Skip to content

Add support for TensorRT Quickly Deployable Plugins #578

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Apr 17, 2025

Conversation

samurdhikaru
Copy link
Collaborator

  • Overloads tp.plugin to allow the addition of TRT QDPs
  • Relaxes the check for the defined plugin field length vs provided field length match
    • Many legacy plugins and QDPs are defined with variable length plugin fields where getFieldNames() of the plugin creator could report zero field length. So the check is enforced only when defined field length is positive.
  • Adds two integration tests for QDPs

@samurdhikaru samurdhikaru force-pushed the dev/add-trt-qdp-support branch 6 times, most recently from ec66b51 to 9d27c30 Compare March 17, 2025 23:38
@samurdhikaru samurdhikaru force-pushed the dev/add-trt-qdp-support branch from 9d27c30 to ec59f03 Compare March 25, 2025 19:28
@pranavm-nvidia pranavm-nvidia force-pushed the dev/add-trt-qdp-support branch 2 times, most recently from 2926157 to 12da90a Compare April 9, 2025 20:23
@samurdhikaru samurdhikaru force-pushed the dev/add-trt-qdp-support branch from 12da90a to 3138e9c Compare April 11, 2025 23:18
samurdhikaru and others added 4 commits April 17, 2025 15:06
- Overloads `tp.plugin` to allow the addition of TRT QDPs
- Relaxes the check for the defined plugin field length vs provided field length match
   - Many legacy plugins and QDPs are defined with variable length plugin fields where `getFieldNames()`
     of the plugin creator could report zero field length. So the check is enforced only when defined
     field length is positive.
- Adds two integration tests for QDPs
`triton` requires source code inspection in order to compile the kernel.
Since the generated docs dynamically execute code blocks, this is normally
not possible. This commit adds a special `ExecNamespace` that will inject
source code information as functions are defined.

This commit also adds various tests for QDPs and refactors the guide
and some of the code.
Co-authored-by: Samurdhi Karunaratne <[email protected]>
Signed-off-by: pranavm-nvidia <[email protected]>
@pranavm-nvidia pranavm-nvidia force-pushed the dev/add-trt-qdp-support branch from 598cfd7 to fbd3c73 Compare April 17, 2025 22:06
@pranavm-nvidia pranavm-nvidia merged commit 899052b into main Apr 17, 2025
1 check passed
@pranavm-nvidia pranavm-nvidia deleted the dev/add-trt-qdp-support branch April 17, 2025 22:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants