File size: 19,856 Bytes
e6fad38 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 | #!/usr/bin/env python3
"""
Aggregate metrics from checkpoint JSONs into a single table.
Saves results locally to the 'table_result' directory.
"""
import os
import json
import argparse
import csv
import numpy as np
from typing import Dict, Any, List, Tuple, Set, Union
import pandas as pd
from openpyxl import load_workbook
from openpyxl.styles import PatternFill, Font, Alignment
# Attempt to import the specific helper; if not found, we define a dummy placeholder
# to ensure the script doesn't crash if the user only has this single file.
try:
from evaluate_aime_raw_vs_finetuned import find_best_checkpoint
except ImportError:
def find_best_checkpoint(path):
return None, 0
Scalar = Union[int, float, str]
def is_scalar(x: Any) -> bool:
return isinstance(x, (int, float, str))
def load_metrics_from_json(json_path: str) -> Dict[str, Scalar]:
"""Load scalar metrics from all_cases.json."""
with open(json_path, "r", encoding="utf-8") as f:
data = json.load(f)
if isinstance(data, dict) and "metrics" in data and isinstance(data["metrics"], dict):
metrics_dict = data["metrics"]
else:
metrics_dict = data
out: Dict[str, Scalar] = {}
if isinstance(metrics_dict, dict):
for k, v in metrics_dict.items():
if is_scalar(v):
out[k] = v
return out
def collect_all_rows(root_dir: str, run: str, best_checkpoint: str = None, model_name: str = "qwen2.5-3B") -> Tuple[List[Dict[str, Scalar]], List[str]]:
"""Walk the checkpoints directory and collect rows + column names."""
rows: List[Dict[str, Scalar]] = []
all_metric_cols: Set[str] = set()
if not os.path.isdir(root_dir):
# Graceful exit or warning if root doesn't exist
print(f"[WARN] Root directory not found: {root_dir}")
return [], ["checkpoint"]
# 1. Collect Raw Model Metrics
ckpt_path = os.path.join(root_dir, "raw_model")
row: Dict[str, Scalar] = {"checkpoint": f"{model_name}"}
if os.path.isdir(ckpt_path):
for dataset_name in sorted(os.listdir(ckpt_path)):
if "neu" in dataset_name.lower():
continue
dataset_path = os.path.join(ckpt_path, dataset_name)
dataset_name = dataset_name.lower()
if not os.path.isdir(dataset_path):
continue
json_path = os.path.join(dataset_path, "raw_results_train_all.json")
if not os.path.isfile(json_path):
continue
try:
metrics = load_metrics_from_json(json_path)
except Exception as e:
print(f"[WARN] Failed to read {json_path}: {e}")
continue
_process_metrics_into_row(row, all_metric_cols, dataset_name, metrics)
# Only add raw model row if we actually found data
if len(row) > 1:
rows.append(row)
# 2. Collect Checkpoint Metrics
root_dir = os.path.join(root_dir, run)
if not os.path.isdir(root_dir):
print(f"[WARN] Run directory not found: {root_dir}")
ordered_cols = ["checkpoint"] + sorted(all_metric_cols)
return rows, ordered_cols
try:
# Filter for directories that look like steps (integers)
training_step = []
for d in os.listdir(root_dir):
if "-" in d:
try:
step_val = int(d.split("-")[-1])
training_step.append(step_val)
except ValueError:
continue
for step in sorted(training_step):
ckpt_name = f"checkpoint-{step}"
ckpt_path = os.path.join(root_dir, ckpt_name)
if not os.path.isdir(ckpt_path):
continue
if best_checkpoint and ckpt_name == best_checkpoint:
row: Dict[str, Scalar] = {"checkpoint": ckpt_name+"(best)"}
else:
row: Dict[str, Scalar] = {"checkpoint": ckpt_name}
for dataset_name in sorted(os.listdir(ckpt_path)):
dataset_path = os.path.join(ckpt_path, dataset_name)
dataset_name = dataset_name.lower()
if not os.path.isdir(dataset_path):
continue
# Try common filenames
json_path = os.path.join(dataset_path, "all_cases.json")
if not os.path.isfile(json_path):
json_path = os.path.join(dataset_path, "all_casses.json") # Typo fallback
if not os.path.isfile(json_path):
continue
try:
metrics = load_metrics_from_json(json_path)
except Exception as e:
print(f"[WARN] Failed to read {json_path}: {e}")
continue
_process_metrics_into_row(row, all_metric_cols, dataset_name, metrics)
rows.append(row)
except Exception as e:
print(f"[ERROR] Error walking directories: {e}")
ordered_cols = ["checkpoint"] + sorted(all_metric_cols)
return rows, ordered_cols
def _process_metrics_into_row(row, all_metric_cols, dataset_name, metrics):
"""Helper to standardize metric naming logic."""
f1_flag = False
possible_col_names = []
for metric_name, metric_value in metrics.items():
col_name = None
if "accuracy" in metric_name or "hamming_accuracy" in metric_name:
col_name = f"{dataset_name}_acc"
elif "macro_f1" in metric_name or "f1_macro" in metric_name or "f1" in metric_name:
col_name = f"{dataset_name}_f1"
f1_flag = True
elif "precision" in metric_name:
col_name = f"{dataset_name}_precision"
elif "recall" in metric_name:
col_name = f"{dataset_name}_recall"
elif "exact_match_accuracy" in metric_name:
col_name = f"{dataset_name}_EM"
if col_name:
possible_col_names.append((col_name, metric_value))
for col_name, metric_value in possible_col_names:
# Priority logic: if F1 exists, prioritize it?
# The original logic seemed to allow both, but checked flags.
# Preserving original logic structure:
if f1_flag and ("f1" in col_name or "_f1" in col_name):
if col_name not in row:
row[col_name] = round(metric_value, 4)
all_metric_cols.add(col_name)
elif not f1_flag:
if col_name not in row:
row[col_name] = round(metric_value, 4)
all_metric_cols.add(col_name)
def clean_sheet_name(name):
"""Ensure Excel sheet name is valid."""
if not name: return "Sheet1"
invalid = ['\\', '/', '*', '?', ':', '[', ']']
for c in invalid:
name = name.replace(c, '_')
return name[:31] # Excel limit
def _checkpoint_key(name: str) -> str:
"""Normalize checkpoint name by stripping '(best)' etc."""
return name.split("(")[0].strip() if name else ""
def append_rows_in_place(
rows: List[Dict[str, Scalar]],
columns: List[str],
out_path: str,
sheet_name: str,
best_checkpoint: str,
model_name: str,
old_sheet_name: str
) -> None:
"""Update an existing Excel sheet in-place."""
sheet_name = clean_sheet_name(sheet_name)
final_columns = columns + ["better_datasets_than_raw"]
metric_cols = [c for c in columns if c != "checkpoint"]
dataset_to_cols: Dict[str, List[str]] = {}
for col in metric_cols:
ds = col.split("_", 1)[0]
dataset_to_cols.setdefault(ds, []).append(col)
baseline = None
for r in rows:
if r.get("checkpoint") == model_name:
baseline = r
break
wb = load_workbook(out_path)
if sheet_name in wb.sheetnames:
ws = wb[sheet_name]
else:
ws = wb.create_sheet(title=sheet_name)
for idx, col_name in enumerate(final_columns, start=1):
ws.cell(row=1, column=idx, value=col_name)
# Re-map headers in case columns changed or are in different order in file
header_map: Dict[str, int] = {}
max_col = ws.max_column
# Read existing headers
for idx, cell in enumerate(ws[1], start=1):
if cell.value is not None:
header_map[str(cell.value)] = idx
# Add new columns if missing
for col_name in final_columns:
if col_name not in header_map:
max_col += 1
ws.cell(row=1, column=max_col, value=col_name)
header_map[col_name] = max_col
col_index = header_map
first_data_row = 2
# Map existing checkpoints to row numbers
existing_row_map: Dict[str, int] = {}
mx_row = ws.max_row
if "checkpoint" in col_index:
chk_col = col_index["checkpoint"]
for r_idx in range(first_data_row, ws.max_row + 1):
val = ws.cell(row=r_idx, column=chk_col).value
if val:
ck = _checkpoint_key(str(val))
existing_row_map[ck] = r_idx
else:
# Should not happen if file was created correctly
chk_col = 1
green_fill = PatternFill(start_color="90EE90", end_color="90EE90", fill_type="solid")
no_fill = PatternFill(fill_type=None)
best_font = Font(color="008000", bold=True)
normal_font = Font(color="000000")
for row_data in rows:
ckpt_val = row_data.get("checkpoint")
if ckpt_val is None:
continue
ck_key = _checkpoint_key(str(ckpt_val))
# Calculate metrics comparison
better_datasets_count = 0
metric_better_flags: Dict[str, bool] = {}
if baseline is not None:
for col in metric_cols:
v = row_data.get(col)
b = baseline.get(col)
improved = False
try:
if v is not None and b is not None:
if float(v) > float(b):
improved = True
except Exception:
improved = False
metric_better_flags[col] = improved
for ds, ds_cols in dataset_to_cols.items():
if any(metric_better_flags.get(c, False) for c in ds_cols):
better_datasets_count += 1
# Determine target row
if ck_key in existing_row_map:
excel_row = existing_row_map[ck_key]
else:
mx_row += 1
excel_row = mx_row
existing_row_map[ck_key] = excel_row
# Write data
for col_name in final_columns:
if col_name not in col_index: continue
idx = col_index[col_name]
cell = ws.cell(row=excel_row, column=idx)
if col_name == "better_datasets_than_raw":
value = better_datasets_count
else:
value = row_data.get(col_name, "")
cell.value = value
cell.alignment = Alignment(horizontal='center', vertical='center')
# Conditional formatting for metrics
if col_name in metric_cols:
if metric_better_flags.get(col_name, False):
cell.fill = green_fill
else:
cell.fill = no_fill
# Highlight best checkpoint name
checkpoint_cell = ws.cell(row=excel_row, column=chk_col)
if best_checkpoint and _checkpoint_key(best_checkpoint) == ck_key:
checkpoint_cell.font = best_font
else:
checkpoint_cell.font = normal_font
# Ensure checkpoint name has (best) suffix if applicable
if best_checkpoint and _checkpoint_key(best_checkpoint) == ck_key:
checkpoint_cell.value = best_checkpoint + "(best)" if "(best)" not in str(checkpoint_cell.value) else checkpoint_cell.value
# Auto-adjust column widths (simple approximation)
for col_name, idx in col_index.items():
col_letter = ws.cell(row=1, column=idx).column_letter
ws.column_dimensions[col_letter].width = max(len(col_name) + 2, 12)
wb.save(out_path)
wb.close()
def write_excel(
rows: List[Dict[str, Scalar]],
columns: List[str],
out_path: str,
sheet_name: str = "Sheet1",
old_sheet_name: str = None,
best_checkpoint: str = None,
model_name: str = "qwen2.5-3B"
) -> None:
"""
Write data to a local Excel file.
Creates file if it doesn't exist, otherwise appends/updates via append_rows_in_place.
"""
sheet_name = clean_sheet_name(sheet_name)
if os.path.exists(out_path):
try:
append_rows_in_place(
rows=rows,
columns=columns,
out_path=out_path,
sheet_name=sheet_name,
best_checkpoint=best_checkpoint,
model_name=model_name,
old_sheet_name=old_sheet_name
)
print(f"Excel updated: {out_path} (sheet: {sheet_name})")
return
except Exception as e:
print(f"[WARN] Failed to update existing Excel: {e}. Attempting full rewrite.")
# Create new DataFrame and Excel file
table = [{col: row.get(col, "") for col in columns} for row in rows]
df = pd.DataFrame(table, columns=columns)
metric_cols = [c for c in df.columns if c != "checkpoint"]
if metric_cols:
df[metric_cols] = df[metric_cols].apply(pd.to_numeric, errors="coerce")
# df[metric_cols] = np.round(df[metric_cols], 4)
# Calculate comparisons for new file
has_baseline = False
better_mask = None
if "checkpoint" in df.columns and (df["checkpoint"] == f"{model_name}").any() and metric_cols:
has_baseline = True
raw_idx = df.index[df["checkpoint"] == f"{model_name}"][0]
raw_values = df.loc[raw_idx, metric_cols]
better_mask = df[metric_cols].gt(raw_values).fillna(False)
dataset_to_cols = {}
for col in metric_cols:
dataset_name = col.split("_", 1)[0]
dataset_to_cols.setdefault(dataset_name, []).append(col)
better_datasets_counts = []
for idx, row_bool in better_mask.iterrows():
count = 0
for ds, ds_cols in dataset_to_cols.items():
if row_bool[ds_cols].any():
count += 1
better_datasets_counts.append(count)
df["better_datasets_than_raw"] = better_datasets_counts
else:
df["better_datasets_than_raw"] = 0
final_columns = columns + ["better_datasets_than_raw"]
# Write using Pandas/OpenPyxl
with pd.ExcelWriter(out_path, engine="openpyxl") as writer:
df.to_excel(writer, index=False, sheet_name=sheet_name, columns=final_columns)
ws = writer.sheets[sheet_name]
green_fill = PatternFill(start_color="90EE90", end_color="90EE90", fill_type="solid")
# Formatting
for r_idx, row in enumerate(ws.iter_rows(min_row=2), start=0): # Data rows
for cell in row:
cell.alignment = Alignment(horizontal='center', vertical='center')
# Highlight cells better than baseline
if has_baseline and better_mask is not None:
row_label = df.index[r_idx]
for col_name in metric_cols:
if col_name in df.columns:
col_idx = df.columns.get_loc(col_name) # pandas index
# Map to excel column (1-based, adjusted for list)
# Actually simpler: find column index in final_columns
if col_name in final_columns:
excel_col_idx = final_columns.index(col_name) + 1
if bool(better_mask.loc[row_label, col_name]):
ws.cell(row=r_idx+2, column=excel_col_idx).fill = green_fill
# Highlight best checkpoint
chk_val = df.iloc[r_idx]["checkpoint"]
if best_checkpoint and _checkpoint_key(str(chk_val)) == _checkpoint_key(best_checkpoint):
ws.cell(row=r_idx+2, column=final_columns.index("checkpoint")+1).font = Font(color="008000", bold=True)
print(f"Excel created: {out_path} (sheet: {sheet_name})")
def main():
parser = argparse.ArgumentParser(
description="Create a metrics table from checkpoint JSON files."
)
parser.add_argument(
"--root",
type=str,
default="./SFT/Evaluation/",
help="Root directory containing checkpoints (default: checkpoints)",
)
parser.add_argument(
"--out_filename",
type=str,
default="./SFT/Evaluation//metrics_summary.xlsx",
help="Output Excel file path (default: metrics_summary.xlsx)",
)
parser.add_argument(
"--run",
type=str,
default="SFT_dt12.11.19:13_e6_unsloth_Qwen2.5_14B_Instruct_bnb_4bit_bnb_4bit_lr5e-06_t0.0_r64_b4_SFT_Implementation",
help="Name of the Excel sheet (subsheet) to write results into.",
)
parser.add_argument(
"--best_checkpoint",
type=str,
default=None,
help="Name of checkpoint whose name in the first column will be colored green.",
)
parser.add_argument(
"--base_model_name",
type=str,
default="qwen2.5-3B",
help="Name of the base model we trained on",
)
parser.add_argument(
"--base_result_dir",
type=str,
default="./SFT/results_sft_14b",
help="Directory of the base model we trained on",
)
parser.add_argument(
"--train_data",
type=str,
default="UniADILR",
)
args = parser.parse_args()
# --- Directory Setup ---
# Ensure the table_result directory exists
output_dir = "table_result"
os.makedirs(output_dir, exist_ok=True)
# Construct final path
final_out_path = os.path.join(output_dir, os.path.basename(args.out_filename))
# --- Best Checkpoint Auto-Detection ---
if args.best_checkpoint is None:
BASE_RESULTS_DIR = os.path.expanduser(args.base_result_dir)
TRAINING_DIR = os.path.join(BASE_RESULTS_DIR, f"Training_{args.run}")
FINAL_DIR = os.path.join(BASE_RESULTS_DIR, args.run)
TRAINING_BASE = None
if os.path.isdir(TRAINING_DIR):
TRAINING_BASE = TRAINING_DIR
elif os.path.isdir(FINAL_DIR):
TRAINING_BASE = FINAL_DIR
if TRAINING_BASE:
best_path, _ = find_best_checkpoint(TRAINING_BASE)
if best_path:
args.best_checkpoint = os.path.basename(best_path)
# --- Data Collection ---
print(f"Collecting metrics for run: {args.run}")
rows, columns = collect_all_rows(args.root, args.run, args.best_checkpoint, args.base_model_name)
if not rows:
print("No metrics found. Exiting.")
return
# --- Sheet Naming Logic ---
# Attempt to format the sheet name based on the specific pattern provided in the prompt
# Fallback to args.run if the split/format fails
try:
parts = args.run.split("e20_")
if len(parts) >= 2:
sheet_name = parts[0] + args.train_data + "_e20_" + parts[1]
else:
sheet_name = args.run
except Exception:
sheet_name = args.run
old_sheet_name = args.run
# --- Save to Local Excel ---
write_excel(
rows=rows,
columns=columns,
out_path=final_out_path,
sheet_name=sheet_name,
old_sheet_name=old_sheet_name,
best_checkpoint=args.best_checkpoint,
model_name=args.base_model_name,
)
if __name__ == "__main__":
main() |