[rollout] chore: bump up trtllm image version to 1.3.0rc10#5841

Open

Superjomn wants to merge 23 commits intoverl-project:mainfrom

Superjomn:upgrade-megatron-0-16

Commits on Apr 8, 2026

[docker] feat: upgrade Megatron-LM to core_v0.16.0 in trtllm Dockerfile

Superjomn
and
claude
committed
235B 8 nodes init config

hchings
authored and
Superjomn
committed
update config

hchings
authored and
Superjomn
committed
remove 235b fp8 script

hchings
authored and
Superjomn
committed
[docker] feat: bump TRT-LLM base image to 1.3.0rc10

Superjomn
and
claude
committed
[docker] fix: pin trl==0.27.0 to fix AutoModelForCausalLMWithValueHead import

Superjomn
and
claude
committed
[ci] bump trtllm CI image from 1.3.0rc4 to 1.3.0rc10

Superjomn
and
claude
committed
trigger CI

Superjomn
and
claude
committed
trigger CI

Superjomn
and
claude
committed
[rollout,trtllm] fix ExecutorMemoryType import for trtllm 1.3.0rc10

Superjomn
and
claude
committed
[rollout,trtllm] fix pre-commit: add blank line before try block

Superjomn
and
claude
committed
update

hchings
authored and
Superjomn
committed
[rollout,trtllm] fix pre-commit: add blank line before try block

Superjomn
and
claude
committed
trigger CI

Superjomn
and
claude
committed
[docker] fix: add CUDA forward-compat layer for trtllm rc10 image

Superjomn
and
claude
committed
[rollout,trtllm] fix: remove model_weights/draft_model_weights from fallback _WEIGHTS_TAGS

Superjomn
and
claude
committed
trigger CI

Superjomn
and
claude
committed
[docker] revert: remove unnecessary cuda-compat-13-1 layer from trtllm Dockerfile

Superjomn
and
claude
committed
[docker] fix: clear TORCH_CUDA_ARCH_LIST for FlashInfer runtime arch detection

Superjomn
and
claude
committed
[ci] switch trtllm CI image to DockerHub for faster debug iteration

Superjomn
and
claude
committed
[ci] revert: restore Volcengine registry for trtllm CI image

Superjomn
and
claude
committed
[ci] set TORCH_CUDA_ARCH_LIST to exclude SM 7.0 for FlashInfer compatibility

Superjomn
and
claude
committed
[ci] temporarily disable non-trtllm workflows for faster CI iteration

Superjomn
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[rollout] chore: bump up trtllm image version to 1.3.0rc10#5841

[rollout] chore: bump up trtllm image version to 1.3.0rc10#5841
Superjomn wants to merge 23 commits intoverl-project:mainfrom
Superjomn:upgrade-megatron-0-16

Commits on Apr 8, 2026

[docker] feat: upgrade Megatron-LM to core_v0.16.0 in trtllm Dockerfile

235B 8 nodes init config

update config

remove 235b fp8 script

[docker] feat: bump TRT-LLM base image to 1.3.0rc10

[docker] fix: pin trl==0.27.0 to fix AutoModelForCausalLMWithValueHead import

[ci] bump trtllm CI image from 1.3.0rc4 to 1.3.0rc10

trigger CI

trigger CI

[rollout,trtllm] fix ExecutorMemoryType import for trtllm 1.3.0rc10

[rollout,trtllm] fix pre-commit: add blank line before try block

update

[rollout,trtllm] fix pre-commit: add blank line before try block

trigger CI

[docker] fix: add CUDA forward-compat layer for trtllm rc10 image

[rollout,trtllm] fix: remove model_weights/draft_model_weights from fallback _WEIGHTS_TAGS

trigger CI

[docker] revert: remove unnecessary cuda-compat-13-1 layer from trtllm Dockerfile

[docker] fix: clear TORCH_CUDA_ARCH_LIST for FlashInfer runtime arch detection

[ci] switch trtllm CI image to DockerHub for faster debug iteration

[ci] revert: restore Volcengine registry for trtllm CI image

[ci] set TORCH_CUDA_ARCH_LIST to exclude SM 7.0 for FlashInfer compatibility

[ci] temporarily disable non-trtllm workflows for faster CI iteration