Skip to content

[fix] Fix inference checkpoint resume to skip completed episodes #341

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jbykkk wants to merge 1 commit into
base: main
Choose a base branch
from
+12 −4
Open

[fix] Fix inference checkpoint resume to skip completed episodes #341

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 12 additions & 4 deletions internnav/habitat_extensions/vln/habitat_vln_evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -243,8 +243,9 @@ def parse_actions(self, output):

def resume_from_output_path(self) -> None:
sucs, spls, oss, nes, ndtw = [], [], [], [], []
completed_episodes = set()
if self.rank != 0:
return sucs, spls, oss, nes, ndtw
return sucs, spls, oss, nes, ndtw, completed_episodes

# resume from previous results
if os.path.exists(os.path.join(self.output_path, 'progress.json')):
Expand All @@ -257,13 +258,14 @@ def resume_from_output_path(self) -> None:
nes.append(res['ne'])
if 'ndtw' in res:
ndtw.append(res['ndtw'])
return sucs, spls, oss, nes, ndtw
completed_episodes.add((res['scene_id'], res['episode_id']))
return sucs, spls, oss, nes, ndtw, completed_episodes

def _run_eval_dual_system(self) -> tuple: # noqa: C901
self.model.eval()

# resume from previous results
sucs, spls, oss, nes, ndtw = self.resume_from_output_path()
sucs, spls, oss, nes, ndtw, completed_episodes = self.resume_from_output_path()

# Episode loop is now driven by env.reset() + env.is_running
process_bar = tqdm.tqdm(total=len(self.env.episodes), desc=f"Eval Epoch {self.epoch} Rank {self.rank}")
Expand All @@ -281,6 +283,12 @@ def _run_eval_dual_system(self) -> tuple: # noqa: C901
scene_id = episode.scene_id.split('/')[-2]
episode_id = int(episode.episode_id)
episode_instruction = episode.instruction.instruction_text

# skip already completed episodes
if (scene_id, episode_id) in completed_episodes:
process_bar.update(1)
continue

print("episode start", episode_instruction)

# save first frame per rank to validate sim quality
Expand Down Expand Up @@ -632,7 +640,7 @@ def _run_eval_system2(self) -> tuple:
self.model.eval()

# resume from previous results
sucs, spls, oss, nes, ndtw = self.resume_from_output_path()
sucs, spls, oss, nes, ndtw, _ = self.resume_from_output_path()

# Episode loop is now driven by env.reset() + env.is_running
process_bar = tqdm.tqdm(total=len(self.env.episodes), desc=f"Eval Epoch {self.epoch} Rank {self.rank}")
Expand Down