File size: 60,220 Bytes
07b428c f32b3c6 07b428c 7803d72 07b428c 7803d72 07b428c 7803d72 07b428c f32b3c6 07b428c f32b3c6 07b428c 7803d72 07b428c 7803d72 07b428c f32b3c6 07b428c 44e6b86 07b428c ae8c38d 07b428c f32b3c6 07b428c dc3b370 ae8c38d 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 07b428c f32b3c6 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 | #!/usr/bin/env python3
"""
HexState GGUF Re-Quantizer β GGUF-to-GGUF Q2_K quantization.
Reads a source GGUF (F16/BF16/F32), copies all metadata verbatim,
and re-quantizes eligible weight tensors to Q2_K using numpy.
This bypasses the tokenizer parsing problem entirely β the source GGUF
(from llama.cpp's convert_hf_to_gguf.py) has correct metadata.
Usage:
python3 hexstate_requantize.py input.gguf output.gguf
"""
import struct
import sys
import time
import os
import io
import ctypes
import numpy as np
# βββ HExState C Library (HPC-optimized Q2_K quantization) ββββββββββββββββββ
_HEXSTATE_LIB = None
def _load_hexstate_lib():
"""Try to load the HExState C shared library for HPC-optimized quantization."""
global _HEXSTATE_LIB
if _HEXSTATE_LIB is not None:
return _HEXSTATE_LIB
lib_dir = os.path.dirname(os.path.abspath(__file__))
lib_path = os.path.join(lib_dir, "libhexstate_q2k.so")
if not os.path.exists(lib_path):
return None
try:
lib = ctypes.CDLL(lib_path)
# void hexstate_init(void)
lib.hexstate_init.restype = None
lib.hexstate_init.argtypes = []
# void hexstate_quantize_tensor_q2k(const float*, int64_t, void*, float*, int, int)
lib.hexstate_quantize_tensor_q2k.restype = None
lib.hexstate_quantize_tensor_q2k.argtypes = [
ctypes.POINTER(ctypes.c_float), # weights
ctypes.c_int64, # n_elements
ctypes.c_void_p, # output
ctypes.POINTER(ctypes.c_float), # out_error
ctypes.c_int, # opt_mode (0=HPC, 1=MSE, 2=Hybrid)
ctypes.c_int, # verbose
]
lib.hexstate_q2k_block_bytes.restype = ctypes.c_int
lib.hexstate_q2k_block_bytes.argtypes = []
lib.hexstate_q2k_block_elements.restype = ctypes.c_int
lib.hexstate_q2k_block_elements.argtypes = []
# imatrix-aware version
lib.hexstate_quantize_tensor_q2k_imat.restype = None
lib.hexstate_quantize_tensor_q2k_imat.argtypes = [
ctypes.POINTER(ctypes.c_float), # weights
ctypes.c_int64, # n_elements
ctypes.c_void_p, # output
ctypes.POINTER(ctypes.c_float), # out_error
ctypes.c_int, # opt_mode
ctypes.POINTER(ctypes.c_float), # imat_importance (can be NULL)
ctypes.c_int, # verbose
]
# Q8_0 HPC quantizer (Shor pipeline; tied embeddings / LM head)
if hasattr(lib, 'hexstate_quantize_tensor_q8_0_hpc'):
lib.hexstate_quantize_tensor_q8_0_hpc.restype = None
lib.hexstate_quantize_tensor_q8_0_hpc.argtypes = [
ctypes.POINTER(ctypes.c_float), # weights
ctypes.c_int64, # n_elements
ctypes.c_void_p, # output
ctypes.POINTER(ctypes.c_float), # out_error
ctypes.POINTER(ctypes.c_float), # imat_importance (can be NULL)
ctypes.c_int, # verbose
]
# Q4_0 HPC quantizer (for attention tensors)
if hasattr(lib, 'hexstate_quantize_tensor_q4_0_hpc'):
lib.hexstate_quantize_tensor_q4_0_hpc.restype = None
lib.hexstate_quantize_tensor_q4_0_hpc.argtypes = [
ctypes.POINTER(ctypes.c_float), # weights
ctypes.c_int64, # n_elements
ctypes.c_void_p, # output
ctypes.POINTER(ctypes.c_float), # out_error
ctypes.POINTER(ctypes.c_float), # imat_importance (can be NULL)
ctypes.c_int, # verbose
]
lib.hexstate_init()
_HEXSTATE_LIB = lib
return lib
except Exception as e:
print(f" WARNING: Failed to load HexState library: {e}")
return None
def _skip_gguf_kv_value(f, vtype):
"""Skip a GGUF KV value of the given type."""
import struct as st
size_map = {0:1, 1:1, 2:2, 3:2, 4:4, 5:4, 6:4, 7:1, 10:8, 11:8, 12:8}
if vtype == 8: # string
slen = st.unpack('<Q', f.read(8))[0]
f.read(slen)
elif vtype == 9: # array
arr_type = st.unpack('<I', f.read(4))[0]
arr_len = st.unpack('<Q', f.read(8))[0]
if arr_type == 8: # array of strings
for _ in range(arr_len):
slen = st.unpack('<Q', f.read(8))[0]
f.read(slen)
else:
sz = size_map.get(arr_type, 4)
f.read(arr_len * sz)
else:
sz = size_map.get(vtype, 4)
f.read(sz)
def read_imatrix(path):
"""Read llama.cpp importance matrix file (GGUF or legacy .dat format).
Returns dict: tensor_name -> normalized importance array (float32)
"""
import struct as st
imat = {}
with open(path, 'rb') as f:
magic = st.unpack('<I', f.read(4))[0]
if magic == 0x46554747: # GGUF format (modern llama.cpp)
_ver = st.unpack('<I', f.read(4))[0]
n_tensors = st.unpack('<Q', f.read(8))[0]
n_kv = st.unpack('<Q', f.read(8))[0]
# Skip KV pairs
for _ in range(n_kv):
slen = st.unpack('<Q', f.read(8))[0]
f.read(slen) # key
vtype = st.unpack('<I', f.read(4))[0]
_skip_gguf_kv_value(f, vtype)
# Read tensor infos
tensor_infos = []
for _ in range(n_tensors):
slen = st.unpack('<Q', f.read(8))[0]
name = f.read(slen).decode('utf-8', errors='replace')
n_dims = st.unpack('<I', f.read(4))[0]
dims = [st.unpack('<Q', f.read(8))[0] for _ in range(n_dims)]
ttype = st.unpack('<I', f.read(4))[0]
offset = st.unpack('<Q', f.read(8))[0]
n_el = 1
for d in dims:
n_el *= d
tensor_infos.append((name, n_el, offset))
# Data section start (32-byte aligned)
data_start = ((f.tell() + 31) // 32) * 32
# Group by base tensor name: collect in_sum2 and counts
sum2_data = {}
counts_data = {}
for name, n_el, offset in tensor_infos:
f.seek(data_start + offset)
data = np.frombuffer(f.read(n_el * 4), dtype=np.float32).copy()
if name.endswith('.in_sum2'):
base = name[:-len('.in_sum2')]
sum2_data[base] = data
elif name.endswith('.counts'):
base = name[:-len('.counts')]
counts_data[base] = data
# Compute normalized importance: sqrt(in_sum2 / counts) / mean
for base_name in sum2_data:
in_sum2 = sum2_data[base_name]
count = counts_data.get(base_name, np.array([1.0]))[0]
if count > 0:
importance = np.sqrt(in_sum2 / count)
else:
importance = np.ones_like(in_sum2)
mean = importance.mean()
if mean > 1e-30:
imat[base_name] = importance / mean
else:
imat[base_name] = np.ones_like(importance)
else:
# Legacy format: first 4 bytes were n_entries
f.seek(0)
n_entries = st.unpack('<i', f.read(4))[0]
for _ in range(n_entries):
name_len = st.unpack('<i', f.read(4))[0]
name = f.read(name_len).decode('utf-8')
n_values = st.unpack('<i', f.read(4))[0]
n_samples = st.unpack('<i', f.read(4))[0]
values = np.frombuffer(f.read(n_values * 4), dtype=np.float32).copy()
mean = values.mean()
if mean > 1e-30:
imat[name] = values / mean
else:
imat[name] = np.ones_like(values)
return imat
def quantize_tensor_q2k_hpc(f32_data, opt_mode=2, importance=None):
"""Quantize tensor using HexState HPC-optimized C implementation.
opt_mode: 0=HPC (BP only), 1=MSE (grid search), 2=Hybrid (recommended)
importance: optional per-element importance weights (from imatrix)
Returns: (bytes, n_blocks) same as quantize_tensor_q2k()
"""
lib = _load_hexstate_lib()
if lib is None:
raise RuntimeError("HexState library not available")
n_elements = len(f32_data)
if n_elements % QK_K != 0:
pad_len = QK_K - (n_elements % QK_K)
f32_data = np.concatenate([f32_data, np.zeros(pad_len, dtype=np.float32)])
if importance is not None:
importance = np.concatenate([importance, np.ones(pad_len, dtype=np.float32)])
n_elements = len(f32_data)
n_blocks = n_elements // QK_K
block_bytes = lib.hexstate_q2k_block_bytes() # 84
# Allocate output buffer
output = np.zeros(n_blocks * block_bytes, dtype=np.uint8)
error = ctypes.c_float(0.0)
# Call C quantizer with or without importance weights
f32_contiguous = np.ascontiguousarray(f32_data, dtype=np.float32)
if importance is not None:
imat_contiguous = np.ascontiguousarray(importance, dtype=np.float32)
imat_ptr = imat_contiguous.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
else:
imat_ptr = None
lib.hexstate_quantize_tensor_q2k_imat(
f32_contiguous.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
ctypes.c_int64(n_elements),
output.ctypes.data_as(ctypes.c_void_p),
ctypes.byref(error),
ctypes.c_int(opt_mode),
imat_ptr,
ctypes.c_int(1), # verbose
)
return output.tobytes(), n_blocks
# βββ Constants ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
GGUF_MAGIC = 0x46554747
GGUF_VERSION = 3
ALIGNMENT = 32
QK_K = 256
GGML_TYPE_F32 = 0
GGML_TYPE_F16 = 1
GGML_TYPE_Q4_0 = 2
GGML_TYPE_Q8_0 = 8
GGML_TYPE_Q2_K = 10
GGML_TYPE_BF16 = 30
TYPE_NAME = {
0: "F32", 1: "F16", 2: "Q4_0", 3: "Q4_1", 6: "Q5_0", 7: "Q5_1",
8: "Q8_0", 9: "Q8_1", 10: "Q2_K", 11: "Q3_K", 12: "Q4_K",
13: "Q5_K", 14: "Q6_K", 15: "Q8_K", 30: "BF16",
}
# Block sizes and byte sizes for each type
TYPE_BLOCK_SIZE = {
0: 1, 1: 1, 2: 32, 3: 32, 6: 32, 7: 32,
8: 32, 9: 32, 10: 256, 11: 256, 12: 256,
13: 256, 14: 256, 15: 256, 30: 1,
}
TYPE_BLOCK_BYTES = {
0: 4, 1: 2, 2: 18, 3: 20, 6: 20, 7: 22,
8: 34, 9: 36, 10: 84, 11: 110, 12: 144,
13: 176, 14: 210, 15: 292, 30: 2,
}
def align_offset(offset, alignment=ALIGNMENT):
return (offset + alignment - 1) & ~(alignment - 1)
def read_string(f):
slen = struct.unpack('<Q', f.read(8))[0]
return f.read(slen).decode('utf-8', errors='replace')
def write_string(f, s):
data = s.encode('utf-8')
f.write(struct.pack('<Q', len(data)))
f.write(data)
def read_kv_value(f, vtype):
"""Read a KV value and return (vtype, raw_bytes) for passthrough."""
start = f.tell()
if vtype == 0: f.read(1) # UINT8
elif vtype == 1: f.read(1) # INT8
elif vtype == 2: f.read(2) # UINT16
elif vtype == 3: f.read(2) # INT16
elif vtype == 4: f.read(4) # UINT32
elif vtype == 5: f.read(4) # INT32
elif vtype == 6: f.read(4) # FLOAT32
elif vtype == 7: f.read(1) # BOOL
elif vtype == 8: # STRING
slen = struct.unpack('<Q', f.read(8))[0]
f.read(slen)
elif vtype == 9: # ARRAY
arr_type = struct.unpack('<I', f.read(4))[0]
arr_len = struct.unpack('<Q', f.read(8))[0]
for _ in range(arr_len):
read_kv_value(f, arr_type)
elif vtype == 10: f.read(8) # UINT64
elif vtype == 11: f.read(8) # INT64
elif vtype == 12: f.read(8) # FLOAT64
else:
raise ValueError(f"Unknown KV type {vtype}")
end = f.tell()
f.seek(start)
raw = f.read(end - start)
return raw
# βββ BF16 β F32 conversion βββββββββββββββββββββββββββββββββββββββββββββββββ
def bf16_to_f32(data_bytes, n_elements):
"""Convert BF16 raw bytes to float32 numpy array."""
bf16 = np.frombuffer(data_bytes, dtype=np.uint16)
# BF16 β F32: shift left 16 bits
f32_bits = bf16.astype(np.uint32) << 16
return f32_bits.view(np.float32)
def f16_to_f32(data_bytes, n_elements):
"""Convert F16 raw bytes to float32 numpy array."""
f16 = np.frombuffer(data_bytes, dtype=np.float16)
return f16.astype(np.float32)
def f32_to_f16(f32_array):
"""Convert float32 array to F16 bytes."""
return f32_array.astype(np.float16).tobytes()
def f32_to_bf16(f32_array):
"""Convert float32 array to BF16 bytes."""
f32_bits = f32_array.view(np.uint32)
bf16 = ((f32_bits + 0x8000) >> 16).astype(np.uint16)
return bf16.tobytes()
# βββ Q2_K quantization β faithful port of ggml quantize_row_q2_K_ref βββββββ
# Vectorized with numpy for performance. Uses make_qkx2_quants algorithm:
# - Weighted MAD error with weights[i] = |x[i]|
# - Joint scale+min least-squares solve
# - 16-step grid search for initial iscale
def quantize_tensor_q8_0(f32_data):
"""Vectorized ggml-faithful Q8_0 (fallback when the HPC lib is absent).
Block: 32 weights -> fp16 d + 32 x int8 = 34 bytes; y = q * d.
d = amax/127 (float), q = round(x/d), d stored as fp16 -- matches
ggml quantize_row_q8_0_ref. Returns (bytes, n_blocks, sse)."""
n = len(f32_data)
if n % 32 != 0:
f32_data = np.concatenate(
[f32_data, np.zeros(32 - n % 32, dtype=np.float32)])
n = len(f32_data)
blocks = f32_data.reshape(-1, 32).astype(np.float32)
nb = blocks.shape[0]
amax = np.max(np.abs(blocks), axis=1)
d = amax / 127.0
id_ = np.where(d > 0, 1.0 / np.where(d > 0, d, 1.0), 0.0)
qs = np.clip(np.rint(blocks * id_[:, None]), -127, 127).astype(np.int8)
d16 = d.astype('<f2')
out = np.zeros((nb, 34), dtype=np.uint8)
out[:, 0:2] = d16.view(np.uint8).reshape(nb, 2)
out[:, 2:] = qs.view(np.uint8)
deq = qs.astype(np.float32) * d16.astype(np.float32)[:, None]
sse = float(np.sum((blocks - deq) ** 2))
return out.tobytes(), nb, sse
def quantize_tensor_q2k(f32_data):
"""Quantize an entire tensor to Q2_K format.
Faithful vectorized port of ggml quantize_row_q2_K_ref with
make_qkx2_quants sub-block optimization.
Q2_K block layout (84 bytes, must match ggml block_q2_K):
d : fp16 super-block scale
dmin : fp16 super-block min-scale
scales[16] : packed 4-bit scale + 4-bit min per sub-block
qs[64] : interleaved 2-bit quants (4 weights 32-apart per byte)
"""
n_elements = len(f32_data)
nmax = 3
q4scale = 15.0
# Pad to QK_K (256) multiple
if n_elements % QK_K != 0:
pad_len = QK_K - (n_elements % QK_K)
f32_data = np.concatenate([f32_data, np.zeros(pad_len, dtype=np.float32)])
n_elements = len(f32_data)
n_blocks = n_elements // QK_K
# Reshape: [n_blocks, 16 sub-blocks, 16 weights]
data = f32_data.reshape(n_blocks, 16, 16).astype(np.float64)
# ββ make_qkx2_quants vectorized over all sub-blocks ββ
# Shape key: S = [n_blocks, 16], V = [n_blocks, 16, 16]
weights = np.abs(data) # [n_blocks, 16, 16]
sb_min = data.min(axis=2) # [n_blocks, 16]
sb_max = data.max(axis=2) # [n_blocks, 16]
sb_min = np.minimum(sb_min, 0.0)
# Weighted sums (needed for least-squares solve)
sum_w = weights.sum(axis=2) # [n_blocks, 16]
sum_x = (weights * data).sum(axis=2) # [n_blocks, 16]
sb_range = sb_max - sb_min
degenerate = sb_range < 1e-30 # [n_blocks, 16]
safe_range = np.maximum(sb_range, 1e-30)
# Initial quantization
iscale0 = nmax / safe_range
scale0 = 1.0 / np.maximum(iscale0, 1e-30)
shifted0 = data - sb_min[:, :, None] # [n_blocks, 16, 16]
L0 = np.clip(np.round(iscale0[:, :, None] * shifted0), 0, nmax).astype(np.float64)
# Initial error (MAD): sum(w * |scale*L + min - x|)
recon0 = scale0[:, :, None] * L0 + sb_min[:, :, None]
best_error = (weights * np.abs(recon0 - data)).sum(axis=2) # [n_blocks, 16]
best_L = L0.copy()
best_scale = scale0.copy()
best_min = sb_min.copy()
# Grid search: 16 steps (nstep=15, rmin=-0.5, rdelta=0.1)
rmin, rdelta, nstep = -0.5, 0.1, 15
for ist in range(nstep + 1):
iscale_try = (rmin + rdelta * ist + nmax) / safe_range # [n_blocks, 16]
shifted = data - sb_min[:, :, None] # use original min for quantization
Laux = np.clip(np.round(iscale_try[:, :, None] * shifted), 0, nmax).astype(np.float64)
# Weighted sums for least-squares solve
wL = weights * Laux # [n_blocks, 16, 16]
sum_l = wL.sum(axis=2) # [n_blocks, 16]
sum_l2 = (wL * Laux).sum(axis=2) # [n_blocks, 16]
sum_xl = (wL * data).sum(axis=2) # [n_blocks, 16]
# Solve 2-var system: x[i] β this_scale * L[i] + this_min
D = sum_w * sum_l2 - sum_l * sum_l
valid_D = D > 0
this_scale = np.where(valid_D,
(sum_w * sum_xl - sum_x * sum_l) / np.maximum(D, 1e-30),
0.0)
this_min = np.where(valid_D,
(sum_l2 * sum_x - sum_l * sum_xl) / np.maximum(D, 1e-30),
0.0)
# If this_min > 0, clamp to 0 and recompute scale
pos_min = this_min > 0
this_min = np.where(pos_min, 0.0, this_min)
this_scale = np.where(pos_min & (sum_l2 > 0),
sum_xl / np.maximum(sum_l2, 1e-30),
this_scale)
# Compute error for this trial
recon = this_scale[:, :, None] * Laux + this_min[:, :, None]
cur_error = (weights * np.abs(recon - data)).sum(axis=2)
# Update where this trial is better
better = valid_D & (cur_error < best_error) & ~degenerate
if better.any():
# Expand mask to weight dimension for L update
better3d = better[:, :, None]
best_L = np.where(better3d, Laux, best_L)
best_error = np.where(better, cur_error, best_error)
best_scale = np.where(better, this_scale, best_scale)
best_min = np.where(better, this_min, best_min)
# the_min = -best_min (make positive)
sb_scale = np.maximum(best_scale, 0.0).astype(np.float32) # [n_blocks, 16]
sb_the_min = np.maximum(-best_min, 0.0).astype(np.float32) # [n_blocks, 16]
# Handle degenerate sub-blocks
sb_scale[degenerate] = 0.0
sb_the_min[degenerate] = np.maximum(-sb_min[degenerate], 0.0).astype(np.float32)
# ββ Phase 2: quantize scales/mins to 4-bit ββ
max_scale = sb_scale.max(axis=1) # [n_blocks]
max_min = sb_the_min.max(axis=1) # [n_blocks]
# Quantize sub-block scales to 4-bit
has_scale = max_scale > 0
iscale_s = np.where(has_scale, q4scale / np.maximum(max_scale, 1e-30), 0.0)
scales_q = np.where(has_scale[:, None],
np.clip(np.round(iscale_s[:, None] * sb_scale), 0, 15),
0.0).astype(np.uint8)
# Quantize sub-block mins to 4-bit
has_min = max_min > 0
iscale_m = np.where(has_min, q4scale / np.maximum(max_min, 1e-30), 0.0)
mins_q = np.where(has_min[:, None],
np.clip(np.round(iscale_m[:, None] * sb_the_min), 0, 15),
0.0).astype(np.uint8)
d_fp16 = np.where(has_scale, max_scale / q4scale, 0.0).astype(np.float16)
dmin_fp16 = np.where(has_min, max_min / q4scale, 0.0).astype(np.float16)
# ββ Phase 3: requantize using fp16-truncated d/dmin ββ
scales_packed = scales_q | (mins_q << 4) # [n_blocks, 16]
d_f32 = d_fp16.astype(np.float32)
dmin_f32 = dmin_fp16.astype(np.float32)
d_sub = d_f32[:, None] * (scales_packed & 0xF).astype(np.float32)
dm_sub = dmin_f32[:, None] * (scales_packed >> 4).astype(np.float32)
# l = nearest_int((x + dm) / d), clamp [0,3]
valid_d = d_sub > 0
inv_d = np.where(valid_d, 1.0 / np.maximum(d_sub, 1e-30), 0.0)
q_vals = np.where(valid_d[:, :, None],
np.clip(np.round(
(f32_data.reshape(n_blocks, 16, 16) + dm_sub[:, :, None]) * inv_d[:, :, None]
), 0, 3),
0).astype(np.uint8)
# ββ Phase 4: pack ββ
q_flat = q_vals.reshape(n_blocks, QK_K)
q_groups = q_flat.reshape(n_blocks, 2, 4, 32)
qs_packed = (q_groups[:, :, 0, :] |
(q_groups[:, :, 1, :] << 2) |
(q_groups[:, :, 2, :] << 4) |
(q_groups[:, :, 3, :] << 6)).astype(np.uint8)
qs_packed = qs_packed.reshape(n_blocks, 64)
# Build output: [n_blocks, 84] bytes
# Layout matches ggml block_q2_K: scales[16] | qs[64] | d(fp16) | dmin(fp16)
result = np.zeros((n_blocks, 84), dtype=np.uint8)
result[:, 0:16] = scales_packed
result[:, 16:80] = qs_packed
result[:, 80:82] = d_fp16.view(np.uint8).reshape(n_blocks, 2)
result[:, 82:84] = dmin_fp16.view(np.uint8).reshape(n_blocks, 2)
return result.tobytes(), n_blocks
def dequant_q2k_fast(q2k_bytes, n_blocks):
"""Vectorized Q2_K dequantization for RMSE computation.
Block layout (84 bytes) β same for both C struct and Python writer:
scales[16] (bytes 0-15) | qs[64] (bytes 16-79) | d(fp16, bytes 80-81) | dmin(fp16, bytes 82-83)
The C struct BlockQ2K in gguf_format.h is:
{ uint8_t scales[16]; uint8_t qs[64]; uint16_t d; uint16_t dmin; }
Dequantization follows gguf_dequantize_q2_k_block() exactly:
For each half (0..1), qs_half = qs[half*32 : half*32+32]
For each shift j (0..3):
scale_idx = half*8 + j*2
elements [0..15] use scales[scale_idx], from qs_half[0..15] >> (j*2)
elements [16..31] use scales[scale_idx+1], from qs_half[16..31] >> (j*2)
"""
data = np.frombuffer(q2k_bytes, dtype=np.uint8).reshape(n_blocks, 84)
# Extract fields
scales_packed = data[:, 0:16] # [n_blocks, 16]
qs = data[:, 16:80] # [n_blocks, 64]
d_fp16 = data[:, 80:82].copy().view(np.float16).astype(np.float32).reshape(n_blocks)
dmin_fp16 = data[:, 82:84].copy().view(np.float16).astype(np.float32).reshape(n_blocks)
# Extract scale (low 4 bits) and min (high 4 bits) per sub-block
sc = (scales_packed & 0xF).astype(np.float32) # [n_blocks, 16]
mn = (scales_packed >> 4).astype(np.float32) # [n_blocks, 16]
# Compute per-sub-block d_sub and m_sub
d_sub = d_fp16[:, np.newaxis] * sc # [n_blocks, 16]
m_sub = dmin_fp16[:, np.newaxis] * mn # [n_blocks, 16]
# Unpack 2-bit quants from qs[64] into 256 values per block.
# Matches C reference: two scales per 32-byte extraction (16 elements each).
# half=0: qs[0..31], half=1: qs[32..63]
# shift j=0..3: scale_idx = half*8 + j*2 (first 16), +1 (second 16)
result = np.zeros((n_blocks, QK_K), dtype=np.float32)
for half in range(2):
qs_half = qs[:, half * 32:(half + 1) * 32] # [n_blocks, 32]
for sub in range(4):
# Extract 2-bit quants at this shift position
q_vals = ((qs_half >> (sub * 2)) & 3).astype(np.float32) # [n_blocks, 32]
base_idx = half * 128 + sub * 32
# First 16 elements: qs_half[0..15], scale index = half*8 + sub*2
si_0 = half * 8 + sub * 2
result[:, base_idx:base_idx + 16] = (
d_sub[:, si_0:si_0+1] * q_vals[:, :16] - m_sub[:, si_0:si_0+1]
)
# Second 16 elements: qs_half[16..31], scale index = si_0 + 1
si_1 = si_0 + 1
result[:, base_idx + 16:base_idx + 32] = (
d_sub[:, si_1:si_1+1] * q_vals[:, 16:] - m_sub[:, si_1:si_1+1]
)
return result.reshape(-1)
def is_attention_tensor(name):
"""Detect attention Q/K/V/O projection tensors.
These are the most sensitive to quantization and get promoted to Q4_0."""
attn_patterns = [
'attn_q.weight', 'attn_k.weight', 'attn_v.weight', 'attn_output.weight',
'attn_qkv.weight', 'attn_gate.weight',
'self_attn.q_proj.weight', 'self_attn.k_proj.weight',
'self_attn.v_proj.weight', 'self_attn.o_proj.weight',
# Qwen 3.6 DeltaNet SSM projections β treat as attention-class
'ssm_in_qkv.weight', 'ssm_in_z.weight', 'ssm_out.weight',
'linear_attn.in_proj_qkv.weight', 'linear_attn.in_proj_z.weight',
'linear_attn.out_proj.weight',
]
for pat in attn_patterns:
if pat in name:
return True
return False
def should_quantize(name, n_dims, dims, tied_embeddings=False):
"""Should this tensor be quantized to Q2_K?
With iMatrix importance weighting, Q2_K is applied to ALL eligible
tensors including embeddings for maximum compression.
Tensors kept as-is:
- 1D tensors (norms, biases) β always kept
- _norm, .bias β normalization layers
- ffn_gate_inp β MoE routing gate
- layer_output_scale β per-layer scaling factor (scalar)
- altup, laurel β small Gemma-specific tensors
- token_embd.weight / output.weight β always excluded here.
When embeddings are TIED, main() routes token_embd.weight to
Q8_0 (HPC Shor pipeline) instead: the same tensor serves as both
embedding lookup AND LM head, and Q2_K/Q4_0 there destroys logit
precision β looping / repetitive generation. --keep-embd keeps
it at source precision instead.
"""
n_elements = 1
for d in dims:
n_elements *= d
if n_dims < 2:
return False
if 'norm' in name:
return False
if '.bias' in name:
return False
if 'ffn_gate_inp' in name:
return False
if 'altup' in name or 'laurel' in name:
return False
if 'layer_output_scale' in name:
return False
# Embedding table β this is a lookup, not a matmul; Q2_K destroys
# token distinctions. Keep at source precision (F16/BF16).
if 'token_embd' in name:
return False
# LM head output projection β logit precision is critical for generation.
# (When tied with embeddings, this is the same tensor and also skipped above.)
if name == 'output.weight':
return False
# DeltaNet state-space parameters β keep at full precision
if 'ssm_a' in name or 'A_log' in name:
return False
if 'ssm_dt' in name or 'dt_bias' in name:
return False
if 'ssm_conv1d' in name or 'conv1d.weight' in name:
return False
# When embeddings are tied, token_embd.weight doubles as the output
# projection (LM head). It gets routed to Q4_0 in the quant plan
# instead of Q2_K β handled in main(), not here.
# Skip vision/audio encoder tensors
if 'v.' in name and name.startswith('v.'):
return False
if name.startswith('mm.') or name.startswith('a.'):
return False
# Small tensors are not worth quantizing
if n_elements < QK_K:
return False
# Must be divisible by QK_K
if n_elements % QK_K != 0:
return False
return True
def main():
if len(sys.argv) < 3:
print("Usage: python3 hexstate_requantize.py <input.gguf> <output.gguf>"
" [--keep-metadata] [--imatrix FILE] [--keep-embd] [--q2all]")
sys.exit(1)
input_path = sys.argv[1]
output_path = sys.argv[2]
keep_metadata = '--keep-metadata' in sys.argv
quantize_none = '--quantize-none' in sys.argv
q2all = '--q2all' in sys.argv
keep_embd = '--keep-embd' in sys.argv # keep tied embedding at source precision instead of Q8_0
# Check for imatrix
imatrix_data = None
for i, arg in enumerate(sys.argv):
if arg == '--imatrix' and i + 1 < len(sys.argv):
imat_path = sys.argv[i + 1]
if os.path.exists(imat_path):
imatrix_data = read_imatrix(imat_path)
print(f" Loaded imatrix: {len(imatrix_data)} tensors from {imat_path}")
else:
print(f" WARNING: imatrix file not found: {imat_path}")
break
# Check for HPC C library
use_hpc = _load_hexstate_lib() is not None
print()
print(" ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ")
print(" β HExState GGUF Re-Quantizer β")
print(" β GGUF β Q2_K GGUF with metadata passthrough β")
if q2all:
print(" β Mode: --q2all ALL eligible tensors β Q2_K (test mode) β")
if use_hpc and imatrix_data:
print(" β Engine: HPC + iMatrix (calibrated sensitivity propagation) β")
elif use_hpc:
print(" β Engine: HPC (BP + MSE Grid + Sensitivity Propagation) β")
else:
print(" β Engine: Python (numpy vectorized) β")
print(" ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ")
print()
start_time = time.time()
file_size = os.path.getsize(input_path)
print(f" Input: {input_path}")
print(f" Size: {file_size / 1024**3:.2f} GB")
print(f" Output: {output_path}")
print()
with open(input_path, 'rb') as fin:
# ββ Read Header ββ
magic = struct.unpack('<I', fin.read(4))[0]
assert magic == GGUF_MAGIC, f"Bad GGUF magic: 0x{magic:08X}"
version = struct.unpack('<I', fin.read(4))[0]
n_tensors = struct.unpack('<Q', fin.read(8))[0]
n_kv = struct.unpack('<Q', fin.read(8))[0]
print(f" GGUF v{version}: {n_tensors} tensors, {n_kv} KV pairs")
print()
# ββ Read KV pairs (store as raw bytes for passthrough) ββ
kv_pairs = []
for i in range(n_kv):
key = read_string(fin)
vtype = struct.unpack('<I', fin.read(4))[0]
raw_value = read_kv_value(fin, vtype)
kv_pairs.append((key, vtype, raw_value))
# ββ Read Tensor Info ββ
tensor_infos = []
for i in range(n_tensors):
name = read_string(fin)
n_dims = struct.unpack('<I', fin.read(4))[0]
dims = [struct.unpack('<Q', fin.read(8))[0] for _ in range(n_dims)]
ttype = struct.unpack('<I', fin.read(4))[0]
offset = struct.unpack('<Q', fin.read(8))[0]
n_elements = 1
for d in dims:
n_elements *= d
blk_sz = TYPE_BLOCK_SIZE.get(ttype, 1)
blk_bytes = TYPE_BLOCK_BYTES.get(ttype, 4)
n_blocks = (n_elements + blk_sz - 1) // blk_sz
data_size = n_blocks * blk_bytes
tensor_infos.append({
'name': name, 'n_dims': n_dims, 'dims': dims,
'type': ttype, 'offset': offset,
'n_elements': n_elements, 'data_size': data_size,
})
# Calculate data section start
pos_after_info = fin.tell()
data_section_start = align_offset(pos_after_info)
print(f" Data section starts at: {data_section_start:,}")
print()
# ββ Detect tied embeddings ββ
# If no separate output.weight tensor exists, token_embd.weight
# doubles as the LM head. Must preserve it at full precision.
tensor_names = {ti['name'] for ti in tensor_infos}
has_output_weight = 'output.weight' in tensor_names
tied_embeddings = not has_output_weight and 'token_embd.weight' in tensor_names
if tied_embeddings:
if keep_embd:
print(" β Tied embeddings detected β token_embd.weight kept at source precision (--keep-embd)")
else:
print(" β Tied embeddings detected β token_embd.weight β Q8_0 via Shor pipeline (serves as LM head;")
print(" Q2_K/Q4_0 here destroys logit precision β classic looping-output symptom)")
print()
# ββ Determine output types ββ
quant_plan = []
total_quant = 0
total_attn = 0
total_keep = 0
total_embd = 0
for ti in tensor_infos:
if quantize_none:
will_quant = False
elif (tied_embeddings and ti['name'] == 'token_embd.weight'
and not keep_embd and ti['n_elements'] % 32 == 0):
# Tied embedding doubles as the LM head. NOTE: the old
# 'promote to Q4_0' branch below should_quantize() was dead
# code (should_quantize always returned False for
# token_embd), so the tensor was silently kept at F16/BF16.
# Now: Q8_0 (8.5 bpw, ~2x smaller than F16) via the HPC
# Shor pipeline β transparent for both embedding lookup
# and logit projection.
will_quant = 'EMBD_Q8'
total_embd += 1
elif should_quantize(ti['name'], ti['n_dims'], ti['dims'], tied_embeddings):
if q2all:
# --q2all: ALL eligible tensors β Q2_K, no exceptions
# (tied embedding stays on the Q8_0 route above).
will_quant = True
total_quant += 1
elif is_attention_tensor(ti['name']):
will_quant = 'ATTN_Q4' # Promote attention to Q4_0 HPC
total_attn += 1
else:
will_quant = True
total_quant += 1
else:
will_quant = False
total_keep += 1
quant_plan.append(will_quant)
if q2all:
print(f" Mode: --q2all β all eligible tensors forced to Q2_K")
print(f" Tensors to quantize (Q2_K): {total_quant}")
print(f" Tensors to keep as-is: {total_keep}")
else:
print(f" Tensors to quantize (Q2_K): {total_quant}")
print(f" Tensors to promote (Q4_0Β·HPC): {total_attn}")
print(f" Tied embd β Q8_0 (ShorΒ·HPC): {total_embd}")
print(f" Tensors to keep as-is: {total_keep}")
print()
# ββ Compute output tensor sizes and offsets ββ
out_tensor_infos = []
out_data_offset = 0
for i, ti in enumerate(tensor_infos):
if quant_plan[i]:
out_dims = list(ti['dims'])
dim0 = out_dims[0] if ti['n_dims'] >= 2 else ti['n_elements']
if quant_plan[i] == 'EMBD_Q8':
# Tied embedding / LM head β Q8_0 (8.5 bpw, 34 B / 32 w)
out_type = GGML_TYPE_Q8_0
n_blocks = ti['n_elements'] // 32
out_size = n_blocks * 34
print(f" [EMBDβQ8_0Β·Shor] {ti['name']} ({ti['n_elements']:,} elements)")
elif quant_plan[i] == 'ATTN_Q4':
# Attention tensor β Q4_0 HPC (4.5 bpw)
out_type = GGML_TYPE_Q4_0
n_blocks = (ti['n_elements'] + 31) // 32
out_size = n_blocks * 18
print(f" [ATTNβQ4_0Β·HPC] {ti['name']} ({ti['n_elements']} elements)")
elif dim0 % QK_K == 0 or q2all:
# Q2_K (2.6 bpw, block_size=256)
# --q2all forces Q2_K even when dim0 isn't a clean multiple;
# the quantizer pads internally to the next QK_K boundary.
out_type = GGML_TYPE_Q2_K
n_blocks = (ti['n_elements'] + QK_K - 1) // QK_K
out_size = n_blocks * 84
if q2all and dim0 % QK_K != 0:
print(f" [Q2_KΒ·PADDED] {ti['name']} (dim0={dim0}, padded to QK_K boundary)")
elif dim0 % 32 == 0:
# Q4_0 fallback (4.5 bpw, block_size=32)
out_type = GGML_TYPE_Q4_0
n_blocks = ti['n_elements'] // 32
out_size = n_blocks * 18
quant_plan[i] = 'Q4_0'
print(f" Q4_0: {ti['name']} (dims[0]={dim0})")
else:
out_type = ti['type']
out_size = ti['data_size']
quant_plan[i] = False
print(f" Keep: {ti['name']} (dims[0]={dim0})")
else:
out_type = ti['type']
out_size = ti['data_size']
out_dims = list(ti['dims'])
out_tensor_infos.append({
'name': ti['name'],
'n_dims': ti['n_dims'],
'dims': out_dims,
'type': out_type,
'offset': out_data_offset,
'data_size': out_size,
})
out_data_offset += out_size
out_data_offset = align_offset(out_data_offset)
# ββ Update KV pairs ββ
updated_kv = []
if keep_metadata:
print(" --keep-metadata: passing through ALL KV pairs unchanged")
updated_kv = list(kv_pairs)
else:
for key, vtype, raw_value in kv_pairs:
if key == 'general.file_type' and vtype == 4: # UINT32
# file_type=10 means Q2_K in llama.cpp
updated_kv.append((key, vtype, struct.pack('<I', 10)))
elif key == 'general.quantization_version' and vtype == 4:
updated_kv.append((key, vtype, struct.pack('<I', 2)))
elif key == 'tokenizer.ggml.token_type' and vtype == 9:
# ββ Fix Gemma 4 token types ββ
# convert_hf_to_gguf.py incorrectly marks control tokens as
# NORMAL (1), causing llama.cpp to sample them (e.g. <unused24>
# spam). Fix: read the tokens array to find control-looking
# tokens, then patch their types to CONTROL (3).
# See: https://github.com/ggml-org/llama.cpp/issues/21321
tokens_kv = next((v for k, vt, v in kv_pairs
if k == 'tokenizer.ggml.tokens' and vt == 9), None)
token_names = []
if tokens_kv:
bio = io.BytesIO(tokens_kv)
arr_type = struct.unpack('<I', bio.read(4))[0]
arr_len = struct.unpack('<Q', bio.read(8))[0]
for _ in range(arr_len):
slen = struct.unpack('<Q', bio.read(8))[0]
token_names.append(bio.read(slen).decode('utf-8', errors='replace'))
# Parse the token_type array
bio2 = io.BytesIO(raw_value)
arr_type2 = struct.unpack('<I', bio2.read(4))[0]
arr_len2 = struct.unpack('<Q', bio2.read(8))[0]
ttypes = list(struct.unpack(f'<{arr_len2}i', bio2.read(arr_len2 * 4)))
# Patch control-looking tokens
n_fixed = 0
CONTROL_TYPE = 3
import re
for i, tname in enumerate(token_names):
if ttypes[i] == CONTROL_TYPE:
continue # already correct
if ttypes[i] == 6:
continue # BYTE type β leave as-is
# Only fix tokens that are genuine control/special tokens:
# - <eos>, <bos>, <unk>, <mask>, </s> β sentence markers
# - <|turn>, <turn|>, <|tool_*|> etc β delimiters
# NOTE: do NOT mark <unused*> as CONTROL β Gemma 4 uses
# these tokens internally for thinking/channel markers
# (e.g. <unused24> = <|channel>). The llama.cpp parser
# handles them via the peg-gemma4 format instead.
is_control = False
if tname in ('<eos>', '<bos>', '<unk>', '<mask>', '</s>',
'<pad>', '<s>'):
is_control = True
elif re.match(r'^<\|.*\|?>$', tname) or re.match(r'^<.*\|>$', tname):
is_control = True
if is_control and ttypes[i] != CONTROL_TYPE:
ttypes[i] = CONTROL_TYPE
n_fixed += 1
print(f" Fixed {n_fixed} token types to CONTROL (Gemma 4 <unused> fix)")
# Rebuild the raw value
new_raw = struct.pack('<I', arr_type2)
new_raw += struct.pack('<Q', arr_len2)
new_raw += struct.pack(f'<{arr_len2}i', *ttypes)
updated_kv.append((key, vtype, new_raw))
elif key == 'tokenizer.chat_template' and vtype == 8:
# ββ Replace chat template with fixed Gemma 4 template ββ
# The HF-exported template doesn't handle thinking mode, causing
# the model to emit <unused24> tokens. The fixed template from
# llama.cpp PR #21418 pre-fills an empty thought block when
# thinking is disabled: <|channel>thought\n<channel|>
# See: https://github.com/ggml-org/llama.cpp/pull/21418
script_dir = os.path.dirname(os.path.abspath(__file__))
workspace_dir = os.path.dirname(script_dir)
template_path = os.path.join(workspace_dir, 'llama-cpp-latest',
'models', 'templates', 'google-gemma-4-31B-it.jinja')
if os.path.exists(template_path):
with open(template_path, 'r') as tf:
new_template = tf.read()
new_raw = struct.pack('<Q', len(new_template.encode('utf-8')))
new_raw += new_template.encode('utf-8')
updated_kv.append((key, vtype, new_raw))
print(f" Replaced chat template with fixed Gemma 4 template ({len(new_template)} chars)")
else:
print(f" WARNING: Fixed template not found at {template_path}, keeping original")
updated_kv.append((key, vtype, raw_value))
else:
updated_kv.append((key, vtype, raw_value))
# ββ Write output GGUF ββ
print(" Writing output GGUF...")
with open(output_path, 'wb') as fout:
# Header
fout.write(struct.pack('<I', GGUF_MAGIC))
fout.write(struct.pack('<I', GGUF_VERSION))
fout.write(struct.pack('<Q', n_tensors))
fout.write(struct.pack('<Q', n_kv))
# KV pairs (passthrough)
for key, vtype, raw_value in updated_kv:
write_string(fout, key)
fout.write(struct.pack('<I', vtype))
fout.write(raw_value)
# Tensor info
for oti in out_tensor_infos:
write_string(fout, oti['name'])
fout.write(struct.pack('<I', oti['n_dims']))
for d in oti['dims']:
fout.write(struct.pack('<Q', d))
fout.write(struct.pack('<I', oti['type']))
fout.write(struct.pack('<Q', oti['offset']))
# Alignment padding before data
pos = fout.tell()
aligned = align_offset(pos)
if aligned > pos:
fout.write(b'\x00' * (aligned - pos))
# ββ Write tensor data ββ
quant_count = 0
total_quant_bytes = 0
total_keep_bytes = 0
total_rmse = 0.0
q2k_rmse_sum = 0.0
q2k_tensor_count = 0
for i, ti in enumerate(tensor_infos):
# Progress bar
pct = (i + 1) / n_tensors * 100
bar_width = 40
filled = int(bar_width * (i + 1) / n_tensors)
bar = 'β' * filled + 'β' * (bar_width - filled)
elapsed = time.time() - start_time
eta = elapsed / max(i + 1, 1) * (n_tensors - i - 1)
sys.stdout.write(f"\r [{bar}] {pct:5.1f}% ({i+1}/{n_tensors}) {elapsed:.0f}s ETA:{eta:.0f}s {ti['name'][:50]}")
sys.stdout.flush()
# Read source tensor data
abs_offset = data_section_start + ti['offset']
fin.seek(abs_offset)
raw_data = fin.read(ti['data_size'])
if quant_plan[i] == 'EMBD_Q8':
# ββ Tied embedding β Q8_0 via the HPC Shor pipeline ββ
if ti['type'] == GGML_TYPE_BF16:
f32 = bf16_to_f32(raw_data, ti['n_elements'])
elif ti['type'] == GGML_TYPE_F16:
f32 = f16_to_f32(raw_data, ti['n_elements'])
elif ti['type'] == GGML_TYPE_F32:
f32 = np.frombuffer(raw_data, dtype=np.float32).copy()
else:
# Can't re-quantize from quantized source β keep
fout.write(raw_data)
pad = align_offset(fout.tell()) - fout.tell()
if pad > 0: fout.write(b'\x00' * pad)
continue
n_el = ti['n_elements']
n_blocks_q8 = n_el // 32
if use_hpc and hasattr(_HEXSTATE_LIB, 'hexstate_quantize_tensor_q8_0_hpc'):
output_buf = np.zeros(n_blocks_q8 * 34, dtype=np.uint8)
error = ctypes.c_float(0.0)
f32_c = np.ascontiguousarray(f32, dtype=np.float32)
imat_ptr = None
if imatrix_data and ti['name'] in imatrix_data:
iw = imatrix_data[ti['name']]
n_cols = iw.shape[0]
n_rows = n_el // n_cols if n_cols > 0 else 1
imat_full = np.tile(iw, n_rows)[:n_el].astype(np.float32)
imat_c = np.ascontiguousarray(imat_full)
imat_ptr = imat_c.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
_HEXSTATE_LIB.hexstate_quantize_tensor_q8_0_hpc(
f32_c.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
ctypes.c_int64(n_el),
output_buf.ctypes.data_as(ctypes.c_void_p),
ctypes.byref(error),
imat_ptr,
ctypes.c_int(0),
)
fout.write(output_buf.tobytes())
rmse8 = float(np.sqrt(error.value / max(n_el, 1)))
print(f"\n [Q8_0Β·Shor] {ti['name']} RMSE={rmse8:.6e}")
else:
q8_bytes, n_blocks_q8, sse8 = quantize_tensor_q8_0(f32)
fout.write(q8_bytes)
rmse8 = float(np.sqrt(sse8 / max(n_el, 1)))
print(f"\n [Q8_0] {ti['name']} RMSE={rmse8:.6e} (numpy fallback)")
quant_count += 1
total_quant_bytes += n_blocks_q8 * 34
elif quant_plan[i] in ('Q4_0', 'ATTN_Q4'):
# ββ Q4_0 quantization (fallback or attention HPC) ββ
if ti['type'] == GGML_TYPE_BF16:
f32 = bf16_to_f32(raw_data, ti['n_elements'])
elif ti['type'] == GGML_TYPE_F16:
f32 = f16_to_f32(raw_data, ti['n_elements'])
elif ti['type'] == GGML_TYPE_F32:
f32 = np.frombuffer(raw_data, dtype=np.float32).copy()
else:
fout.write(raw_data)
pad = align_offset(fout.tell()) - fout.tell()
if pad > 0: fout.write(b'\x00' * pad)
continue
# Pad to 32-element boundary
n_el = len(f32)
pad_to = ((n_el + 31) // 32) * 32
if pad_to > n_el:
f32 = np.concatenate([f32, np.zeros(pad_to - n_el, dtype=np.float32)])
n_el = pad_to
n_blocks_q4 = n_el // 32
# Use HPC for attention tensors if available
if quant_plan[i] == 'ATTN_Q4' and use_hpc and hasattr(_HEXSTATE_LIB, 'hexstate_quantize_tensor_q4_0_hpc'):
output_buf = np.zeros(n_blocks_q4 * 18, dtype=np.uint8)
error = ctypes.c_float(0.0)
f32_c = np.ascontiguousarray(f32, dtype=np.float32)
# Look up imatrix importance
imat_ptr = None
if imatrix_data and ti['name'] in imatrix_data:
iw = imatrix_data[ti['name']]
n_cols = iw.shape[0]
n_rows = n_el // n_cols if n_cols > 0 else 1
imat_full = np.tile(iw, n_rows)[:n_el].astype(np.float32)
imat_c = np.ascontiguousarray(imat_full)
imat_ptr = imat_c.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
_HEXSTATE_LIB.hexstate_quantize_tensor_q4_0_hpc(
f32_c.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
ctypes.c_int64(n_el),
output_buf.ctypes.data_as(ctypes.c_void_p),
ctypes.byref(error),
imat_ptr,
ctypes.c_int(1), # verbose
)
fout.write(output_buf.tobytes())
print(f"\n [Q4_0Β·HPC] {ti['name']} RMSE={np.sqrt(error.value / ti['n_elements']):.6e}")
else:
# Vectorized Q4_0: process all blocks at once
blocks = f32.reshape(-1, 32)
amax = np.max(np.abs(blocks), axis=1)
d = amax / 7.0
d[d == 0] = 1.0 # avoid div by zero
qs = np.clip(np.round(blocks / d[:, None]) + 8, 0, 15).astype(np.uint8)
d_orig = amax / 7.0 # restore zeros
d_fp16 = d_orig.astype(np.float16)
out_buf = bytearray(n_blocks_q4 * 18)
for b in range(n_blocks_q4):
off = b * 18
struct.pack_into('<e', out_buf, off, float(d_fp16[b]))
for j in range(16):
out_buf[off + 2 + j] = int(qs[b, j]) | (int(qs[b, j + 16]) << 4)
fout.write(bytes(out_buf))
quant_count += 1
total_quant_bytes += n_blocks_q4 * 18
elif quant_plan[i]:
# Convert to F32 for quantization
if ti['type'] == GGML_TYPE_BF16:
f32 = bf16_to_f32(raw_data, ti['n_elements'])
elif ti['type'] == GGML_TYPE_F16:
f32 = f16_to_f32(raw_data, ti['n_elements'])
elif ti['type'] == GGML_TYPE_F32:
f32 = np.frombuffer(raw_data, dtype=np.float32).copy()
else:
# Can't re-quantize from quantized format β keep as-is
fout.write(raw_data)
pad = align_offset(fout.tell()) - fout.tell()
if pad > 0:
fout.write(b'\x00' * pad)
continue
# Quantize to Q2_K β always use HPC with chunked processing
# Each chunk gets full HPC treatment (no size threshold)
HPC_CHUNK = 50_000_000 # 50M elements per HPC chunk
HPC_CHUNK = (HPC_CHUNK // QK_K) * QK_K # align to QK_K
# Look up imatrix importance for this tensor
imat_full = None
if imatrix_data and ti['name'] in imatrix_data:
iw = imatrix_data[ti['name']]
n_cols = iw.shape[0]
n_rows = ti['n_elements'] // n_cols if n_cols > 0 else 1
imat_full = np.tile(iw, n_rows)[:ti['n_elements']]
n_el = ti['n_elements']
if use_hpc and n_el <= HPC_CHUNK:
# Small tensor β single HPC pass
q2k_data, n_blocks = quantize_tensor_q2k_hpc(f32, opt_mode=2, importance=imat_full)
elif use_hpc:
# Large tensor β chunked HPC (each chunk gets BP)
chunks = []
processed = 0
while processed < n_el:
end = min(processed + HPC_CHUNK, n_el)
chunk_f32 = f32[processed:end]
if len(chunk_f32) % QK_K != 0:
pad_len = QK_K - (len(chunk_f32) % QK_K)
chunk_f32 = np.concatenate([chunk_f32, np.zeros(pad_len, dtype=np.float32)])
chunk_imp = imat_full[processed:end] if imat_full is not None else None
if chunk_imp is not None and len(chunk_imp) < len(chunk_f32):
chunk_imp = np.concatenate([chunk_imp, np.ones(len(chunk_f32) - len(chunk_imp), dtype=np.float32)])
chunk_data, _ = quantize_tensor_q2k_hpc(chunk_f32, opt_mode=2, importance=chunk_imp)
actual_blocks = (end - processed + QK_K - 1) // QK_K
chunks.append(chunk_data[:actual_blocks * 84])
processed = end
pct = 100.0 * processed / n_el
print(f"\r β {processed/1e6:.0f}M/{n_el/1e6:.0f}M ({pct:.0f}%)", end='', flush=True)
print()
q2k_data = b''.join(chunks)
n_blocks = n_el // QK_K
else:
# No HPC available β python fallback
CHUNK_SIZE = 10_000_000
CHUNK_SIZE = (CHUNK_SIZE // QK_K) * QK_K
chunks = []
processed = 0
while processed < n_el:
end = min(processed + CHUNK_SIZE, n_el)
chunk_data, _ = quantize_tensor_q2k(f32[processed:end])
chunks.append(chunk_data)
processed = end
pct = 100.0 * processed / n_el
print(f"\r β {processed/1e6:.0f}M/{n_el/1e6:.0f}M ({pct:.0f}%)", end='', flush=True)
print()
q2k_data = b''.join(chunks)
n_blocks = n_el // QK_K
fout.write(q2k_data)
# ββ Compute and report exact per-tensor RMSE ββ
try:
CHUNK_BLK = 100_000 # blocks per chunk to bound memory
total_se = 0.0
total_n = 0
for ci in range(0, n_blocks, CHUNK_BLK):
ce = min(ci + CHUNK_BLK, n_blocks)
chunk_q = q2k_data[ci*84:ce*84]
deq_chunk = dequant_q2k_fast(chunk_q, ce - ci)
orig_chunk = f32[ci*QK_K:ce*QK_K]
n_valid = min(len(orig_chunk), len(deq_chunk))
diff = orig_chunk[:n_valid] - deq_chunk[:n_valid]
total_se += np.sum(diff ** 2)
total_n += n_valid
tensor_rmse = np.sqrt(total_se / max(total_n, 1))
q2k_rmse_sum += tensor_rmse
q2k_tensor_count += 1
print(f"\n [Q2_K] {ti['name'][:55]} RMSE={tensor_rmse:.6e}")
except Exception as e:
print(f"\n [Q2_K] {ti['name'][:55]} RMSE=err({e})")
quant_count += 1
total_quant_bytes += len(q2k_data)
else:
# Keep as-is (passthrough)
fout.write(raw_data)
total_keep_bytes += len(raw_data)
# Alignment padding
pad = align_offset(fout.tell()) - fout.tell()
if pad > 0:
fout.write(b'\x00' * pad)
final_size = fout.tell()
elapsed = time.time() - start_time
print(f"\r {'β' * 40} 100.0% ({n_tensors}/{n_tensors}) {elapsed:.0f}s" + " " * 60)
print()
# ββ Summary ββ
original_bytes = sum(ti['data_size'] for ti in tensor_infos)
compression = original_bytes / max(final_size, 1)
print(" ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ")
print(" β RE-QUANTIZATION SUMMARY β")
print(" β βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ£")
print(f" β Tensors quantized (Q2_K): {quant_count:<33d} β")
print(f" β Tensors kept as-is: {total_keep:<33d} β")
print(f" β Q2_K data: {total_quant_bytes:>12,} bytes ({total_quant_bytes/1024**2:>7.1f} MB) β")
print(f" β Kept data: {total_keep_bytes:>12,} bytes ({total_keep_bytes/1024**2:>7.1f} MB) β")
print(f" β Original size: {file_size:>12,} bytes ({file_size/1024**3:>7.2f} GB) β")
print(f" β Output size: {final_size:>12,} bytes ({final_size/1024**3:>7.2f} GB) β")
print(f" β Compression: {compression:>42.1f}x β")
if q2k_tensor_count > 0:
mean_rmse = q2k_rmse_sum / q2k_tensor_count
print(f" β Mean Q2_K RMSE: {mean_rmse:>12.6e} β")
print(f" β Total time: {elapsed:>39.1f} sec β")
print(" ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ")
print()
print(f" Output: {output_path}")
print()
if __name__ == '__main__':
main() |