ianalin123 Claude Opus 4.6 commited on
Commit
23c1b75
·
1 Parent(s): 6d522e2

fix: switch eval to A100-80GB and track completion bonus

Browse files

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Files changed (1) hide show
  1. modal_eval_v3.py +4 -2
modal_eval_v3.py CHANGED
@@ -22,7 +22,7 @@ ALL_TASKS = ["triangle", "half_fold", "quarter_fold", "letter_fold", "waterbomb_
22
 
23
  @app.function(
24
  image=image,
25
- gpu="B200",
26
  timeout=3600,
27
  volumes={OUTPUTS_DIR: volume},
28
  )
@@ -137,14 +137,16 @@ def evaluate(
137
  break
138
 
139
  final_progress = 0.0
 
140
  if obs.reward_breakdown:
141
  final_progress = obs.reward_breakdown.get("progress", 0)
 
142
 
143
  episodes.append({
144
  "total_reward": ep_reward,
145
  "steps": steps,
146
  "n_steps": len(steps),
147
- "completed": final_progress > 0.9,
148
  "final_progress": final_progress,
149
  })
150
 
 
22
 
23
  @app.function(
24
  image=image,
25
+ gpu="A100-80GB",
26
  timeout=3600,
27
  volumes={OUTPUTS_DIR: volume},
28
  )
 
137
  break
138
 
139
  final_progress = 0.0
140
+ got_completion = False
141
  if obs.reward_breakdown:
142
  final_progress = obs.reward_breakdown.get("progress", 0)
143
+ got_completion = obs.reward_breakdown.get("completion", 0) > 0
144
 
145
  episodes.append({
146
  "total_reward": ep_reward,
147
  "steps": steps,
148
  "n_steps": len(steps),
149
+ "completed": got_completion,
150
  "final_progress": final_progress,
151
  })
152