File size: 3,499 Bytes
93f12cc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 | #!/usr/bin/env python3
"""
Debug script to inspect rollout data structure.
Usage:
python scripts/debug_rollout.py <path_to_pkl_file>
"""
import sys
import json
from pathlib import Path
from verl import DataProto
def inspect_rollout(pkl_path: str):
"""Inspect the structure of a rollout PKL file."""
print(f"Loading: {pkl_path}")
data = DataProto.load_from_disk(pkl_path)
total = len(data)
print(f"\nTotal trajectories: {total}")
if total == 0:
print("No trajectories found!")
return
# Inspect first trajectory
print("\n" + "="*80)
print("INSPECTING FIRST TRAJECTORY (index 0)")
print("="*80)
item = data[0]
# Check meta_info
print("\n--- meta_info ---")
if item.meta_info:
for k, v in item.meta_info.items():
print(f" {k}: {type(v)} = {v if not isinstance(v, (list, dict)) or len(str(v)) < 100 else f'{type(v)} (length {len(v)})'}")
else:
print(" (None)")
# Check batch
print("\n--- batch ---")
if item.batch is not None:
try:
print(f" Type: {type(item.batch)}")
if hasattr(item.batch, 'keys'):
for k in item.batch.keys():
try:
v = item.batch[k]
print(f" {k}: {type(v)} shape={getattr(v, 'shape', 'N/A')}")
except Exception as e:
print(f" {k}: Error accessing - {e}")
except Exception as e:
print(f" Error inspecting batch: {e}")
else:
print(" (None)")
# Check non_tensor_batch
print("\n--- non_tensor_batch ---")
ntb = item.non_tensor_batch
if ntb:
for k, v in ntb.items():
if k == 'history':
print(f" history: list with {len(v)} entries")
if len(v) > 0:
print(f" First entry keys: {list(v[0].keys())}")
print(f" First entry: {json.dumps(v[0], indent=4, default=str)[:500]}...")
elif k == 'metrics':
print(f" metrics: {type(v)}")
if isinstance(v, dict):
for mk, mv in v.items():
print(f" {mk}: {mv}")
else:
val_str = str(v) if len(str(v)) < 100 else f"{type(v)} (length {len(v) if hasattr(v, '__len__') else 'N/A'})"
print(f" {k}: {type(v)} = {val_str}")
else:
print(" (None)")
# Inspect a few more trajectories
print("\n" + "="*80)
print("SUMMARY OF ALL TRAJECTORIES")
print("="*80)
histories_found = 0
non_empty_histories = 0
for idx in range(min(5, total)):
item = data[idx]
ntb = item.non_tensor_batch or {}
history = ntb.get('history', [])
if history is not None:
histories_found += 1
if len(history) > 0:
non_empty_histories += 1
print(f"Traj {idx}: history={len(history) if history else 0} entries")
print(f"\nOut of {total} trajectories:")
print(f" - {histories_found} have 'history' field")
print(f" - {non_empty_histories} have non-empty history")
if __name__ == "__main__":
if len(sys.argv) < 2:
print("Usage: python scripts/debug_rollout.py <path_to_pkl_file>")
sys.exit(1)
pkl_path = sys.argv[1]
if not Path(pkl_path).exists():
print(f"Error: File not found: {pkl_path}")
sys.exit(1)
inspect_rollout(pkl_path)
|