Hugging Face's logo

software-mansion
/
gemma-4-E2B-it-ExecuTorch

Audio-Text-to-Text
ExecuTorch
google
deepmind
gemma4
edge
Model card Files
xet
 Community
Bartosz Hanc commited on
Commit
25617f4
·
1 Parent(s): 2049254

Update Gemma 4 E2B MLX multimodal model (v0.11.0)

Browse files
Files changed (2) hide show
  1. e2b/mlx/config.json +170 -13
  2. e2b/mlx/gemma4_e2b_mlx_int4.pte +2 -2
e2b/mlx/config.json CHANGED
@@ -13,6 +13,8 @@
13
  {
14
  "file": "gemma4_e2b_mlx_int4.pte",
15
  "precision": "int4",
 
 
16
  "methods": {
17
  "enable_dynamic_shape": true,
18
  "get_bos_id": 2,
@@ -24,24 +26,179 @@
24
  "get_max_context_len": 2048,
25
  "get_max_seq_len": 511,
26
  "get_vocab_size": 262144,
27
- "has_ple": true,
28
- "image_placeholder_id": 262145,
29
  "use_kv_cache": true,
30
- "audio_encoder": {
31
- "inputs": [],
32
- "outputs": []
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  },
34
  "text_decoder": {
35
- "inputs": [],
36
- "outputs": []
37
- },
38
- "token_embedding": {
39
- "inputs": [],
40
- "outputs": []
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41
  },
42
  "vision_encoder": {
43
- "inputs": [],
44
- "outputs": []
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
45
  }
46
  }
47
  }
 
13
  {
14
  "file": "gemma4_e2b_mlx_int4.pte",
15
  "precision": "int4",
16
+ "quantized": true,
17
+ "default": true,
18
  "methods": {
19
  "enable_dynamic_shape": true,
20
  "get_bos_id": 2,
 
26
  "get_max_context_len": 2048,
27
  "get_max_seq_len": 511,
28
  "get_vocab_size": 262144,
 
 
29
  "use_kv_cache": true,
30
+ "has_ple": true,
31
+ "image_placeholder_id": 258880,
32
+ "audio_placeholder_id": 258881,
33
+ "audio_sample_rate": 16000,
34
+ "audio_block_samples": 7680,
35
+ "audio_block_k_min": 1,
36
+ "audio_block_k_max": 62,
37
+ "audio_soft_tokens_max": 744,
38
+ "audio_sample_count_min": 7680,
39
+ "audio_sample_count_max": 476160,
40
+ "token_embedding": {
41
+ "inputs": [
42
+ {
43
+ "name": "input_ids",
44
+ "shape": [
45
+ 1,
46
+ {
47
+ "min": 1,
48
+ "max": 511
49
+ }
50
+ ],
51
+ "dtype": "int64"
52
+ }
53
+ ],
54
+ "outputs": [
55
+ {
56
+ "name": "embeds",
57
+ "shape": [
58
+ 1,
59
+ {
60
+ "min": 1,
61
+ "max": 511
62
+ }
63
+ ],
64
+ "dtype": "float32"
65
+ },
66
+ {
67
+ "name": "ple_tokens",
68
+ "shape": [
69
+ 1,
70
+ {
71
+ "min": 1,
72
+ "max": 511
73
+ },
74
+ 35,
75
+ 256
76
+ ],
77
+ "dtype": "float32"
78
+ }
79
+ ]
80
  },
81
  "text_decoder": {
82
+ "inputs": [
83
+ {
84
+ "name": "embeds",
85
+ "shape": [
86
+ 1,
87
+ {
88
+ "min": 1,
89
+ "max": 511
90
+ },
91
+ 1536
92
+ ],
93
+ "dtype": "float32"
94
+ },
95
+ {
96
+ "name": "ple_tokens",
97
+ "shape": [
98
+ 1,
99
+ {
100
+ "min": 1,
101
+ "max": 511
102
+ },
103
+ 35,
104
+ 256
105
+ ],
106
+ "dtype": "float32"
107
+ },
108
+ {
109
+ "name": "input_pos",
110
+ "shape": [
111
+ {
112
+ "min": 1,
113
+ "max": 511
114
+ }
115
+ ],
116
+ "dtype": "int64"
117
+ }
118
+ ],
119
+ "outputs": [
120
+ {
121
+ "name": "logits",
122
+ "shape": [
123
+ 1,
124
+ 1,
125
+ 262144
126
+ ],
127
+ "dtype": "float32"
128
+ }
129
+ ]
130
  },
131
  "vision_encoder": {
132
+ "inputs": [
133
+ {
134
+ "name": "pixel_values",
135
+ "shape": [
136
+ 1,
137
+ 3,
138
+ 448,
139
+ 640
140
+ ],
141
+ "dtype": "float32"
142
+ }
143
+ ],
144
+ "outputs": [
145
+ {
146
+ "name": "embeds",
147
+ "shape": [
148
+ 1,
149
+ 280,
150
+ 1536
151
+ ],
152
+ "dtype": "float32"
153
+ }
154
+ ]
155
+ },
156
+ "audio_encoder": {
157
+ "inputs": [
158
+ {
159
+ "name": "waveform",
160
+ "shape": [
161
+ 1,
162
+ {
163
+ "min": 7680,
164
+ "max": 476160
165
+ }
166
+ ],
167
+ "dtype": "float32"
168
+ },
169
+ {
170
+ "name": "num_valid_samples",
171
+ "shape": [
172
+ 1
173
+ ],
174
+ "dtype": "int64"
175
+ }
176
+ ],
177
+ "outputs": [
178
+ {
179
+ "name": "audio_embeds",
180
+ "shape": [
181
+ 1,
182
+ {
183
+ "min": 12,
184
+ "max": 744
185
+ },
186
+ 1536
187
+ ],
188
+ "dtype": "float32"
189
+ },
190
+ {
191
+ "name": "audio_mask",
192
+ "shape": [
193
+ 1,
194
+ {
195
+ "min": 12,
196
+ "max": 744
197
+ }
198
+ ],
199
+ "dtype": "float32"
200
+ }
201
+ ]
202
  }
203
  }
204
  }
e2b/mlx/gemma4_e2b_mlx_int4.pte CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8914ade9e9e3e09a4939a1676fb46254d80c38c7768d3a4108bcd0820e1981f4
3
- size 3180710400
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c09b22d8601464c84610d8b60d774d97023d670aa2d4d9982776f560a94d04b
3
+ size 2728359680