avanigupta Claude Opus 4.6 (1M context) commited on
Commit
a1f98bf
·
1 Parent(s): 8560706

remove ambiguous LR fix — identify-only, any valid LR works

Browse files

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

dataqa_env/server/gradio_ui.py CHANGED
@@ -113,11 +113,11 @@ AGENT_TRAJECTORIES = {
113
  "row:12,col:test_accuracy,issue:statistical_outlier",
114
  ],
115
  "fixes": [
116
- # All deterministic:
117
  "row:9,col:batch_size,fix:256", # 250 → nearest power of 2
118
  "row:14,col:training_time_hours,fix:72.0", # -72.0 → remove negative sign
119
  "row:15,col:model_name,fix:whisper-small", # "whsiper-small" → fix spelling
120
- "row:13,col:learning_rate,fix:0.000025", # 2.5 likely 2.5e-5
121
  ],
122
  },
123
  ],
 
113
  "row:12,col:test_accuracy,issue:statistical_outlier",
114
  ],
115
  "fixes": [
116
+ # Only deterministic fixes:
117
  "row:9,col:batch_size,fix:256", # 250 → nearest power of 2
118
  "row:14,col:training_time_hours,fix:72.0", # -72.0 → remove negative sign
119
  "row:15,col:model_name,fix:whisper-small", # "whsiper-small" → fix spelling
120
+ # NOT proposed: row:13 LR (2.5 is out of range but any valid LR works)
121
  ],
122
  },
123
  ],
dataqa_env/server/tasks.py CHANGED
@@ -432,11 +432,11 @@ EXP-030,llama2-13b,oasst1,84437,4401,4401,0.00001,2,3,0.78,0.88,0.0,52.0,12.0,20
432
  description="Negative training time -72.0 — likely sign typo (should be 72.0)",
433
  difficulty=1.0))
434
 
435
- # Issue 7: Learning rate in wrong notation (deterministic: "2.5e1" intended as "2.5e-5" "0.000025")
436
  r = 12 # EXP-013
437
- data[r][6] = "2.5" # clearly missing the "e-5" part
438
  issues.append(PlantedIssue(row=r + 1, col="learning_rate", issue_type="out_of_range",
439
- description="Learning rate 2.5 exceeds maximum 1.0 — likely truncated scientific notation (e.g. 2.5e-5 → 0.000025)",
440
  difficulty=1.5))
441
 
442
  # Issue 8: Model name misspelling (deterministic: "whsiper-small" → "whisper-small")
 
432
  description="Negative training time -72.0 — likely sign typo (should be 72.0)",
433
  difficulty=1.0))
434
 
435
+ # Issue 7: Learning rate out of range (identify-only any valid LR would work)
436
  r = 12 # EXP-013
437
+ data[r][6] = "2.5" # exceeds max 1.0
438
  issues.append(PlantedIssue(row=r + 1, col="learning_rate", issue_type="out_of_range",
439
+ description="Learning rate 2.5 exceeds maximum of 1.0",
440
  difficulty=1.5))
441
 
442
  # Issue 8: Model name misspelling (deterministic: "whsiper-small" → "whisper-small")