liji-nv

Import pycuda.autoprimaryctx or pycuda.autoinit to init pycuda execut…

3

…ion environment Signed-off-by: Jin Li # What does this PR do ? Import pycuda.autoprimaryctx or pycuda.autoinit to init pycuda execution environment to fix "invalid device context - no currently active...

[https://nvbugs/5123103][fix] Fix torch compile for DeepSeekV3

89

# PR title Please write the PR title by following template: [JIRA ticket link/nvbug link/github issue link][fix/feat/doc/infra/...] \ For example, assume I have a PR hope to support a new...

[https://nvbugs/5667774][fix] Refine Piecewise Cuda Graph Condition for DP

10

…or DP ## Summary by CodeRabbit * **Bug Fixes** * Improved handling of context requests across multi-GPU configurations for more accurate padding calculations and CUDA graph execution decisions. ✏️ Tip:...

[TRTLLM-7073][feat] Support torch compile for PP for Llama and DeepSeekV3

22

…ekV3 ## Summary by CodeRabbit - New Features - Improved multi-GPU pipeline-parallel support with torch.compile, including backend awareness of distributed mappings for better fusions. - Models may now return (hidden_states,...