From 18bba97cd69e1425ce6b281ace3879a8ef175aad Mon Sep 17 00:00:00 2001 From: Jade Choghari Date: Sun, 25 Jan 2026 19:48:54 +0100 Subject: [PATCH] change default to 256 latent video --- src/lerobot/policies/videovla/configuration_pi05.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/lerobot/policies/videovla/configuration_pi05.py b/src/lerobot/policies/videovla/configuration_pi05.py index 541db40e6..fa969e08c 100644 --- a/src/lerobot/policies/videovla/configuration_pi05.py +++ b/src/lerobot/policies/videovla/configuration_pi05.py @@ -47,7 +47,7 @@ class PI05VideoConfig(PreTrainedConfig): # Which camera to use for video encoding (None = first camera, or specify key like "observation.images.top") video_encoder_camera_key: str | None = None # Perceiver Resampler settings to reduce video tokens (4096 -> video_num_latents) - video_num_latents: int = 128 # Number of latent tokens for video resampler + video_num_latents: int = 256 # Number of latent tokens for video resampler video_resampler_num_heads: int = 8 # Number of attention heads in resampler # Shorter state and action vectors will be padded to these dimensions