Spaces:
Sleeping
Sleeping
File size: 35,283 Bytes
a690cfc 4419bdb a690cfc 4419bdb a690cfc 4419bdb a690cfc 4419bdb a690cfc 4419bdb a690cfc 4419bdb a690cfc 4419bdb a690cfc 4419bdb a690cfc 029e89b 4419bdb 029e89b 4419bdb 029e89b 4419bdb 029e89b 4419bdb 029e89b 4419bdb 029e89b 4419bdb a690cfc 029e89b 4419bdb 029e89b a690cfc 029e89b 4419bdb 029e89b 4419bdb a690cfc 4419bdb a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b 4419bdb 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 4419bdb 029e89b 4419bdb 029e89b 4419bdb 029e89b 4419bdb 029e89b a690cfc 029e89b a690cfc 029e89b 4419bdb 029e89b 4419bdb 029e89b 4419bdb 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b 4419bdb 029e89b 4419bdb 029e89b 4419bdb 029e89b a690cfc 029e89b a690cfc 029e89b 4419bdb a690cfc 029e89b a690cfc 029e89b 4419bdb a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc 029e89b a690cfc bc7b711 a690cfc | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 | import gradio as gr
import torch
import os
import gc
import shutil
import requests
import json
import struct
import numpy as np
import re
from pathlib import Path
from typing import Dict, Any, Optional, List
from huggingface_hub import HfApi, hf_hub_download, list_repo_files, login
from safetensors.torch import load_file, save_file
from tqdm import tqdm
# --- Memory Efficient Safetensors ---
class MemoryEfficientSafeOpen:
"""
Reads safetensors metadata and tensors without mmap, keeping RAM usage low.
"""
def __init__(self, filename):
self.filename = filename
self.file = open(filename, "rb")
self.header, self.header_size = self._read_header()
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.file.close()
def keys(self) -> list[str]:
return [k for k in self.header.keys() if k != "__metadata__"]
def metadata(self) -> Dict[str, str]:
return self.header.get("__metadata__", {})
def get_tensor(self, key):
if key not in self.header:
raise KeyError(f"Tensor '{key}' not found in the file")
metadata = self.header[key]
offset_start, offset_end = metadata["data_offsets"]
self.file.seek(self.header_size + 8 + offset_start)
tensor_bytes = self.file.read(offset_end - offset_start)
return self._deserialize_tensor(tensor_bytes, metadata)
def _read_header(self):
header_size = struct.unpack("<Q", self.file.read(8))[0]
header_json = self.file.read(header_size).decode("utf-8")
return json.loads(header_json), header_size
def _deserialize_tensor(self, tensor_bytes, metadata):
dtype_map = {
"F32": torch.float32, "F16": torch.float16, "BF16": torch.bfloat16,
"I64": torch.int64, "I32": torch.int32, "I16": torch.int16, "I8": torch.int8,
"U8": torch.uint8, "BOOL": torch.bool
}
dtype = dtype_map[metadata["dtype"]]
shape = metadata["shape"]
return torch.frombuffer(tensor_bytes, dtype=torch.uint8).view(dtype).reshape(shape)
# --- Constants & Setup ---
try:
TempDir = Path("/tmp/temp_tool")
os.makedirs(TempDir, exist_ok=True)
except:
TempDir = Path("./temp_tool")
os.makedirs(TempDir, exist_ok=True)
api = HfApi()
def cleanup_temp():
if TempDir.exists():
shutil.rmtree(TempDir)
os.makedirs(TempDir, exist_ok=True)
gc.collect()
def download_file(input_path, token, filename=None):
local_path = TempDir / (filename if filename else "model.safetensors")
if input_path.startswith("http"):
print(f"Downloading {filename} from URL...")
try:
response = requests.get(input_path, stream=True, timeout=30)
response.raise_for_status()
with open(local_path, 'wb') as f:
for chunk in response.iter_content(chunk_size=8192):
f.write(chunk)
except Exception as e: raise ValueError(f"Download failed: {e}")
else:
print(f"Downloading {filename} from Hub...")
if not filename:
try:
files = list_repo_files(repo_id=input_path, token=token)
safetensors = [f for f in files if f.endswith(".safetensors")]
filename = safetensors[0] if safetensors else "adapter_model.safetensors"
except: filename = "adapter_model.safetensors"
try:
hf_hub_download(repo_id=input_path, filename=filename, token=token, local_dir=TempDir, local_dir_use_symlinks=False)
if not (TempDir / filename).exists():
found = list(TempDir.rglob(filename))
if found: shutil.move(found[0], local_path)
except Exception as e: raise ValueError(f"Hub download failed: {e}")
return local_path
def get_key_stem(key):
key = key.replace(".weight", "").replace(".bias", "")
key = key.replace(".lora_down", "").replace(".lora_up", "")
key = key.replace(".lora_A", "").replace(".lora_B", "")
key = key.replace(".alpha", "")
prefixes = [
"model.diffusion_model.", "diffusion_model.", "model.",
"transformer.", "text_encoder.", "lora_unet_", "lora_te_", "base_model.model."
]
changed = True
while changed:
changed = False
for p in prefixes:
if key.startswith(p):
key = key[len(p):]
changed = True
return key
# =================================================================================
# TAB 1: MERGE & RESHARD (Fixes Folder Structure & Aux Files)
# =================================================================================
def load_lora_to_memory(lora_path, precision_dtype=torch.bfloat16):
print(f"Loading LoRA from {lora_path}...")
state_dict = load_file(lora_path, device="cpu")
pairs = {}
alphas = {}
for k, v in state_dict.items():
stem = get_key_stem(k)
if "alpha" in k:
alphas[stem] = v.item() if isinstance(v, torch.Tensor) else v
else:
if stem not in pairs: pairs[stem] = {}
if "lora_down" in k or "lora_A" in k:
pairs[stem]["down"] = v.to(dtype=precision_dtype)
pairs[stem]["rank"] = v.shape[0]
elif "lora_up" in k or "lora_B" in k:
pairs[stem]["up"] = v.to(dtype=precision_dtype)
for stem in pairs:
pairs[stem]["alpha"] = alphas.get(stem, float(pairs[stem].get("rank", 1.0)))
return pairs
class ShardBuffer:
def __init__(self, max_size_gb, output_dir, output_repo, subfolder, hf_token, filename_prefix="model"):
self.max_bytes = int(max_size_gb * 1024**3)
self.output_dir = output_dir
self.output_repo = output_repo
self.subfolder = subfolder
self.hf_token = hf_token
self.filename_prefix = filename_prefix
self.buffer = []
self.current_bytes = 0
self.shard_count = 0
self.index_map = {}
self.total_size = 0 # Accumulates total model size for index.json
def add_tensor(self, key, tensor):
# Determine bytes for size calculation and storage
if tensor.dtype == torch.bfloat16:
raw_bytes = tensor.view(torch.int16).numpy().tobytes()
dtype_str = "BF16"
elif tensor.dtype == torch.float16:
raw_bytes = tensor.numpy().tobytes()
dtype_str = "F16"
else:
raw_bytes = tensor.numpy().tobytes()
dtype_str = "F32"
size = len(raw_bytes)
self.buffer.append({
"key": key,
"data": raw_bytes,
"dtype": dtype_str,
"shape": tensor.shape
})
self.current_bytes += size
self.total_size += size # Explicitly increment total size
if self.current_bytes >= self.max_bytes:
self.flush()
def flush(self):
if not self.buffer: return
self.shard_count += 1
# Naming: prefix-0000X.safetensors
# This is standard for indexed loading.
filename = f"{self.filename_prefix}-{self.shard_count:05d}.safetensors"
# Proper Subfolder Handling
path_in_repo = f"{self.subfolder}/{filename}" if self.subfolder else filename
print(f"Flushing {path_in_repo} ({self.current_bytes / 1024**3:.2f} GB)...")
header = {"__metadata__": {"format": "pt"}}
current_offset = 0
for item in self.buffer:
header[item["key"]] = {
"dtype": item["dtype"],
"shape": item["shape"],
"data_offsets": [current_offset, current_offset + len(item["data"])]
}
current_offset += len(item["data"])
self.index_map[item["key"]] = filename # Relative filename for index
header_json = json.dumps(header).encode('utf-8')
out_path = self.output_dir / filename
with open(out_path, 'wb') as f:
f.write(struct.pack('<Q', len(header_json)))
f.write(header_json)
for item in self.buffer:
f.write(item["data"])
print(f"Uploading {path_in_repo}...")
api.upload_file(path_or_fileobj=out_path, path_in_repo=path_in_repo, repo_id=self.output_repo, token=self.hf_token)
os.remove(out_path)
self.buffer = []
self.current_bytes = 0
gc.collect()
# =================================================================================
# ROBUST RESHARDING LOGIC (Plan -> Execute)
# =================================================================================
def download_lora_smart(input_str, token):
"""Robust LoRA downloader that handles Direct URLs and Repo IDs."""
local_path = TempDir / "adapter.safetensors"
if local_path.exists(): os.remove(local_path)
# 1. Try as Direct URL
if input_str.startswith("http"):
print(f"Downloading LoRA from URL: {input_str}")
headers = {"Authorization": f"Bearer {token}"} if token else {}
try:
response = requests.get(input_str, stream=True, headers=headers, timeout=60)
response.raise_for_status()
with open(local_path, 'wb') as f:
for chunk in response.iter_content(chunk_size=8192):
f.write(chunk)
if verify_safetensors(local_path): return local_path
except Exception as e:
print(f"URL download failed: {e}. Trying as Repo ID...")
# 2. Try as Repo ID
print(f"Attempting download from Hub Repo: {input_str}")
try:
# Check if user provided a filename in the repo string (e.g. user/repo/file.safetensors)
if ".safetensors" in input_str and "/" in input_str:
# splitting repo_id and filename might be needed, but hf_hub_download expects valid repo_id
pass
# Try to find the adapter file automatically
files = list_repo_files(repo_id=input_str, token=token)
candidates = ["adapter_model.safetensors", "model.safetensors"]
target = next((f for f in files if f in candidates), None)
# If no standard name, take the first safetensors found
if not target:
safes = [f for f in files if f.endswith(".safetensors")]
if safes: target = safes[0]
if not target: raise ValueError("No .safetensors found")
hf_hub_download(repo_id=input_str, filename=target, token=token, local_dir=TempDir)
# Move to standard location
downloaded = TempDir / target
if downloaded != local_path:
shutil.move(downloaded, local_path)
return local_path
except Exception as e:
raise ValueError(f"Could not download LoRA. Checked URL and Repo. Error: {e}")
def get_tensor_byte_size(shape, dtype_str):
"""Calculates byte size of a tensor based on shape and dtype."""
# F32=4, F16/BF16=2, I8=1, etc.
bytes_per = 4 if "F32" in dtype_str else 2 if "16" in dtype_str else 1
numel = 1
for d in shape: numel *= d
return numel * bytes_per
def plan_resharding(input_shards, max_shard_size_gb, filename_prefix):
"""
Pass 1: Reads headers ONLY. Groups tensors into virtual shards of max_shard_size_gb.
Returns a Plan (List of ShardDefinitions).
"""
print(f"Planning resharding (Max {max_shard_size_gb} GB)...")
max_bytes = int(max_shard_size_gb * 1024**3)
all_tensors = []
# 1. Scan all inputs
for p in input_shards:
with MemoryEfficientSafeOpen(p) as f:
for k in f.keys():
shape = f.header[k]['shape']
dtype = f.header[k]['dtype']
size = get_tensor_byte_size(shape, dtype)
all_tensors.append({
"key": k,
"shape": shape,
"dtype": dtype,
"size": size,
"source": p
})
# 2. Sort tensors (Crucial for deterministic output)
all_tensors.sort(key=lambda x: x["key"])
# 3. Bucket into Shards
plan = []
current_shard = []
current_size = 0
for t in all_tensors:
# If adding this tensor exceeds limit AND we have stuff in the bucket, close bucket
if current_size + t['size'] > max_bytes and current_shard:
plan.append(current_shard)
current_shard = []
current_size = 0
current_shard.append(t)
current_size += t['size']
if current_shard:
plan.append(current_shard)
total_shards = len(plan)
total_model_size = sum(t['size'] for shard in plan for t in shard)
print(f"Plan created: {total_shards} shards. Total size: {total_model_size / 1024**3:.2f} GB")
# 4. Format Plan
final_plan = []
for i, shard_tensors in enumerate(plan):
# Naming: prefix-00001-of-00005.safetensors
name = f"{filename_prefix}-{i+1:05d}-of-{total_shards:05d}.safetensors"
final_plan.append({
"filename": name,
"tensors": shard_tensors
})
return final_plan, total_model_size
def copy_auxiliary_configs(hf_token, base_repo, base_subfolder, output_repo, output_subfolder):
"""
Downloads NON-WEIGHT files (json, txt, model) from Base Repo and uploads to Output.
"""
print(f"Copying config files from {base_repo}...")
try:
files = list_repo_files(repo_id=base_repo, token=hf_token)
# Extensions to KEEP (Configs, Tokenizers, etc.)
allowed_ext = ['.json', '.txt', '.model', '.py', '.yml', '.yaml']
# Extensions to SKIP (Weights, we are generating these)
blocked_ext = ['.safetensors', '.bin', '.pt', '.pth', '.msgpack', '.h5']
for f in files:
# Filter by subfolder if needed
if base_subfolder and not f.startswith(base_subfolder):
continue
ext = os.path.splitext(f)[1]
if ext in blocked_ext: continue
if ext not in allowed_ext: continue # Skip unknown types to be safe? Or allow?
# Download
print(f"Transferring {f}...")
local = hf_hub_download(repo_id=base_repo, filename=f, token=hf_token, local_dir=TempDir)
# Determine path in new repo
if base_subfolder:
# Remove base_subfolder prefix for the rel path
rel_name = f[len(base_subfolder):].lstrip('/')
else:
rel_name = f
# Add output_subfolder prefix
target_path = f"{output_subfolder}/{rel_name}" if output_subfolder else rel_name
api.upload_file(path_or_fileobj=local, path_in_repo=target_path, repo_id=output_repo, token=hf_token)
os.remove(local)
except Exception as e:
print(f"Config copy warning: {e}")
def task_merge(hf_token, base_repo, base_subfolder, lora_input, scale, precision, shard_size, output_repo, structure_repo, private, progress=gr.Progress()):
cleanup_temp()
if not hf_token: return "Error: Token missing."
login(hf_token)
# 1. Output Setup
try:
api.create_repo(repo_id=output_repo, private=private, exist_ok=True, token=hf_token)
except Exception as e: return f"Error creating repo: {e}"
# Determine Folder Logic
# If base_subfolder is "qint4", and we want output to be "transformer", user needs to specify that.
# But usually, if base has a subfolder, we maintain a subfolder structure.
# ADAPTIVE: If base_subfolder is "qint4", we treat it as the source of weights.
# Since you merged into "transformer", I assume you want the output in "transformer".
# For general LLMs (root), both are empty.
# Heuristic: If base has subfolder, use "transformer" as target if it looks like a DiT, else keep original name.
if base_subfolder:
output_subfolder = "transformer" if "qint" in base_subfolder or "transformer" in base_subfolder else base_subfolder
else:
output_subfolder = ""
# 2. Copy Configs (The missing step from previous run)
copy_auxiliary_configs(hf_token, base_repo, base_subfolder, output_repo, output_subfolder)
# 3. Structure Repo (Only needed if Base doesn't have everything, e.g. VAE)
if structure_repo:
print(f"Copying extras from {structure_repo}...")
# We assume structure repo is a standard diffusers repo
# We copy text_encoder, vae, scheduler, tokenizer, etc.
# We SKIP 'transformer' or 'unet' because we are building that.
streaming_copy_structure(hf_token, structure_repo, output_repo, ignore_prefix="transformer")
# 4. Download ALL Input Shards (Needed for Planning)
progress(0.1, desc="Downloading Input Model...")
files = list_repo_files(repo_id=base_repo, token=hf_token)
input_shards = []
for f in files:
if f.endswith(".safetensors"):
if base_subfolder and not f.startswith(base_subfolder): continue
local = TempDir / "inputs" / os.path.basename(f)
os.makedirs(local.parent, exist_ok=True)
hf_hub_download(repo_id=base_repo, filename=f, token=hf_token, local_dir=local.parent, local_dir_use_symlinks=False)
# Handle nesting
found = list(local.parent.rglob(os.path.basename(f)))
if found: input_shards.append(found[0])
if not input_shards: return "No safetensors found."
input_shards.sort()
# 5. Detect Naming Convention (Adaptive)
sample_name = os.path.basename(input_shards[0])
if "diffusion_pytorch_model" in sample_name or output_subfolder == "transformer":
prefix = "diffusion_pytorch_model"
index_file = "diffusion_pytorch_model.safetensors.index.json"
else:
prefix = "model"
index_file = "model.safetensors.index.json"
# 6. Create Plan (Pass 1)
# This calculates total shards and size BEFORE processing
progress(0.2, desc="Planning Shards...")
plan, total_model_size = plan_resharding(input_shards, shard_size, prefix)
# 7. Load LoRA
dtype = torch.bfloat16 if precision == "bf16" else torch.float16 if precision == "fp16" else torch.float32
try:
progress(0.25, desc="Loading LoRA...")
lora_path = download_lora_smart(lora_input, hf_token)
lora_pairs = load_lora_to_memory(lora_path, precision_dtype=dtype)
except Exception as e: return f"LoRA Error: {e}"
# 8. Execute Plan (Pass 2)
index_map = {}
for i, shard_plan in enumerate(plan):
filename = shard_plan['filename']
tensors_to_write = shard_plan['tensors']
progress(0.3 + (0.7 * i / len(plan)), desc=f"Merging {filename}")
print(f"Generating {filename} ({len(tensors_to_write)} tensors)...")
# Prepare Header
header = {"__metadata__": {"format": "pt"}}
current_offset = 0
for t in tensors_to_write:
# Recalculate dtype string for header based on TARGET dtype
tgt_dtype_str = "BF16" if dtype == torch.bfloat16 else "F16" if dtype == torch.float16 else "F32"
# Calculate output size (might differ from input size if we change precision)
# Input size in plan was source size. We need target size.
out_size = get_tensor_byte_size(t['shape'], tgt_dtype_str)
header[t['key']] = {
"dtype": tgt_dtype_str,
"shape": t['shape'],
"data_offsets": [current_offset, current_offset + out_size]
}
current_offset += out_size
index_map[t['key']] = filename
header_json = json.dumps(header).encode('utf-8')
out_path = TempDir / filename
with open(out_path, 'wb') as f_out:
f_out.write(struct.pack('<Q', len(header_json)))
f_out.write(header_json)
# Open source files as needed
open_files = {}
for t_plan in tqdm(tensors_to_write, leave=False):
src = t_plan['source']
if src not in open_files: open_files[src] = MemoryEfficientSafeOpen(src)
# Load Tensor
v = open_files[src].get_tensor(t_plan['key'])
k = t_plan['key']
# --- MERGE LOGIC ---
base_stem = get_key_stem(k)
match = None
# Check match (Same logic as before)
if base_stem in lora_pairs: match = lora_pairs[base_stem]
# ... [QKV Logic omitted for brevity, same as previous] ...
if not match:
if "to_q" in base_stem:
qkv = base_stem.replace("to_q", "qkv")
if qkv in lora_pairs: match = lora_pairs[qkv]
elif "to_k" in base_stem:
qkv = base_stem.replace("to_k", "qkv")
if qkv in lora_pairs: match = lora_pairs[qkv]
elif "to_v" in base_stem:
qkv = base_stem.replace("to_v", "qkv")
if qkv in lora_pairs: match = lora_pairs[qkv]
if match:
down = match["down"]
up = match["up"]
# ... [Matmul Logic, same as previous] ...
scaling = scale * (match["alpha"] / match["rank"])
if len(v.shape) == 4 and len(down.shape) == 2:
down = down.unsqueeze(-1).unsqueeze(-1)
up = up.unsqueeze(-1).unsqueeze(-1)
try:
if len(up.shape) == 4:
delta = (up.squeeze() @ down.squeeze()).reshape(up.shape[0], down.shape[1], 1, 1)
else:
delta = up @ down
except: delta = up.T @ down
delta = delta * scaling
# Slicing
valid = True
if delta.shape == v.shape: pass
elif delta.shape[0] == v.shape[0] * 3:
chunk = v.shape[0]
if "to_q" in k: delta = delta[0:chunk, ...]
elif "to_k" in k: delta = delta[chunk:2*chunk, ...]
elif "to_v" in k: delta = delta[2*chunk:, ...]
else: valid = False
elif delta.numel() == v.numel(): delta = delta.reshape(v.shape)
else: valid = False
if valid:
v = v.to(dtype)
delta = delta.to(dtype)
v.add_(delta)
del delta
# --- END MERGE ---
# Write
if v.dtype != dtype: v = v.to(dtype)
if dtype == torch.bfloat16:
raw = v.view(torch.int16).numpy().tobytes()
else:
raw = v.numpy().tobytes()
f_out.write(raw)
del v
# Close handles
for fh in open_files.values(): fh.file.close()
# Upload Shard
path_in_repo = f"{output_subfolder}/{filename}" if output_subfolder else filename
api.upload_file(path_or_fileobj=out_path, path_in_repo=path_in_repo, repo_id=output_repo, token=hf_token)
os.remove(out_path)
gc.collect()
# 9. Upload Index
# Update total size to reflect the TARGET dtype size, not source
# We recalculate total_size based on what we actually wrote
final_total_size = 0
for t_list in plan:
for t in t_list['tensors']:
tgt_dtype_str = "BF16" if dtype == torch.bfloat16 else "F16" if dtype == torch.float16 else "F32"
final_total_size += get_tensor_byte_size(t['shape'], tgt_dtype_str)
index_data = {"metadata": {"total_size": final_total_size}, "weight_map": index_map}
with open(TempDir / index_file, "w") as f:
json.dump(index_data, f, indent=4)
path_in_repo = f"{output_subfolder}/{index_file}" if output_subfolder else index_file
api.upload_file(path_or_fileobj=TempDir / index_file, path_in_repo=path_in_repo, repo_id=output_repo, token=hf_token)
cleanup_temp()
return f"Success! {len(plan)} shards created at {output_repo}"
# =================================================================================
# TAB 2: EXTRACT LORA
# =================================================================================
def extract_lora_layer_by_layer(model_org, model_tuned, rank, clamp):
org = MemoryEfficientSafeOpen(model_org)
tuned = MemoryEfficientSafeOpen(model_tuned)
lora_sd = {}
print("Calculating diffs...")
for key in tqdm(org.keys()):
if key not in tuned.keys(): continue
mat_org = org.get_tensor(key).float()
mat_tuned = tuned.get_tensor(key).float()
diff = mat_tuned - mat_org
if torch.max(torch.abs(diff)) < 1e-4: continue
out_dim, in_dim = diff.shape[:2]
r = min(rank, in_dim, out_dim)
is_conv = len(diff.shape) == 4
if is_conv: diff = diff.flatten(start_dim=1)
try:
U, S, Vh = torch.linalg.svd(diff, full_matrices=False)
U, S, Vh = U[:, :r], S[:r], Vh[:r, :]
U = U @ torch.diag(S)
dist = torch.cat([U.flatten(), Vh.flatten()])
hi_val = torch.quantile(dist, clamp)
U = U.clamp(-hi_val, hi_val)
Vh = Vh.clamp(-hi_val, hi_val)
if is_conv:
U = U.reshape(out_dim, r, 1, 1)
Vh = Vh.reshape(r, in_dim, mat_org.shape[2], mat_org.shape[3])
else:
U = U.reshape(out_dim, r)
Vh = Vh.reshape(r, in_dim)
stem = key.replace(".weight", "")
lora_sd[f"{stem}.lora_up.weight"] = U
lora_sd[f"{stem}.lora_down.weight"] = Vh
lora_sd[f"{stem}.alpha"] = torch.tensor(r).float()
except: pass
out = TempDir / "extracted.safetensors"
save_file(lora_sd, out)
return str(out)
def task_extract(hf_token, org, tun, rank, out):
cleanup_temp()
login(hf_token)
try:
p1 = download_file(org, hf_token, filename="org.safetensors")
p2 = download_file(tun, hf_token, filename="tun.safetensors")
f = extract_lora_layer_by_layer(p1, p2, int(rank), 0.99)
api.create_repo(repo_id=out, exist_ok=True, token=hf_token)
api.upload_file(path_or_fileobj=f, path_in_repo="extracted.safetensors", repo_id=out, token=hf_token)
return "Done"
except Exception as e: return f"Error: {e}"
# =================================================================================
# TAB 3: MERGE ADAPTERS (EMA) with Sigma Rel
# =================================================================================
def sigma_rel_to_gamma(sigma_rel):
t = sigma_rel**-2
coeffs = [1, 7, 16 - t, 12 - t]
roots = np.roots(coeffs)
gamma = roots[np.isreal(roots) & (roots.real >= 0)].real.max()
return gamma
def task_merge_adapters(hf_token, lora_urls, beta, sigma_rel, out_repo):
cleanup_temp()
login(hf_token)
urls = [u.strip() for u in lora_urls.split(",") if u.strip()]
paths = []
try:
for i, url in enumerate(urls):
paths.append(download_file(url, hf_token, filename=f"a_{i}.safetensors"))
except Exception as e: return f"Download Error: {e}"
if not paths: return "No models found"
base_sd = load_file(paths[0], device="cpu")
for k in base_sd:
if base_sd[k].dtype.is_floating_point: base_sd[k] = base_sd[k].float()
gamma = None
if sigma_rel > 0:
gamma = sigma_rel_to_gamma(sigma_rel)
for i, path in enumerate(paths[1:]):
print(f"Merging {path}")
if gamma is not None:
t = i + 1
current_beta = (1 - 1 / t) ** (gamma + 1)
else:
current_beta = beta
curr = load_file(path, device="cpu")
for k in base_sd:
if k in curr and "alpha" not in k:
base_sd[k] = base_sd[k] * current_beta + curr[k].float() * (1 - current_beta)
out = TempDir / "merged_adapters.safetensors"
save_file(base_sd, out)
api.create_repo(repo_id=out_repo, exist_ok=True, token=hf_token)
api.upload_file(path_or_fileobj=out, path_in_repo="merged_adapters.safetensors", repo_id=out_repo, token=hf_token)
return "Done"
# =================================================================================
# TAB 4: RESIZE
# =================================================================================
def index_sv_ratio(S, target):
max_sv = S[0]
min_sv = max_sv / target
index = int(torch.sum(S > min_sv).item())
return max(1, min(index, len(S) - 1))
def task_resize(hf_token, lora_input, new_rank, dynamic_method, dynamic_param, out_repo):
cleanup_temp()
login(hf_token)
try:
path = download_file(lora_input, hf_token)
except Exception as e: return f"Error: {e}"
state = load_file(path, device="cpu")
new_state = {}
groups = {}
for k in state:
stem = get_key_stem(k)
simple = k.split(".lora_")[0]
if simple not in groups: groups[simple] = {}
if "lora_down" in k or "lora_A" in k: groups[simple]["down"] = state[k]
if "lora_up" in k or "lora_B" in k: groups[simple]["up"] = state[k]
if "alpha" in k: groups[simple]["alpha"] = state[k]
for stem, g in tqdm(groups.items()):
if "down" in g and "up" in g:
down, up = g["down"].float(), g["up"].float()
if len(down.shape) == 4:
merged = (up.squeeze() @ down.squeeze()).reshape(up.shape[0], down.shape[1], down.shape[2], down.shape[3])
flat = merged.flatten(1)
else:
merged = up @ down
flat = merged
U, S, Vh = torch.linalg.svd(flat, full_matrices=False)
target_rank = int(new_rank)
if dynamic_method == "sv_ratio":
target_rank = index_sv_ratio(S, dynamic_param)
target_rank = min(target_rank, S.shape[0])
U = U[:, :target_rank]
S = S[:target_rank]
U = U @ torch.diag(S)
Vh = Vh[:target_rank, :]
if len(down.shape) == 4:
U = U.reshape(up.shape[0], target_rank, 1, 1)
Vh = Vh.reshape(target_rank, down.shape[1], down.shape[2], down.shape[3])
new_state[f"{stem}.lora_down.weight"] = Vh
new_state[f"{stem}.lora_up.weight"] = U
new_state[f"{stem}.alpha"] = torch.tensor(target_rank).float()
out = TempDir / "resized.safetensors"
save_file(new_state, out)
api.create_repo(repo_id=out_repo, exist_ok=True, token=hf_token)
api.upload_file(path_or_fileobj=out, path_in_repo="resized.safetensors", repo_id=out_repo, token=hf_token)
return "Done"
# =================================================================================
# UI
# =================================================================================
css = ".container { max-width: 900px; margin: auto; }"
with gr.Blocks() as demo:
gr.Markdown("# 🧰SOONmerge® LoRA Toolkit")
with gr.Tabs():
with gr.Tab("Merge to Base + Reshard Output"):
t1_token = gr.Textbox(label="Token", type="password")
t1_base = gr.Textbox(label="Base Repo (Diffusers)", value="ostris/Z-Image-De-Turbo")
t1_sub = gr.Textbox(label="Subfolder", value="transformer")
t1_lora = gr.Textbox(label="LoRA Repo as (name/repo)", value="GuangyuanSD/Z-Image-Re-Turbo-LoRA")
with gr.Row():
t1_scale = gr.Slider(label="Scale", value=1.0, minimum=0, maximum=3.0, step=0.1)
t1_prec = gr.Radio(["bf16", "fp16", "float32"], value="bf16", label="Precision")
t1_shard = gr.Slider(label="Shard Size (GB)", value=2.0, minimum=0.1, maximum=10.0, step=0.1)
t1_out = gr.Textbox(label="Output Repo")
t1_struct = gr.Textbox(label="Diffusers Extras (Copies VAE/TextEnc/etc)", value="Tongyi-MAI/Z-Image-Turbo")
t1_priv = gr.Checkbox(label="Private", value=True)
t1_btn = gr.Button("Merge")
t1_res = gr.Textbox(label="Result")
t1_btn.click(task_merge, [t1_token, t1_base, t1_sub, t1_lora, t1_scale, t1_prec, t1_shard, t1_out, t1_struct, t1_priv], t1_res)
with gr.Tab("Extract Adapter"):
t2_token = gr.Textbox(label="Token", type="password")
t2_org = gr.Textbox(label="Original Model")
t2_tun = gr.Textbox(label="Tuned Model")
t2_rank = gr.Number(label="Extract At Rank", value=32, minimum=1, maximum=1024, step=1)
t2_out = gr.Textbox(label="Output Repo")
t2_btn = gr.Button("Extract")
t2_res = gr.Textbox(label="Result")
t2_btn.click(task_extract, [t2_token, t2_org, t2_tun, t2_rank, t2_out], t2_res)
with gr.Tab("Merge Multiple Adapters"):
t3_token = gr.Textbox(label="Token", type="password")
t3_urls = gr.Textbox(label="URLs")
with gr.Row():
t3_beta = gr.Slider(label="Beta", value=0.95, minimum=0.01, maximum=1.00, step=0.01)
t3_sigma = gr.Slider(label="Sigma Rel (Overrides Beta)", value=0.21, minimum=0.01, maximum=1.00, step=0.01)
t3_out = gr.Textbox(label="Output Repo")
t3_btn = gr.Button("Merge")
t3_res = gr.Textbox(label="Result")
t3_btn.click(task_merge_adapters, [t3_token, t3_urls, t3_beta, t3_sigma, t3_out], t3_res)
with gr.Tab("Resize Adapter"):
t4_token = gr.Textbox(label="Token", type="password")
t4_in = gr.Textbox(label="LoRA")
with gr.Row():
t4_rank = gr.Number(label="To Rank (Lower Only!)", value=8, minimum=1, maximum=256, step=1)
t4_method = gr.Dropdown(["None", "sv_ratio"], value="None", label="Dynamic Method")
t4_param = gr.Number(label="Dynamic Param", value=4.0)
t4_out = gr.Textbox(label="Output")
t4_btn = gr.Button("Resize")
t4_res = gr.Textbox(label="Result")
t4_btn.click(task_resize, [t4_token, t4_in, t4_rank, t4_method, t4_param, t4_out], t4_res)
if __name__ == "__main__":
demo.queue().launch(css=css, ssr_mode=False) |