lthn commited on
Commit
ac18253
·
verified ·
1 Parent(s): 882673e

Replace bf16 with Q4 MLX — consumer-ready quantisation

Browse files
config.json CHANGED
@@ -18,6 +18,16 @@
18
  "image_token_id": 258880,
19
  "initializer_range": 0.02,
20
  "model_type": "gemma4",
 
 
 
 
 
 
 
 
 
 
21
  "text_config": {
22
  "attention_bias": false,
23
  "attention_dropout": 0.0,
 
18
  "image_token_id": 258880,
19
  "initializer_range": 0.02,
20
  "model_type": "gemma4",
21
+ "quantization": {
22
+ "group_size": 64,
23
+ "bits": 4,
24
+ "mode": "affine"
25
+ },
26
+ "quantization_config": {
27
+ "group_size": 64,
28
+ "bits": 4,
29
+ "mode": "affine"
30
+ },
31
  "text_config": {
32
  "attention_bias": false,
33
  "attention_dropout": 0.0,
model-00001-of-00012.safetensors → model-00001-of-00004.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:462ed563d4cbf840aba9ff33fc90f264cd84802b086390316997629e6b8495ae
3
- size 5229866535
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:56d3c9369c48dc4f89afa4746fcc9c32f184d6a2471dd57d83fdf3211fd7889d
3
+ size 5366617512
model-00002-of-00012.safetensors → model-00002-of-00004.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a3a51f95fc25992817ce12c27a8514642c7546fd54473eb744819e9b9c4d4967
3
- size 5362167404
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ae88f3ac4ceb64dd129de643570875baf5ee8a6cc0ec21bf9fd7a060a7625b9
3
+ size 5361642573
model-00003-of-00012.safetensors → model-00003-of-00004.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4af3b05a75840c7ec9132961add6da894e785f2eb7637ae8be0dd4eb2e3c15af
3
- size 5163944494
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2d206c897deaa7585ac6393e4a27dd17024e064121d55deaa906c0c839a2162d
3
+ size 5367276094
model-00004-of-00012.safetensors → model-00004-of-00004.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:85ede963d97f0d2603f0330e9480403f210a9c756d04078e8f1e05e7109d7ce2
3
- size 5362123860
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9893d5a09967478543422e055a5cb2df652d14ebdf8ff24c6ae7e03f2738778b
3
+ size 1173848301
model-00005-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:b385ca3b7485c52aa8fe05e6120ce264ce2032ad1626d9dea2312916fda5f96d
3
- size 5307117538
 
 
 
 
model-00006-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:02ab7a3ba5d1050d32048764e35d1c81e82b8bd7d0487c9e1e1910bd11441477
3
- size 5218994488
 
 
 
 
model-00007-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:8a804a6e92a51b7c52b0a3d0df9599c493197d91c0f7e6828060c58d08cbad34
3
- size 5252066087
 
 
 
 
model-00008-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:a78c0fe9d06cbeaeb01f1394c7da905926a3dfa7644ee37a29a9c7e534fa82ad
3
- size 5163944506
 
 
 
 
model-00009-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:fbe5fa924b7675c7443fe62609da02a648dbad8468077f04d578cb592406a8a2
3
- size 5362123852
 
 
 
 
model-00010-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:44b39b35c0198a2cc0f54a62f1308b39920f43ca89f6bc6402f88b0962a62857
3
- size 5307117526
 
 
 
 
model-00011-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:bcdfb5e62f1f8400d68c0412de162fe5acc854e8956d26f5dc1a9bb090f8c344
3
- size 5218994502
 
 
 
 
model-00012-of-00012.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:5ee556e4915e48186c43eb815eeffc54fb9f1f5a9517eb990a199c6ab4c412ef
3
- size 3446338270
 
 
 
 
model.safetensors.index.json CHANGED
The diff for this file is too large to render. See raw diff