hqfang commited on
Commit
953e671
·
1 Parent(s): 5ff40da

update weights from bf16 to fp32

Browse files
config.json CHANGED
@@ -183,7 +183,7 @@
183
  }
184
  },
185
  "tie_word_embeddings": false,
186
- "torch_dtype": "bfloat16",
187
  "transformers_version": "4.52.3",
188
  "use_cache": true,
189
  "vit_config": {
 
183
  }
184
  },
185
  "tie_word_embeddings": false,
186
+ "torch_dtype": "float32",
187
  "transformers_version": "4.52.3",
188
  "use_cache": true,
189
  "vit_config": {
model-00001-of-00004.safetensors → model-00001-of-00007.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5ae6fc6b4d0e55be7083752d084e2b712bf9b2794b9f91512a04341cf366447b
3
- size 4878581216
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d091a7fcf85d7e5e5b9cf7a2bcfd33dd2a5165594d40cbc4b577556e2efbe41
3
+ size 4978520816
model-00002-of-00004.safetensors → model-00002-of-00007.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1db7d9da0b0b4284c1145576ebdf3a0c2e51fde4c9463350b28128b2faa3fc7f
3
- size 4932745864
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2469da3db18f67a7f2a0c8c32115b825ea1276568463e4311af31449451dfeb9
3
+ size 4778633920
model-00003-of-00004.safetensors → model-00003-of-00007.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:64f7efd5c440baf14fda858f3ff1832e9a359c0524795bf612118017b5ac6d0e
3
- size 4994552920
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f09e1217e65a6a54749ec5af4b681a321038f3dfc23a95d78824f72a2ee2e20b
3
+ size 4661160168
model-00004-of-00004.safetensors → model-00004-of-00007.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:340050ceda4128e93ce60caf2d2d5e999d7f3067beb2f0634224d837c0eafb9f
3
- size 1433042592
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:01b29df00b9972d5a548527774daaaae540c44cddf380bd2d3f7c922f633d6a5
3
+ size 4661160192
model-00005-of-00007.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f027c8de3a20c6bd6e13b2d2485aef7a114cc8d2a2878df6361ba6477645d40
3
+ size 4661160192
model-00006-of-00007.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a40ccc85e6fa499f8af46ca0b1d7bdfe241df001d29df351280fc8f41013c395
3
+ size 4997750712
model-00007-of-00007.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a8db94d0998ccf9a252d84bc12ea065cc8f66280c7177a7c9e7e69c505efe8c
3
+ size 3739371680
model.safetensors.index.json CHANGED
The diff for this file is too large to render. See raw diff
 
model.yaml ADDED
@@ -0,0 +1,253 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model_name: molmo
2
+ llm:
3
+ d_model: 3584
4
+ n_heads: 28
5
+ n_kv_heads: 4
6
+ head_dim: null
7
+ qkv_bias: true
8
+ clip_qkv: null
9
+ n_layers: 28
10
+ mlp_ratio: 4
11
+ mlp_hidden_size: 37888
12
+ activation_type: swiglu
13
+ block_type: sequential
14
+ rope: true
15
+ rope_full_precision: true
16
+ rope_theta: 1000000.0
17
+ rope_type: default
18
+ rope_factor: null
19
+ rope_high_freq_factor: null
20
+ rope_low_freq_factor: null
21
+ rope_original_max_position_embeddings: null
22
+ attention_type: sdpa
23
+ float32_attention: true
24
+ attention_dropout: 0.0
25
+ attention_layer_norm: false
26
+ attention_layer_norm_type: olmo
27
+ residual_dropout: 0.1
28
+ response_residual_dropout: 0.0
29
+ layer_norm_type: rms
30
+ layer_norm_with_affine: true
31
+ layer_norm_eps: 1.0e-06
32
+ attention_layer_norm_with_affine: true
33
+ max_sequence_length: 4096
34
+ max_position_embeddings: null
35
+ include_bias: false
36
+ bias_for_layer_norm: null
37
+ norm_after: false
38
+ moe_num_experts: 8
39
+ moe_top_k: 2
40
+ moe_mlp_impl: sparse
41
+ moe_log_expert_assignment: false
42
+ moe_shared_expert: false
43
+ moe_lbl_in_fp32: false
44
+ moe_interleave: false
45
+ moe_loss_weight: 0.1
46
+ moe_zloss_weight: null
47
+ moe_dropless: true
48
+ moe_capacity_factor: 1.25
49
+ embedding_dropout: 0.0
50
+ scale_logits: false
51
+ vocab_size: 152064
52
+ additional_vocab_size: 128
53
+ weight_tying: false
54
+ embedding_size: 152064
55
+ use_position_ids: true
56
+ tokenizer:
57
+ identifier: Qwen/Qwen2.5-7B
58
+ tokenizer_dir: null
59
+ depth_tokens: true
60
+ init_path: gs://mm-olmo/pretrained_llms/qwen2.5-7b.pt
61
+ init_incremental: null
62
+ new_embedding_init_range: 0.02
63
+ initializer_range: 0.02
64
+ normalize_input_embeds: false
65
+ activation_checkpoint: whole_layer
66
+ compile: blocks
67
+ fix_pad_tokenizer: false
68
+ resize_vocab: false
69
+ init_std: 0.02
70
+ init_fn: normal
71
+ init_cutoff_factor: null
72
+ vision_backbone:
73
+ vit:
74
+ image_model_type: siglip
75
+ image_default_input_size:
76
+ - 378
77
+ - 378
78
+ image_patch_size: 14
79
+ image_pos_patch_size: 14
80
+ image_emb_dim: 1152
81
+ image_num_heads: 16
82
+ image_num_key_value_heads: 16
83
+ image_num_layers: 27
84
+ image_head_dim: 72
85
+ image_mlp_dim: 4304
86
+ image_mlp_activations: gelu_pytorch_tanh
87
+ image_dropout_rate: 0.0
88
+ image_num_pos: 729
89
+ image_norm_eps: 1.0e-06
90
+ attention_dropout: 0.0
91
+ residual_dropout: 0.0
92
+ initializer_range: 0.02
93
+ float32_attention: true
94
+ attention_type: sdpa
95
+ activation_checkpointing: true
96
+ init_path: gs://mm-olmo/pretrained_image_encoders/siglip2-so400m-14-384.pt
97
+ resize_mode: siglip
98
+ pad_value: 0.0
99
+ normalize: siglip
100
+ image_pooling_2d: attention_meanq
101
+ pooling_attention_mask: false
102
+ image_projector: mlp
103
+ image_padding_embed: null
104
+ vit_layers:
105
+ - -3
106
+ - -9
107
+ skip_unused_layers: true
108
+ image_feature_dropout: 0.0
109
+ connector_activation_checkpointing: true
110
+ compile_vit: blocks
111
+ data_formatter:
112
+ prompt_templates: uber_model
113
+ message_format: role
114
+ system_prompt: demo_or_style
115
+ always_start_with_space: false
116
+ default_inference_len: 65
117
+ select_answer: best
118
+ debug: false
119
+ image_last: false
120
+ format_message_list: null
121
+ p_one_message: 0.0
122
+ mm_preprocessor:
123
+ crop_mode: overlap-and-resize-c2
124
+ max_crops: 8
125
+ max_images: 2
126
+ max_multi_image_crops: 8
127
+ pooling_w: 2
128
+ pooling_h: 2
129
+ overlap_margins:
130
+ - 4
131
+ - 4
132
+ use_col_tokens: true
133
+ loss_token_weighting: root_subsegments
134
+ legacy_image_mask: false
135
+ max_answer_len: null
136
+ img_aug: false
137
+ bi_directional_attn: null
138
+ lora_enable: false
139
+ lora_rank: 64
140
+ lora_alpha: 16
141
+ lora_dropout: 0.05
142
+ lora_bias: none
143
+ n_action_bins: 256
144
+ norm_stats:
145
+ fractal20220817_data:
146
+ action:
147
+ mean:
148
+ - 0.006987582892179489
149
+ - 0.006265917327255011
150
+ - -0.01262515690177679
151
+ - 0.04333311319351196
152
+ - -0.005756212864071131
153
+ - 0.0009130256366916001
154
+ - 0.5354204773902893
155
+ std:
156
+ - 0.0692116990685463
157
+ - 0.05970962345600128
158
+ - 0.07353084534406662
159
+ - 0.15610496699810028
160
+ - 0.13164450228214264
161
+ - 0.14593800902366638
162
+ - 0.497110515832901
163
+ max:
164
+ - 2.9984593391418457
165
+ - 22.09052848815918
166
+ - 2.7507524490356445
167
+ - 1.570636510848999
168
+ - 1.5321086645126343
169
+ - 1.5691522359848022
170
+ - 1.0
171
+ min:
172
+ - -2.0204520225524902
173
+ - -5.497899532318115
174
+ - -2.031663417816162
175
+ - -1.569917917251587
176
+ - -1.569892168045044
177
+ - -1.570419430732727
178
+ - 0.0
179
+ q01:
180
+ - -0.22453527510166169
181
+ - -0.14820013284683228
182
+ - -0.231589707583189
183
+ - -0.3517994859814644
184
+ - -0.4193011274933815
185
+ - -0.43643461108207704
186
+ - 0.0
187
+ q99:
188
+ - 0.17824687153100965
189
+ - 0.14938379630446405
190
+ - 0.21842354819178575
191
+ - 0.5892666035890578
192
+ - 0.35272657424211445
193
+ - 0.44796681255102094
194
+ - 1.0
195
+ mask:
196
+ - true
197
+ - true
198
+ - true
199
+ - true
200
+ - true
201
+ - true
202
+ - false
203
+ proprio:
204
+ mean:
205
+ - 0.0
206
+ - 0.0
207
+ - 0.0
208
+ - 0.0
209
+ - 0.0
210
+ - 0.0
211
+ - 0.0
212
+ std:
213
+ - 0.0
214
+ - 0.0
215
+ - 0.0
216
+ - 0.0
217
+ - 0.0
218
+ - 0.0
219
+ - 0.0
220
+ max:
221
+ - 0.0
222
+ - 0.0
223
+ - 0.0
224
+ - 0.0
225
+ - 0.0
226
+ - 0.0
227
+ - 0.0
228
+ min:
229
+ - 0.0
230
+ - 0.0
231
+ - 0.0
232
+ - 0.0
233
+ - 0.0
234
+ - 0.0
235
+ - 0.0
236
+ q01:
237
+ - 0.0
238
+ - 0.0
239
+ - 0.0
240
+ - 0.0
241
+ - 0.0
242
+ - 0.0
243
+ - 0.0
244
+ q99:
245
+ - 0.0
246
+ - 0.0
247
+ - 0.0
248
+ - 0.0
249
+ - 0.0
250
+ - 0.0
251
+ - 0.0
252
+ num_transitions: 3786400
253
+ num_trajectories: 87212