f
refine doc (#4422)
juney-nvidia
authored
ddf01f6

Commits on May 19, 2025

Remove vila test (#4376)
Tabrizian
authored
7de90a6
[TRTLLM-4618][feat] Add Nemotron Super 49B FP8 test on RTX6000 Pro (SM120) (#4363)
farazkh80
authored
791c209
tests: add qa test mentioned in docs (#4357)
crazydemo
authored
c4a0d76
[Infra] - Always push the release images in the post-merge job (#4426)
chzblych
authored
a28cf32
tests: Add test cases for rcca cases (#4347)
crazydemo
authored
58d2508
chore: cleanup perf_evaluator code (#3833)
Superjomn
authored
5b1c88d
feat: Add pp support for hybrid attn/mamba model (#4358)
yuxianq
authored
cf6cd94
fix: wrong argument name `enable_overlap_scheduler` (#4433)
kaiyux
authored
a439146
Update "Roadmap" link under README.md to the issues with Roadmap label (#4425)
AdamzNV
authored
3640fba
fix potential issues in allreduce fusion kernel and ut (#4226)

yilin-void
and
AIDC-AIB
authored
62bb7f9
[TRTLLM-4638] feat(scaffolding): update Reward Controller to PRM specific controller with step split (#4337)
dc3671
authored
e70a205
feat: NIXL interface integration (#3934)
Shixiaowei02
authored
df2798e
Downgrade the logger level for fallback tactic warning. (#4440)
hyukn
authored
98018f3
Test: Improve model re-use in C++ DGX tests for CI stability (#4263)
DomBrown
authored
c45f414
fix: temp disable the problem test (#4445)
Shixiaowei02
authored
001704c
Add llama4 disagg accuracy tests (#4336)
Tabrizian
authored
c6074c4
[https://nvbugs/5123103][fix] Fix torch compile for DeepSeekV3 (#3952)
liji-nv
authored
58e4056
[Docs] - Reapply #4220 (#4434)
chzblych
authored
942ac5c
[TRTLLM-4618][feat] Fix cutlass MoE GEMM fallback failure on FP8 + add e2e test for Mixtral 8x7B FP8 on RTX6000 Pro (SM120) (#4335)
farazkh80
authored
7656af1
[Feat] add chunked-attention kernels on Hopper (for llama4) (#4291)

PerkzZheng
and
schetlur-nv
authored
1c5b0d6
add e2e small build unit test for llama4 vlm example, update graph utils for nested subgraphs
Fridah-nv
committed
553e7f9
shrink the test model
Fridah-nv
committed
f53fb08
debug unit test failure of thread leakage
Fridah-nv
committed
5d5b3b1
change mixtral small build test to WS=0
Fridah-nv
committed
459e4bd
minor update, revert ad_build_small test
Fridah-nv
committed
13c19c0
move to integration test
Fridah-nv
committed
c715c0b
iterage graphs in reverse order
Fridah-nv
committed
7f1f847
minor
Fridah-nv
committed
963a69d
minor
Fridah-nv
committed
9c80b2d

Navigation Menu

Uh oh!

feat:[AutoDeploy] E2E build example for llama4 VLM - #9

feat:[AutoDeploy] E2E build example for llama4 VLM#9
Fridah-nv wants to merge 32 commits into
feat/ad_coverage_week1from
user/fridah/vllama4

Commits on May 18, 2025

Refine doc (#4420)

Refine doc (#4421)

refine doc (#4422)

Commits on May 19, 2025

Remove vila test (#4376)

[TRTLLM-4618][feat] Add Nemotron Super 49B FP8 test on RTX6000 Pro (SM120) (#4363)

tests: add qa test mentioned in docs (#4357)

[Infra] - Always push the release images in the post-merge job (#4426)

tests: Add test cases for rcca cases (#4347)

chore: cleanup perf_evaluator code (#3833)

feat: Add pp support for hybrid attn/mamba model (#4358)

fix: wrong argument name `enable_overlap_scheduler` (#4433)

Update "Roadmap" link under README.md to the issues with Roadmap label (#4425)

fix potential issues in allreduce fusion kernel and ut (#4226)

[TRTLLM-4638] feat(scaffolding): update Reward Controller to PRM specific controller with step split (#4337)

feat: NIXL interface integration (#3934)

Downgrade the logger level for fallback tactic warning. (#4440)

Test: Improve model re-use in C++ DGX tests for CI stability (#4263)

fix: temp disable the problem test (#4445)

Add llama4 disagg accuracy tests (#4336)

[https://nvbugs/5123103][fix] Fix torch compile for DeepSeekV3 (#3952)

[Docs] - Reapply #4220 (#4434)

[TRTLLM-4618][feat] Fix cutlass MoE GEMM fallback failure on FP8 + add e2e test for Mixtral 8x7B FP8 on RTX6000 Pro (SM120) (#4335)

[Feat] add chunked-attention kernels on Hopper (for llama4) (#4291)

add e2e small build unit test for llama4 vlm example, update graph utils for nested subgraphs

shrink the test model

debug unit test failure of thread leakage

change mixtral small build test to WS=0

minor update, revert ad_build_small test

move to integration test

iterage graphs in reverse order

minor

minor