mirror of
https://github.com/huggingface/lerobot.git
synced 2026-07-06 09:37:06 +00:00
chore(pi052): retune subtask_mem blend weights
Shift the recipe mix toward low-level execution (0.55->0.60) and away from high-level subtask (0.30->0.25); matches the blend used for the pi052 robocasa runs. Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
@@ -16,13 +16,13 @@
|
||||
blend:
|
||||
|
||||
high_level_subtask:
|
||||
weight: 0.30
|
||||
weight: 0.25
|
||||
messages:
|
||||
- {role: user, content: "${task}", stream: high_level}
|
||||
- {role: assistant, content: "${subtask}", stream: high_level, target: true, if_present: subtask}
|
||||
|
||||
low_level_execution:
|
||||
weight: 0.55
|
||||
weight: 0.60
|
||||
messages:
|
||||
# The action expert is conditioned on the SUBTASK — at inference
|
||||
# `HighLevelSubtaskFwd` generates it via the LM head and feeds it
|
||||
|
||||
Reference in New Issue
Block a user