admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-05-23 12:40:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix traacking

Browse Source
This commit is contained in:
Eugene Mironov
2025-11-06 02:30:53 +07:00
parent 55c4cc1b27
commit b27570039c
4 changed files with 11 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/rtc/eval_dataset.py
+8 -1
View File
@@ -68,7 +68,7 @@ class RTCEvalConfig(HubMixin):
default_factory=lambda: RTCConfig( default_factory=lambda: RTCConfig(
enabled=True, enabled=True,
execution_horizon=20, execution_horizon=20,
max_guidance_weight=5.0, max_guidance_weight=10.0,
prefix_attention_schedule=RTCAttentionSchedule.EXP, prefix_attention_schedule=RTCAttentionSchedule.EXP,
debug=True, debug=True,
debug_maxlen=1000, debug_maxlen=1000,
@@ -184,6 +184,10 @@ class RTCEvaluator:
preprocessed_first_sample, preprocessed_first_sample,
)[:, :25, :].squeeze(0) )[:, :25, :].squeeze(0)
self.policy.rtc_processor.reset_tracker()
logging.info("Resetting tracker")
# Sample noise (use same noise for both RTC and non-RTC for fair comparison) # Sample noise (use same noise for both RTC and non-RTC for fair comparison)
noise_size = (1, self.policy.config.chunk_size, self.policy.config.max_action_dim) noise_size = (1, self.policy.config.chunk_size, self.policy.config.max_action_dim)
noise = self.policy.model.sample_noise(noise_size, self.device) noise = self.policy.model.sample_noise(noise_size, self.device)
@@ -300,6 +304,9 @@ class RTCEvaluator:
num_steps: Total number of denoising steps for colormap num_steps: Total number of denoising steps for colormap
""" """
logging.info("=" * 80)
logging.info(f"Plotting {len(tracked_steps)} steps")
debug_steps = tracked_steps debug_steps = tracked_steps
if not debug_steps: if not debug_steps:
return return
src/lerobot/policies/rtc/configuration_rtc.py
+1 -1
View File
@@ -40,7 +40,7 @@ class RTCConfig:
# Core RTC settings # Core RTC settings
# Todo change to exp # Todo change to exp
prefix_attention_schedule: RTCAttentionSchedule = RTCAttentionSchedule.LINEAR prefix_attention_schedule: RTCAttentionSchedule = RTCAttentionSchedule.LINEAR
max_guidance_weight: float = 5.0 max_guidance_weight: float = 10.0
execution_horizon: int = 10 execution_horizon: int = 10
# Debug settings # Debug settings
src/lerobot/policies/rtc/modeling_rtc.py
-1
View File
@@ -260,7 +260,6 @@ class RTCProcessor:
x1_t = x1_t.squeeze(0) x1_t = x1_t.squeeze(0)
err = err.squeeze(0) err = err.squeeze(0)
# Record debug information (all params except x_t which is recorded externally)
self.track( self.track(
time=time, time=time,
x1_t=x1_t, x1_t=x1_t,
src/lerobot/policies/smolvla/modeling_smolvla.py
+2 -1
View File
@@ -798,12 +798,13 @@ class VLAFlowMatching(nn.Module):
# Euler step # Euler step
x_t += dt * v_t x_t += dt * v_t
time += dt
# Record x_t and v_t after Euler step (other params are recorded in rtc_processor.denoise_step) # Record x_t and v_t after Euler step (other params are recorded in rtc_processor.denoise_step)
if self.rtc_processor is not None and self.rtc_processor.is_debug_enabled(): if self.rtc_processor is not None and self.rtc_processor.is_debug_enabled():
self.rtc_processor.track(time=time, x_t=x_t, v_t=v_t) self.rtc_processor.track(time=time, x_t=x_t, v_t=v_t)
time += dt
return x_t return x_t
def denoise_step( def denoise_step(