nodes/city96-ComfyUI-GGUF
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Use int64 for index in dequant

Browse Source 
This is the fix provided by @LSXAxeller in #385
This commit is contained in:
City
2025-12-18 09:11:56 +01:00
parent 8493db6481
commit 795e45156e
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
dequant.py
+2 -2
View File
@@ -247,7 +247,7 @@ def dequantize_blocks_IQ4_NL(blocks, block_size, type_size, dtype=None):
d = d.view(torch.float16).to(dtype)
qs = qs.reshape((n_blocks, -1, 1, block_size//2)) >> torch.tensor([0, 4], device=d.device, dtype=torch.uint8).reshape((1, 1, 2, 1))
qs = (qs & 0x0F).reshape((n_blocks, -1, 1)).to(torch.int32)
qs = (qs & 0x0F).reshape((n_blocks, -1, 1)).to(torch.int64)
kvalues = KVALUES.to(qs.device).expand(*qs.shape[:-1], 16)
qs = torch.gather(kvalues, dim=-1, index=qs).reshape((n_blocks, -1))
@@ -277,7 +277,7 @@ def dequantize_blocks_IQ4_XS(blocks, block_size, type_size, dtype=None):
qs = qs.reshape((n_blocks, -1, 32, 1)) & 0x0F
kvalues = KVALUES.to(qs.device).expand(*qs.shape[:-1], 16)
qs = torch.gather(kvalues, dim=-1, index=qs.to(torch.int32)).reshape((n_blocks, -1, 32))
qs = torch.gather(kvalues, dim=-1, index=qs.to(torch.int64)).reshape((n_blocks, -1, 32))
del kvalues # see IQ4_NL
del shift_a
del shift_b