Repository navigation
Expand file tree
/
Copy pathrepack_experts.py
More file actions
661 lines (559 loc) · 26.4 KB
/
Copy pathrepack_experts.py
File metadata and controls
661 lines (559 loc) · 26.4 KB
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
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
#!/usr/bin/env python3
"""Repack expert weights from scattered safetensors into contiguous per-layer binary files.
Creates one binary file per layer: packed_experts/layer_XX.bin
Each file = N_EXPERTS x EXPERT_SIZE bytes
Within each expert block, 9 components packed in fixed order:
gate_proj.weight, gate_proj.scales, gate_proj.biases,
up_proj.weight, up_proj.scales, up_proj.biases,
down_proj.weight, down_proj.scales, down_proj.biases
Usage:
python repack_experts.py --model ~/qwen35-397b-4bit
python repack_experts.py --model ~/qwen3-coder-480b-4bit --profile qwen3-coder
python repack_experts.py --model ~/qwen35-397b-4bit --layers 0-4
python repack_experts.py --model ~/qwen35-397b-4bit --dry-run
python repack_experts.py --model ~/qwen35-397b-4bit --verify-only 0
python repack_experts.py --model ~/qwen35-397b-4bit --validate-model
"""
import argparse
import json
import os
import re
import struct
import time
import sys
# ============================================================================
# Model profiles
# ============================================================================
MODEL_PROFILES = {
"qwen35-397b": {
"components": [
{"name": "gate_proj.weight", "offset": 0, "size": 2097152, "dtype": "U32", "shape": [1024, 512]},
{"name": "gate_proj.scales", "offset": 2097152, "size": 131072, "dtype": "BF16", "shape": [1024, 64]},
{"name": "gate_proj.biases", "offset": 2228224, "size": 131072, "dtype": "BF16", "shape": [1024, 64]},
{"name": "up_proj.weight", "offset": 2359296, "size": 2097152, "dtype": "U32", "shape": [1024, 512]},
{"name": "up_proj.scales", "offset": 4456448, "size": 131072, "dtype": "BF16", "shape": [1024, 64]},
{"name": "up_proj.biases", "offset": 4587520, "size": 131072, "dtype": "BF16", "shape": [1024, 64]},
{"name": "down_proj.weight", "offset": 4718592, "size": 2097152, "dtype": "U32", "shape": [4096, 128]},
{"name": "down_proj.scales", "offset": 6815744, "size": 131072, "dtype": "BF16", "shape": [4096, 16]},
{"name": "down_proj.biases", "offset": 6946816, "size": 131072, "dtype": "BF16", "shape": [4096, 16]},
],
"expert_size": 7077888,
"num_experts": 512,
"num_layers": 60,
"weight_prefix": "language_model.",
"supported_config": {
"hidden_size": 4096,
"num_hidden_layers": 60,
"num_experts": 512,
"num_experts_per_tok": 10,
},
"supported_shard_count": 46,
"supported_bits": 4,
# regex: group(1)=layer, group(2)=proj, group(3)=part (weight/scales/biases)
"expert_key_pattern": r"language_model\.model\.layers\.(\d+)\.mlp\.experts\.(\d+)\.(\w+)\.(weight|scales|biases)",
"expert_key_type": "individual", # one tensor per expert
},
"qwen3-coder": {
"components": [
{"name": "gate_proj.weight", "offset": 0, "size": 7864320, "dtype": "U32", "shape": [2560, 768]},
{"name": "gate_proj.scales", "offset": 7864320, "size": 491520, "dtype": "BF16", "shape": [2560, 96]},
{"name": "gate_proj.biases", "offset": 8355840, "size": 491520, "dtype": "BF16", "shape": [2560, 96]},
{"name": "up_proj.weight", "offset": 8847360, "size": 7864320, "dtype": "U32", "shape": [2560, 768]},
{"name": "up_proj.scales", "offset": 16711680, "size": 491520, "dtype": "BF16", "shape": [2560, 96]},
{"name": "up_proj.biases", "offset": 17203200, "size": 491520, "dtype": "BF16", "shape": [2560, 96]},
{"name": "down_proj.weight", "offset": 17694720, "size": 7864320, "dtype": "U32", "shape": [6144, 320]},
{"name": "down_proj.scales", "offset": 25559040, "size": 491520, "dtype": "BF16", "shape": [6144, 40]},
{"name": "down_proj.biases", "offset": 26050560, "size": 491520, "dtype": "BF16", "shape": [6144, 40]},
],
"expert_size": 26542080,
"num_experts": 160,
"num_layers": 62,
"weight_prefix": "",
"supported_config": {
"hidden_size": 6144,
"num_hidden_layers": 62,
"num_experts": 160,
"num_experts_per_tok": 8,
},
"supported_shard_count": None, # not validated
"supported_bits": 4,
# regex: group(1)=layer, group(2)=proj, group(3)=part (weight/scales/biases)
"expert_key_pattern": r"model\.layers\.(\d+)\.mlp\.switch_mlp\.(\w+)\.(weight|scales|biases)",
"expert_key_type": "packed", # all experts in one tensor [N, ...]
},
}
# Default profile
ACTIVE_PROFILE = None
COMPONENTS = None
EXPERT_SIZE = None
NUM_EXPERTS = None
NUM_LAYERS = None
LAYER_SIZE = None
WEIGHT_PREFIX = None
SUPPORTED_MODEL_CONFIG = None
SUPPORTED_SHARD_COUNT = None
SUPPORTED_BITS = None
EXPERT_KEY_PATTERN = None
EXPERT_KEY_TYPE = None
def set_profile(name):
"""Set the active model profile."""
global ACTIVE_PROFILE, COMPONENTS, EXPERT_SIZE, NUM_EXPERTS, NUM_LAYERS
global LAYER_SIZE, WEIGHT_PREFIX, SUPPORTED_MODEL_CONFIG
global SUPPORTED_SHARD_COUNT, SUPPORTED_BITS
global EXPERT_KEY_PATTERN, EXPERT_KEY_TYPE
if name not in MODEL_PROFILES:
print(f"ERROR: unknown model profile '{name}'. Available: {', '.join(MODEL_PROFILES.keys())}", file=sys.stderr)
sys.exit(1)
p = MODEL_PROFILES[name]
ACTIVE_PROFILE = name
COMPONENTS = p["components"]
EXPERT_SIZE = p["expert_size"]
NUM_EXPERTS = p["num_experts"]
NUM_LAYERS = p["num_layers"]
LAYER_SIZE = NUM_EXPERTS * EXPERT_SIZE
WEIGHT_PREFIX = p["weight_prefix"]
SUPPORTED_MODEL_CONFIG = p["supported_config"]
SUPPORTED_SHARD_COUNT = p["supported_shard_count"]
SUPPORTED_BITS = p["supported_bits"]
EXPERT_KEY_PATTERN = p["expert_key_pattern"]
EXPERT_KEY_TYPE = p["expert_key_type"]
def parse_layers(spec):
"""Parse layer specification like '0-4' or '0,5,10' or 'all'."""
if spec is None or spec == 'all':
return list(range(NUM_LAYERS))
layers = []
for part in spec.split(','):
part = part.strip()
if '-' in part:
a, b = part.split('-', 1)
layers.extend(range(int(a), int(b) + 1))
else:
layers.append(int(part))
return sorted(set(layers))
def load_index(index_path):
"""Load expert_index.json and return expert_reads dict."""
with open(index_path) as f:
idx = json.load(f)
return idx['expert_reads']
def generate_expert_index(model_path):
"""Generate expert_reads from safetensors headers for packed-expert models.
For models like Qwen3-Coder where all experts are stored in a single tensor
per component per layer (shape [N_experts, ...]), we scan safetensors headers
to build the same index structure as pre-built expert_index.json.
"""
index_path = os.path.join(model_path, "model.safetensors.index.json")
with open(index_path) as f:
weight_index = json.load(f)
weight_map = weight_index["weight_map"]
pattern = re.compile(EXPERT_KEY_PATTERN)
file_headers = {}
expert_reads = {}
if EXPERT_KEY_TYPE == "packed":
# Packed format: model.layers.X.mlp.switch_mlp.{proj}.{part}
# Each tensor has shape [N_experts, ...], all experts contiguous
for key in sorted(weight_map.keys()):
m = pattern.match(key)
if not m:
continue
layer_idx, proj_name, part_name = m.groups()
comp_name = f"{proj_name}.{part_name}"
shard_file = weight_map[key]
if layer_idx not in expert_reads:
expert_reads[layer_idx] = {}
if shard_file not in file_headers:
shard_path = os.path.join(model_path, shard_file)
with open(shard_path, "rb") as f:
header_size = struct.unpack("<Q", f.read(8))[0]
header = json.loads(f.read(header_size))
file_headers[shard_file] = (header, header_size)
header, h_size = file_headers[shard_file]
tensor_info = header[key]
shape = tensor_info["shape"]
data_start = tensor_info["data_offsets"][0] + 8 + h_size
data_end = tensor_info["data_offsets"][1] + 8 + h_size
total_size = data_end - data_start
num_experts = shape[0]
expert_size = total_size // num_experts
expert_reads[layer_idx][comp_name] = {
"file": shard_file,
"abs_offset": data_start,
"expert_stride": expert_size,
"expert_size": expert_size,
"total_size": total_size,
"shape": shape,
}
else:
# Individual format: each expert is a separate tensor
# Pattern captures layer, expert_idx, proj, part
for key in sorted(weight_map.keys()):
m = pattern.match(key)
if not m:
continue
layer_idx, expert_idx_str, proj_name, part_name = m.groups()
expert_idx = int(expert_idx_str)
comp_name = f"{proj_name}.{part_name}"
shard_file = weight_map[key]
if expert_idx != 0:
continue # only need expert 0 to determine strides
if layer_idx not in expert_reads:
expert_reads[layer_idx] = {}
if shard_file not in file_headers:
shard_path = os.path.join(model_path, shard_file)
with open(shard_path, "rb") as f:
header_size = struct.unpack("<Q", f.read(8))[0]
header = json.loads(f.read(header_size))
file_headers[shard_file] = (header, header_size)
header, h_size = file_headers[shard_file]
tensor_info = header[key]
shape = tensor_info["shape"]
data_start = tensor_info["data_offsets"][0] + 8 + h_size
data_end = tensor_info["data_offsets"][1] + 8 + h_size
expert_size = data_end - data_start
expert_reads[layer_idx][comp_name] = {
"file": shard_file,
"abs_offset": data_start,
"expert_stride": expert_size,
"expert_size": expert_size,
"total_size": expert_size * NUM_EXPERTS,
"shape": [NUM_EXPERTS] + shape,
}
print(f"Generated expert index: {len(expert_reads)} layers, "
f"{len(next(iter(expert_reads.values())))} components/layer")
return expert_reads
def load_model_metadata(model_path):
"""Load config/index metadata for user-facing compatibility checks."""
config_path = os.path.join(model_path, "config.json")
index_path = os.path.join(model_path, "model.safetensors.index.json")
metadata = {
"config_path": config_path,
"index_path": index_path,
"config": {},
"weight_index": {},
"bits": None,
"group_size": None,
"hidden_size": None,
"num_hidden_layers": None,
"num_experts": None,
"num_experts_per_tok": None,
"shard_count": None,
"total_parameters": None,
}
if os.path.isfile(config_path):
with open(config_path) as f:
metadata["config"] = json.load(f)
quant = metadata["config"].get("quantization_config") or metadata["config"].get("quantization") or {}
metadata["bits"] = quant.get("bits")
metadata["group_size"] = quant.get("group_size")
# Multimodal models (e.g. Qwen3.5) nest text params under text_config
text_cfg = metadata["config"].get("text_config", {})
for key in SUPPORTED_MODEL_CONFIG:
metadata[key] = metadata["config"].get(key) or text_cfg.get(key)
if os.path.isfile(index_path):
with open(index_path) as f:
metadata["weight_index"] = json.load(f)
shard_files = sorted(set(metadata["weight_index"].get("weight_map", {}).values()))
metadata["shard_count"] = len(shard_files)
metadata["total_parameters"] = metadata["weight_index"].get("metadata", {}).get("total_parameters")
return metadata
def validate_supported_model(model_path, expert_reads):
"""Fail fast with a clear error when the model directory is incompatible."""
metadata = load_model_metadata(model_path)
problems = []
for key, expected in SUPPORTED_MODEL_CONFIG.items():
actual = metadata.get(key)
if actual != expected:
problems.append(f"{key}={actual!r} (expected {expected})")
if metadata["bits"] != SUPPORTED_BITS:
problems.append(f"quantization bits={metadata['bits']!r} (expected {SUPPORTED_BITS})")
if SUPPORTED_SHARD_COUNT is not None and metadata["shard_count"] != SUPPORTED_SHARD_COUNT:
problems.append(f"shard_count={metadata['shard_count']!r} (expected {SUPPORTED_SHARD_COUNT})")
missing_files = []
if expert_reads is not None:
expected_files = sorted({
info["file"]
for layer_info in expert_reads.values()
for info in layer_info.values()
})
missing_files = [
fname for fname in expected_files
if not os.path.isfile(os.path.join(model_path, fname))
]
if missing_files:
problems.append(
f"missing {len(missing_files)} shard(s) referenced by expert_index.json"
)
if problems:
detected = [
f"hidden_size={metadata['hidden_size']!r}",
f"num_hidden_layers={metadata['num_hidden_layers']!r}",
f"num_experts={metadata['num_experts']!r}",
f"num_experts_per_tok={metadata['num_experts_per_tok']!r}",
f"bits={metadata['bits']!r}",
f"shards={metadata['shard_count']!r}",
]
if metadata["total_parameters"] is not None:
detected.append(f"total_parameters={metadata['total_parameters']}")
print(f"ERROR: unsupported model directory: {model_path}", file=sys.stderr)
print(f"Active profile: {ACTIVE_PROFILE}", file=sys.stderr)
print(f"Detected metadata: {', '.join(detected)}", file=sys.stderr)
print("Compatibility check failed:", file=sys.stderr)
for problem in problems:
print(f" - {problem}", file=sys.stderr)
if SUPPORTED_SHARD_COUNT is not None and metadata["shard_count"] not in (None, SUPPORTED_SHARD_COUNT):
print(
f"Hint: --model/--model-dir points at a different MLX model. "
f"Check that --profile matches your model.",
file=sys.stderr,
)
elif missing_files:
preview = ", ".join(missing_files[:3])
if len(missing_files) > 3:
preview += ", ..."
print(f"Missing files include: {preview}", file=sys.stderr)
print(
"Hint: the model download is incomplete. Re-run hf download to fetch the missing shards.",
file=sys.stderr,
)
return False
print(f"Model metadata verified (profile: {ACTIVE_PROFILE})")
return True
def verify_component_sizes(expert_reads):
"""Verify that component sizes in the index match expected sizes."""
expected = {c['name']: c['size'] for c in COMPONENTS}
for layer_key, comps in expert_reads.items():
for comp_name, info in comps.items():
if comp_name not in expected:
print(f"WARNING: unknown component {comp_name} in layer {layer_key}")
continue
if info['expert_size'] != expected[comp_name]:
print(f"MISMATCH: layer {layer_key}, {comp_name}: "
f"index says {info['expert_size']}, expected {expected[comp_name]}")
return False
print("Component sizes verified: all match expected layout")
return True
def open_source_files(expert_reads, model_path, layers):
"""Open all needed safetensors files, return {filename: fd}."""
needed_files = set()
for layer_idx in layers:
layer_key = str(layer_idx)
if layer_key not in expert_reads:
print(f"WARNING: layer {layer_idx} not found in expert_reads")
continue
for info in expert_reads[layer_key].values():
needed_files.add(info['file'])
fds = {}
for fname in sorted(needed_files):
path = os.path.join(model_path, fname)
fds[fname] = os.open(path, os.O_RDONLY)
print(f"Opened {len(fds)} source safetensors files")
return fds
def repack_layer(layer_idx, expert_reads, fds, output_dir, dry_run=False):
"""Repack all 512 experts for one layer into a contiguous binary file.
Returns (bytes_written, elapsed_seconds).
"""
layer_key = str(layer_idx)
if layer_key not in expert_reads:
print(f" Layer {layer_idx}: NOT FOUND in index, skipping")
return 0, 0.0
layer_info = expert_reads[layer_key]
out_path = os.path.join(output_dir, f"layer_{layer_idx:02d}.bin")
if dry_run:
# Just verify we can compute all offsets
for expert_idx in range(NUM_EXPERTS):
for comp in COMPONENTS:
info = layer_info[comp['name']]
src_offset = info['abs_offset'] + expert_idx * info['expert_stride']
dst_offset = expert_idx * EXPERT_SIZE + comp['offset']
print(f" Layer {layer_idx:2d}: DRY RUN OK — would write {LAYER_SIZE:,} bytes to {out_path}")
return LAYER_SIZE, 0.0
t0 = time.monotonic()
# Pre-allocate output file with zeros
fd_out = os.open(out_path, os.O_RDWR | os.O_CREAT | os.O_TRUNC, 0o644)
os.ftruncate(fd_out, LAYER_SIZE)
bytes_written = 0
# Build read plan: group reads by source file for better locality
# Each entry: (src_fd, src_offset, dst_offset, size)
read_plan = []
for expert_idx in range(NUM_EXPERTS):
for comp in COMPONENTS:
info = layer_info[comp['name']]
src_fd = fds[info['file']]
src_offset = info['abs_offset'] + expert_idx * info['expert_stride']
dst_offset = expert_idx * EXPERT_SIZE + comp['offset']
read_plan.append((src_fd, src_offset, dst_offset, comp['size']))
# Sort by (src_fd, src_offset) for sequential read locality
read_plan.sort(key=lambda x: (x[0], x[1]))
# Execute reads and writes
for src_fd, src_offset, dst_offset, size in read_plan:
data = os.pread(src_fd, size, src_offset)
if len(data) != size:
raise IOError(f"Short read: expected {size}, got {len(data)} "
f"at offset {src_offset}")
os.pwrite(fd_out, data, dst_offset)
bytes_written += size
os.close(fd_out)
elapsed = time.monotonic() - t0
return bytes_written, elapsed
def verify_layer(layer_idx, expert_reads, fds, output_dir):
"""Read back several experts from packed file and compare to originals."""
layer_key = str(layer_idx)
layer_info = expert_reads[layer_key]
out_path = os.path.join(output_dir, f"layer_{layer_idx:02d}.bin")
if not os.path.exists(out_path):
print(f" Layer {layer_idx}: packed file not found")
return False
fd_packed = os.open(out_path, os.O_RDONLY)
# Spot-check: first, second, middle, last experts
check_indices = sorted(set([0, 1, NUM_EXPERTS // 2, NUM_EXPERTS - 1]))
mismatches = 0
for expert_idx in check_indices:
for comp in COMPONENTS:
info = layer_info[comp['name']]
src_fd = fds[info['file']]
src_offset = info['abs_offset'] + expert_idx * info['expert_stride']
dst_offset = expert_idx * EXPERT_SIZE + comp['offset']
original = os.pread(src_fd, comp['size'], src_offset)
packed = os.pread(fd_packed, comp['size'], dst_offset)
if original != packed:
print(f" MISMATCH: layer {layer_idx}, expert {expert_idx}, {comp['name']}")
mismatches += 1
os.close(fd_packed)
indices_str = ", ".join(str(i) for i in check_indices)
if mismatches == 0:
print(f" Layer {layer_idx}: verification PASSED (experts {indices_str})")
else:
print(f" Layer {layer_idx}: verification FAILED ({mismatches} mismatches)")
return mismatches == 0
def write_layout(output_dir):
"""Write layout.json describing the packed format."""
layout = {
"expert_size": EXPERT_SIZE,
"num_layers": NUM_LAYERS,
"num_experts": NUM_EXPERTS,
"components": COMPONENTS,
}
path = os.path.join(output_dir, "layout.json")
with open(path, 'w') as f:
json.dump(layout, f, indent=2)
print(f"Wrote {path}")
def main():
parser = argparse.ArgumentParser(description="Repack expert weights into contiguous per-layer binary files")
default_index = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'expert_index.json')
parser.add_argument('--model', required=True,
help='Path to model directory containing safetensors shards')
parser.add_argument('--profile', default='qwen35-397b',
choices=list(MODEL_PROFILES.keys()),
help='Model profile (default: qwen35-397b)')
parser.add_argument('--index', default=default_index,
help='Path to expert_index.json')
parser.add_argument('--layers', default=None,
help='Layer spec: "all", "0-4", "0,5,10" (default: all)')
parser.add_argument('--dry-run', action='store_true',
help='Verify offsets without writing')
parser.add_argument('--verify-only', type=int, default=None, metavar='LAYER',
help='Verify a specific layer against originals')
parser.add_argument('--validate-model', action='store_true',
help='Check that --model matches the supported layout and exit')
args = parser.parse_args()
set_profile(args.profile)
print(f"Model profile: {args.profile}")
print(f"Expert size: {EXPERT_SIZE:,} bytes, Experts/layer: {NUM_EXPERTS}, Layers: {NUM_LAYERS}")
model_path = os.path.abspath(os.path.expanduser(args.model))
index_path = os.path.abspath(os.path.expanduser(args.index))
if not os.path.isdir(model_path):
print(f"ERROR: model directory not found: {model_path}", file=sys.stderr)
sys.exit(1)
# --validate-model: only check config.json metadata, no index needed
if args.validate_model:
if not validate_supported_model(model_path, expert_reads=None):
sys.exit(1)
return
# Load or generate expert index
use_auto_index = EXPERT_KEY_TYPE == "packed"
if use_auto_index:
print("Generating expert index from safetensors headers...")
expert_reads = generate_expert_index(model_path)
else:
if not os.path.isfile(index_path):
print(f"ERROR: expert index not found: {index_path}", file=sys.stderr)
sys.exit(1)
print("Loading expert index...")
expert_reads = load_index(index_path)
print(f"Index path: {index_path}")
print(f"Model path: {model_path}")
print(f"Layers in index: {len(expert_reads)}")
# Verify component sizes
if not verify_component_sizes(expert_reads):
print("ABORTING: component size mismatch")
sys.exit(1)
if not validate_supported_model(model_path, expert_reads):
sys.exit(1)
output_dir = os.path.join(model_path, "packed_experts")
os.makedirs(output_dir, exist_ok=True)
print(f"Output directory: {output_dir}")
# Determine which layers to process
if args.verify_only is not None:
layers = [args.verify_only]
else:
layers = parse_layers(args.layers)
print(f"Layers to process: {layers[0]}-{layers[-1]} ({len(layers)} layers)")
if not args.dry_run and args.verify_only is None:
total_bytes = len(layers) * LAYER_SIZE
print(f"Total data to write: {total_bytes / (1024**3):.1f} GB")
# Check free disk space
stat = os.statvfs(output_dir)
free_bytes = stat.f_bavail * stat.f_frsize
free_gb = free_bytes / (1024**3)
needed_gb = total_bytes / (1024**3)
print(f"Free disk space: {free_gb:.1f} GB, needed: {needed_gb:.1f} GB")
if free_bytes < total_bytes:
print(f"WARNING: Not enough free space! Need {needed_gb:.1f} GB but only {free_gb:.1f} GB free.")
print(f"Hint: use --layers to process a subset, e.g. --layers 0-{int(free_gb / 3.63) - 1}")
sys.exit(1)
# Open source files
fds = open_source_files(expert_reads, model_path, layers)
if args.verify_only is not None:
verify_layer(args.verify_only, expert_reads, fds, output_dir)
for fd in fds.values():
os.close(fd)
return
# Write layout.json
write_layout(output_dir)
# Repack each layer
t_start = time.monotonic()
total_written = 0
for i, layer_idx in enumerate(layers):
t_layer = time.monotonic()
bytes_written, elapsed = repack_layer(
layer_idx, expert_reads, fds, output_dir, dry_run=args.dry_run
)
total_written += bytes_written
if not args.dry_run and bytes_written > 0:
throughput = bytes_written / elapsed / (1024**3) if elapsed > 0 else float('inf')
overall_elapsed = time.monotonic() - t_start
overall_throughput = total_written / overall_elapsed / (1024**3) if overall_elapsed > 0 else 0
eta = (len(layers) - i - 1) * (overall_elapsed / (i + 1))
print(f" Layer {layer_idx:2d}: {bytes_written/1024**3:.2f} GB in {elapsed:.1f}s "
f"({throughput:.1f} GB/s) | "
f"Total: {total_written/1024**3:.1f}/{len(layers)*LAYER_SIZE/1024**3:.1f} GB "
f"({overall_throughput:.1f} GB/s avg) | "
f"ETA: {eta:.0f}s")
# Verify this layer immediately
if not verify_layer(layer_idx, expert_reads, fds, output_dir):
print(f"ABORTING: verification failed for layer {layer_idx}")
sys.exit(1)
# Close source files
for fd in fds.values():
os.close(fd)
# Final summary
total_elapsed = time.monotonic() - t_start
if not args.dry_run and total_written > 0:
print(f"\n{'='*60}")
print(f"DONE: {total_written:,} bytes ({total_written/1024**3:.1f} GB) written")
print(f"Time: {total_elapsed:.1f}s")
print(f"Throughput: {total_written/total_elapsed/1024**3:.1f} GB/s")
print(f"Output: {output_dir}")
elif args.dry_run:
print(f"\nDRY RUN complete: {len(layers)} layers validated")
if __name__ == '__main__':
main()