Environment Variables#
This page lists TensorRT-related environment variables referenced in the documentation. Today these variables configure the capture-and-replay shim used by Capture and Replay.
Capture-and-replay shim (TRT_SHIM_*)#
Set these variables before launching an application through the TensorRT capture shim (LD_PRELOAD or drop-in replacement). For setup instructions, refer to Capture and Replay.
Variable |
Description |
Type |
Default |
|---|---|---|---|
|
Path to save the captured |
|
|
|
Intercepted TensorRT library name. If unset, |
|
|
|
Print |
|
|
|
Print |
|
|
|
Lock every API call to enforce single-threaded execution. Ignored when |
|
|
|
Inline weights into the |
|
|
|
Skip saving weights with an element count ≥ this threshold (they will be marked as random). |
|
|
|
Flush captured calls to the file after every API call instead of aggregating them. |
|
|
|
Flush captured calls after every |
|
|
|
Path to a tactic-cache file loaded into |
|
|
See also
- Capture and Replay
How to install and run the capture shim.
- Troubleshooting
Common errors and debugging guidance.