admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-16 14:32:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs(pi052): describe merged training optimizations

Browse Source
This commit is contained in:
Pepijn
2026-07-15 15:07:01 +02:00
parent 5242e9195c
commit 2e43ca0d54
1 changed files with 5 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/pi052.mdx
+5 -6
View File
@@ -141,8 +141,7 @@ pass.
## Training performance ## Training performance
Pi052 includes the optimized training paths from Pi052 uses optimized training paths by default:
[PR #3974](https://github.com/huggingface/lerobot/pull/3974). The default path:
- batches repeated flow targets and suffix projections instead of replaying - batches repeated flow targets and suffix projections instead of replaying
small operations in Python; small operations in Python;
@@ -174,10 +173,10 @@ with Flex/manual attention and compiled vision disabled.
### Reported training benchmarks ### Reported training benchmarks
PR #3974 measured complete optimizer steps with three real camera inputs, BF16 These benchmarks measure complete optimizer steps with three real camera
transformer/action execution, FP32 vision, fused AdamW, and no video decoding or inputs, BF16 transformer/action execution, FP32 vision, fused AdamW, and no
network I/O. Results vary with GPU, batch shape, annotation mixture, and video decoding or network I/O. Results vary with GPU, batch shape, annotation
checkpointing: mixture, and checkpointing:
| Workload | RTX PRO 6000 Blackwell | A100 80 GB | | Workload | RTX PRO 6000 Blackwell | A100 80 GB |
| -------------------------- | -------------------------: | -------------------------: | | -------------------------- | -------------------------: | -------------------------: |