Hugging Face's logo

ds4sd
/
SmolDocling-256M-preview-mlx-bf16

Image-Text-to-Text
Transformers
Safetensors
MLX
English
idefics3
conversational
Inference Endpoints
Model card Files Community
1
auerchristoph commited on
Commit
1f62407
·
verified ·
1 Parent(s): d7fc901

Upload folder using huggingface_hub

Browse files
Files changed (14) hide show
  1. README.md +24 -3
  2. added_tokens.json +130 -0
  3. chat_template.json +3 -0
  4. config.json +322 -0
  5. merges.txt +0 -0
  6. model.safetensors +3 -0
  7. model.safetensors.index.json +478 -0
  8. preprocessor_config.json +28 -0
  9. processor_config.json +4 -0
  10. special_tokens_map.json +53 -0
  11. tokenizer.json +0 -0
  12. tokenizer_config.json +1189 -0
  13. vocab.json +0 -0
  14. zero_to_fp32.py +760 -0
README.md CHANGED
@@ -1,3 +1,24 @@
1
- ---
2
- license: cdla-permissive-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model:
3
+ - HuggingFaceTB/SmolVLM-256M-Instruct
4
+ language:
5
+ - en
6
+ library_name: transformers
7
+ license: apache-2.0
8
+ pipeline_tag: image-text-to-text
9
+ tags:
10
+ - mlx
11
+ ---
12
+
13
+ # ds4sd/SmolDocling-256M-preview-mlx-bf16
14
+ This model was converted to MLX format from [`ds4sd/SmolDocling-256M-preview`]() using mlx-vlm version **0.1.18**.
15
+ Refer to the [original model card](https://huggingface.co/ds4sd/SmolDocling-256M-preview) for more details on the model.
16
+ ## Use with mlx
17
+
18
+ ```bash
19
+ pip install -U mlx-vlm
20
+ ```
21
+
22
+ ```bash
23
+ python -m mlx_vlm.generate --model ds4sd/SmolDocling-256M-preview-mlx-bf16 --max-tokens 100 --temperature 0.0 --prompt "Describe this image." --image <path_to_image>
24
+ ```
added_tokens.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</caption>": 49192,
3
+ "</chart>": 49248,
4
+ "</checkbox_selected>": 49211,
5
+ "</checkbox_unselected>": 49213,
6
+ "</doctag>": 49230,
7
+ "</footnote>": 49195,
8
+ "</form>": 49215,
9
+ "</formula>": 49197,
10
+ "</group>": 49228,
11
+ "</key_": 49243,
12
+ "</key_value_region>": 49217,
13
+ "</list_item>": 49199,
14
+ "</ordered_list>": 49224,
15
+ "</otsl>": 49209,
16
+ "</page_footer>": 49201,
17
+ "</page_header>": 49203,
18
+ "</paragraph>": 49220,
19
+ "</picture>": 49205,
20
+ "</reference>": 49222,
21
+ "</section_header_level_": 49207,
22
+ "</smiles>": 49251,
23
+ "</unordered_list>": 49226,
24
+ "</value_": 49245,
25
+ "<caption>": 49191,
26
+ "<chart>": 49247,
27
+ "<checkbox_selected>": 49210,
28
+ "<checkbox_unselected>": 49212,
29
+ "<ched>": 49239,
30
+ "<doctag>": 49229,
31
+ "<ecel>": 49234,
32
+ "<end_of_utterance>": 49279,
33
+ "<fake_token_around_image>": 49189,
34
+ "<fcel>": 49233,
35
+ "<footnote>": 49193,
36
+ "<form>": 49214,
37
+ "<formula>": 49196,
38
+ "<global-img>": 49152,
39
+ "<group>": 49227,
40
+ "<image>": 49190,
41
+ "<key_": 49242,
42
+ "<key_value_region>": 49216,
43
+ "<lcel>": 49235,
44
+ "<link_": 49246,
45
+ "<list_item>": 49198,
46
+ "<loc_": 49218,
47
+ "<nl>": 49238,
48
+ "<ordered_list>": 49223,
49
+ "<otsl>": 49208,
50
+ "<page_": 49231,
51
+ "<page_break>": 49249,
52
+ "<page_footer>": 49200,
53
+ "<page_header>": 49202,
54
+ "<paragraph>": 49219,
55
+ "<picture>": 49204,
56
+ "<reference>": 49221,
57
+ "<rhed>": 49240,
58
+ "<row_1_col_1>": 49153,
59
+ "<row_1_col_2>": 49154,
60
+ "<row_1_col_3>": 49155,
61
+ "<row_1_col_4>": 49156,
62
+ "<row_1_col_5>": 49157,
63
+ "<row_1_col_6>": 49158,
64
+ "<row_2_col_1>": 49159,
65
+ "<row_2_col_2>": 49160,
66
+ "<row_2_col_3>": 49161,
67
+ "<row_2_col_4>": 49162,
68
+ "<row_2_col_5>": 49163,
69
+ "<row_2_col_6>": 49164,
70
+ "<row_3_col_1>": 49165,
71
+ "<row_3_col_2>": 49166,
72
+ "<row_3_col_3>": 49167,
73
+ "<row_3_col_4>": 49168,
74
+ "<row_3_col_5>": 49169,
75
+ "<row_3_col_6>": 49170,
76
+ "<row_4_col_1>": 49171,
77
+ "<row_4_col_2>": 49172,
78
+ "<row_4_col_3>": 49173,
79
+ "<row_4_col_4>": 49174,
80
+ "<row_4_col_5>": 49175,
81
+ "<row_4_col_6>": 49176,
82
+ "<row_5_col_1>": 49177,
83
+ "<row_5_col_2>": 49178,
84
+ "<row_5_col_3>": 49179,
85
+ "<row_5_col_4>": 49180,
86
+ "<row_5_col_5>": 49181,
87
+ "<row_5_col_6>": 49182,
88
+ "<row_6_col_1>": 49183,
89
+ "<row_6_col_2>": 49184,
90
+ "<row_6_col_3>": 49185,
91
+ "<row_6_col_4>": 49186,
92
+ "<row_6_col_5>": 49187,
93
+ "<row_6_col_6>": 49188,
94
+ "<section_header_level_": 49206,
95
+ "<smiles>": 49250,
96
+ "<text_break>": 49232,
97
+ "<ucel>": 49236,
98
+ "<unordered_list>": 49225,
99
+ "<value_": 49244,
100
+ "<xcel>": 49237,
101
+ "<|reserved_special_token_3|>": 49194,
102
+ "<|reserved_special_token_50|>": 49241,
103
+ "<|reserved_special_token_61|>": 49252,
104
+ "<|reserved_special_token_62|>": 49253,
105
+ "<|reserved_special_token_63|>": 49254,
106
+ "<|reserved_special_token_64|>": 49255,
107
+ "<|reserved_special_token_65|>": 49256,
108
+ "<|reserved_special_token_66|>": 49257,
109
+ "<|reserved_special_token_67|>": 49258,
110
+ "<|reserved_special_token_68|>": 49259,
111
+ "<|reserved_special_token_69|>": 49260,
112
+ "<|reserved_special_token_70|>": 49261,
113
+ "<|reserved_special_token_71|>": 49262,
114
+ "<|reserved_special_token_72|>": 49263,
115
+ "<|reserved_special_token_73|>": 49264,
116
+ "<|reserved_special_token_74|>": 49265,
117
+ "<|reserved_special_token_75|>": 49266,
118
+ "<|reserved_special_token_76|>": 49267,
119
+ "<|reserved_special_token_77|>": 49268,
120
+ "<|reserved_special_token_78|>": 49269,
121
+ "<|reserved_special_token_79|>": 49270,
122
+ "<|reserved_special_token_80|>": 49271,
123
+ "<|reserved_special_token_81|>": 49272,
124
+ "<|reserved_special_token_82|>": 49273,
125
+ "<|reserved_special_token_83|>": 49274,
126
+ "<|reserved_special_token_84|>": 49275,
127
+ "<|reserved_special_token_85|>": 49276,
128
+ "<|reserved_special_token_86|>": 49277,
129
+ "<|reserved_special_token_87|>": 49278
130
+ }
chat_template.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "chat_template": "<|im_start|>{% for message in messages %}{{message['role'] | capitalize}}{% if message['content'][0]['type'] == 'image' %}{{':'}}{% else %}{{': '}}{% endif %}{% for line in message['content'] %}{% if line['type'] == 'text' %}{{line['text']}}{% elif line['type'] == 'image' %}{{ '<image>' }}{% endif %}{% endfor %}<end_of_utterance>\n{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}"
3
+ }
config.json ADDED
@@ -0,0 +1,322 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_attn_implementation_autoset": false,
3
+ "add_cross_attention": false,
4
+ "architectures": [
5
+ "Idefics3ForConditionalGeneration"
6
+ ],
7
+ "bad_words_ids": null,
8
+ "begin_suppress_tokens": null,
9
+ "bos_token_id": null,
10
+ "chunk_size_feed_forward": 0,
11
+ "cross_attention_hidden_size": null,
12
+ "decoder_start_token_id": null,
13
+ "diversity_penalty": 0.0,
14
+ "do_sample": false,
15
+ "early_stopping": false,
16
+ "encoder_no_repeat_ngram_size": 0,
17
+ "eos_token_id": null,
18
+ "exponential_decay_length_penalty": null,
19
+ "finetuning_task": null,
20
+ "forced_bos_token_id": null,
21
+ "forced_eos_token_id": null,
22
+ "id2label": {
23
+ "0": "LABEL_0",
24
+ "1": "LABEL_1"
25
+ },
26
+ "image_token_id": 49190,
27
+ "is_decoder": false,
28
+ "is_encoder_decoder": false,
29
+ "label2id": {
30
+ "LABEL_0": 0,
31
+ "LABEL_1": 1
32
+ },
33
+ "length_penalty": 1.0,
34
+ "max_length": 20,
35
+ "min_length": 0,
36
+ "model_type": "idefics3",
37
+ "no_repeat_ngram_size": 0,
38
+ "num_beam_groups": 1,
39
+ "num_beams": 1,
40
+ "num_return_sequences": 1,
41
+ "output_attentions": false,
42
+ "output_hidden_states": false,
43
+ "output_scores": false,
44
+ "pad_token_id": 128002,
45
+ "prefix": null,
46
+ "problem_type": null,
47
+ "pruned_heads": {},
48
+ "remove_invalid_values": false,
49
+ "repetition_penalty": 1.0,
50
+ "return_dict": true,
51
+ "return_dict_in_generate": false,
52
+ "scale_factor": 4,
53
+ "sep_token_id": null,
54
+ "suppress_tokens": null,
55
+ "task_specific_params": null,
56
+ "temperature": 1.0,
57
+ "text_config": {
58
+ "vocab_size": 49280,
59
+ "max_position_embeddings": 8192,
60
+ "hidden_size": 576,
61
+ "intermediate_size": 1536,
62
+ "num_hidden_layers": 30,
63
+ "num_attention_heads": 9,
64
+ "num_key_value_heads": 3,
65
+ "hidden_act": "silu",
66
+ "initializer_range": 0.041666666666666664,
67
+ "rms_norm_eps": 1e-05,
68
+ "pretraining_tp": 1,
69
+ "use_cache": true,
70
+ "rope_theta": 100000,
71
+ "rope_scaling": null,
72
+ "attention_bias": false,
73
+ "attention_dropout": 0.0,
74
+ "mlp_bias": false,
75
+ "head_dim": 64,
76
+ "return_dict": true,
77
+ "output_hidden_states": false,
78
+ "output_attentions": false,
79
+ "torchscript": false,
80
+ "torch_dtype": "bfloat16",
81
+ "use_bfloat16": false,
82
+ "tf_legacy_loss": false,
83
+ "pruned_heads": {},
84
+ "tie_word_embeddings": false,
85
+ "chunk_size_feed_forward": 0,
86
+ "is_encoder_decoder": false,
87
+ "is_decoder": false,
88
+ "cross_attention_hidden_size": null,
89
+ "add_cross_attention": false,
90
+ "tie_encoder_decoder": false,
91
+ "max_length": 20,
92
+ "min_length": 0,
93
+ "do_sample": false,
94
+ "early_stopping": false,
95
+ "num_beams": 1,
96
+ "num_beam_groups": 1,
97
+ "diversity_penalty": 0.0,
98
+ "temperature": 1.0,
99
+ "top_k": 50,
100
+ "top_p": 1.0,
101
+ "typical_p": 1.0,
102
+ "repetition_penalty": 1.0,
103
+ "length_penalty": 1.0,
104
+ "no_repeat_ngram_size": 0,
105
+ "encoder_no_repeat_ngram_size": 0,
106
+ "bad_words_ids": null,
107
+ "num_return_sequences": 1,
108
+ "output_scores": false,
109
+ "return_dict_in_generate": false,
110
+ "forced_bos_token_id": null,
111
+ "forced_eos_token_id": null,
112
+ "remove_invalid_values": false,
113
+ "exponential_decay_length_penalty": null,
114
+ "suppress_tokens": null,
115
+ "begin_suppress_tokens": null,
116
+ "architectures": [
117
+ "VLlama3ForCausalLM"
118
+ ],
119
+ "finetuning_task": null,
120
+ "id2label": {
121
+ "0": "LABEL_0",
122
+ "1": "LABEL_1"
123
+ },
124
+ "label2id": {
125
+ "LABEL_0": 0,
126
+ "LABEL_1": 1
127
+ },
128
+ "tokenizer_class": null,
129
+ "prefix": null,
130
+ "bos_token_id": 1,
131
+ "pad_token_id": 2,
132
+ "eos_token_id": 2,
133
+ "sep_token_id": null,
134
+ "decoder_start_token_id": null,
135
+ "task_specific_params": null,
136
+ "problem_type": null,
137
+ "_name_or_path": "None",
138
+ "_attn_implementation_autoset": false,
139
+ "_flash_attn_2_enabled": true,
140
+ "is_llama_config": true,
141
+ "model_type": "llama",
142
+ "neftune_noise_alpha": 0.0,
143
+ "perceiver_config": {
144
+ "_attn_implementation_autoset": false,
145
+ "_name_or_path": "",
146
+ "add_cross_attention": false,
147
+ "architectures": null,
148
+ "attention_dropout": 0.0,
149
+ "bad_words_ids": null,
150
+ "begin_suppress_tokens": null,
151
+ "bos_token_id": null,
152
+ "chunk_size_feed_forward": 0,
153
+ "cross_attention_hidden_size": null,
154
+ "decoder_start_token_id": null,
155
+ "diversity_penalty": 0.0,
156
+ "do_sample": false,
157
+ "early_stopping": false,
158
+ "encoder_no_repeat_ngram_size": 0,
159
+ "eos_token_id": null,
160
+ "exponential_decay_length_penalty": null,
161
+ "finetuning_task": null,
162
+ "forced_bos_token_id": null,
163
+ "forced_eos_token_id": null,
164
+ "hidden_act": "silu",
165
+ "id2label": {
166
+ "0": "LABEL_0",
167
+ "1": "LABEL_1"
168
+ },
169
+ "is_decoder": false,
170
+ "is_encoder_decoder": false,
171
+ "label2id": {
172
+ "LABEL_0": 0,
173
+ "LABEL_1": 1
174
+ },
175
+ "length_penalty": 1.0,
176
+ "max_length": 20,
177
+ "min_length": 0,
178
+ "model_type": "vllama3",
179
+ "no_repeat_ngram_size": 0,
180
+ "num_beam_groups": 1,
181
+ "num_beams": 1,
182
+ "num_key_value_heads": 1,
183
+ "num_return_sequences": 1,
184
+ "output_attentions": false,
185
+ "output_hidden_states": false,
186
+ "output_scores": false,
187
+ "pad_token_id": null,
188
+ "prefix": null,
189
+ "problem_type": null,
190
+ "pruned_heads": {},
191
+ "qk_layer_norms_perceiver": false,
192
+ "remove_invalid_values": false,
193
+ "repetition_penalty": 1.0,
194
+ "resampler_depth": 6,
195
+ "resampler_head_dim": 96,
196
+ "resampler_n_heads": 16,
197
+ "resampler_n_latents": 64,
198
+ "return_dict": true,
199
+ "return_dict_in_generate": false,
200
+ "sep_token_id": null,
201
+ "suppress_tokens": null,
202
+ "task_specific_params": null,
203
+ "temperature": 1.0,
204
+ "tf_legacy_loss": false,
205
+ "tie_encoder_decoder": false,
206
+ "tie_word_embeddings": true,
207
+ "tokenizer_class": null,
208
+ "top_k": 50,
209
+ "top_p": 1.0,
210
+ "torch_dtype": null,
211
+ "torchscript": false,
212
+ "transformers_version": "4.46.0",
213
+ "typical_p": 1.0,
214
+ "use_bfloat16": false
215
+ },
216
+ "pixel_shuffle_factor": 4,
217
+ "qk_layer_norms": false,
218
+ "rope_interleaved": false,
219
+ "transformers.js_config": {
220
+ "kv_cache_dtype": {
221
+ "fp16": "float16",
222
+ "q4f16": "float16"
223
+ }
224
+ },
225
+ "use_resampler": false
226
+ },
227
+ "tf_legacy_loss": false,
228
+ "tie_encoder_decoder": false,
229
+ "tie_word_embeddings": false,
230
+ "tokenizer_class": null,
231
+ "top_k": 50,
232
+ "top_p": 1.0,
233
+ "torch_dtype": "bfloat16",
234
+ "torchscript": false,
235
+ "transformers_version": "4.49.0",
236
+ "typical_p": 1.0,
237
+ "use_bfloat16": false,
238
+ "use_cache": true,
239
+ "vision_config": {
240
+ "return_dict": true,
241
+ "output_hidden_states": false,
242
+ "output_attentions": false,
243
+ "torchscript": false,
244
+ "torch_dtype": "bfloat16",
245
+ "use_bfloat16": false,
246
+ "tf_legacy_loss": false,
247
+ "pruned_heads": {},
248
+ "tie_word_embeddings": false,
249
+ "chunk_size_feed_forward": 0,
250
+ "is_encoder_decoder": false,
251
+ "is_decoder": false,
252
+ "cross_attention_hidden_size": null,
253
+ "add_cross_attention": false,
254
+ "tie_encoder_decoder": false,
255
+ "max_length": 20,
256
+ "min_length": 0,
257
+ "do_sample": false,
258
+ "early_stopping": false,
259
+ "num_beams": 1,
260
+ "num_beam_groups": 1,
261
+ "diversity_penalty": 0.0,
262
+ "temperature": 1.0,
263
+ "top_k": 50,
264
+ "top_p": 1.0,
265
+ "typical_p": 1.0,
266
+ "repetition_penalty": 1.0,
267
+ "length_penalty": 1.0,
268
+ "no_repeat_ngram_size": 0,
269
+ "encoder_no_repeat_ngram_size": 0,
270
+ "bad_words_ids": null,
271
+ "num_return_sequences": 1,
272
+ "output_scores": false,
273
+ "return_dict_in_generate": false,
274
+ "forced_bos_token_id": null,
275
+ "forced_eos_token_id": null,
276
+ "remove_invalid_values": false,
277
+ "exponential_decay_length_penalty": null,
278
+ "suppress_tokens": null,
279
+ "begin_suppress_tokens": null,
280
+ "architectures": null,
281
+ "finetuning_task": null,
282
+ "id2label": {
283
+ "0": "LABEL_0",
284
+ "1": "LABEL_1"
285
+ },
286
+ "label2id": {
287
+ "LABEL_0": 0,
288
+ "LABEL_1": 1
289
+ },
290
+ "tokenizer_class": null,
291
+ "prefix": null,
292
+ "bos_token_id": null,
293
+ "pad_token_id": null,
294
+ "eos_token_id": null,
295
+ "sep_token_id": null,
296
+ "decoder_start_token_id": null,
297
+ "task_specific_params": null,
298
+ "problem_type": null,
299
+ "_name_or_path": "",
300
+ "_attn_implementation_autoset": false,
301
+ "max_image_size": {
302
+ "longest_edge": 512
303
+ },
304
+ "model_type": "idefics3_vision",
305
+ "size": {
306
+ "longest_edge": 2048
307
+ },
308
+ "use_base_siglip": true,
309
+ "hidden_size": 768,
310
+ "intermediate_size": 3072,
311
+ "num_hidden_layers": 12,
312
+ "num_attention_heads": 12,
313
+ "num_channels": 3,
314
+ "patch_size": 16,
315
+ "image_size": 512,
316
+ "attention_dropout": 0.0,
317
+ "layer_norm_eps": 1e-06,
318
+ "hidden_act": "gelu_pytorch_tanh",
319
+ "initializer_range": 0.02
320
+ },
321
+ "vocab_size": 49280
322
+ }
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c7a39b5bdf18205a9c6fd0d2f1fc95bbaaea2cd014de43d6c6450e5c15441dbe
3
+ size 513026834
model.safetensors.index.json ADDED
@@ -0,0 +1,478 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 512969856
4
+ },
5
+ "weight_map": {
6
+ "connector.modality_projection.proj.weight": "model.safetensors",
7
+ "language_model.embed_tokens.weight": "model.safetensors",
8
+ "language_model.layers.0.input_layernorm.weight": "model.safetensors",
9
+ "language_model.layers.0.mlp.down_proj.weight": "model.safetensors",
10
+ "language_model.layers.0.mlp.gate_proj.weight": "model.safetensors",
11
+ "language_model.layers.0.mlp.up_proj.weight": "model.safetensors",
12
+ "language_model.layers.0.post_attention_layernorm.weight": "model.safetensors",
13
+ "language_model.layers.0.self_attn.k_proj.weight": "model.safetensors",
14
+ "language_model.layers.0.self_attn.o_proj.weight": "model.safetensors",
15
+ "language_model.layers.0.self_attn.q_proj.weight": "model.safetensors",
16
+ "language_model.layers.0.self_attn.v_proj.weight": "model.safetensors",
17
+ "language_model.layers.1.input_layernorm.weight": "model.safetensors",
18
+ "language_model.layers.1.mlp.down_proj.weight": "model.safetensors",
19
+ "language_model.layers.1.mlp.gate_proj.weight": "model.safetensors",
20
+ "language_model.layers.1.mlp.up_proj.weight": "model.safetensors",
21
+ "language_model.layers.1.post_attention_layernorm.weight": "model.safetensors",
22
+ "language_model.layers.1.self_attn.k_proj.weight": "model.safetensors",
23
+ "language_model.layers.1.self_attn.o_proj.weight": "model.safetensors",
24
+ "language_model.layers.1.self_attn.q_proj.weight": "model.safetensors",
25
+ "language_model.layers.1.self_attn.v_proj.weight": "model.safetensors",
26
+ "language_model.layers.10.input_layernorm.weight": "model.safetensors",
27
+ "language_model.layers.10.mlp.down_proj.weight": "model.safetensors",
28
+ "language_model.layers.10.mlp.gate_proj.weight": "model.safetensors",
29
+ "language_model.layers.10.mlp.up_proj.weight": "model.safetensors",
30
+ "language_model.layers.10.post_attention_layernorm.weight": "model.safetensors",
31
+ "language_model.layers.10.self_attn.k_proj.weight": "model.safetensors",
32
+ "language_model.layers.10.self_attn.o_proj.weight": "model.safetensors",
33
+ "language_model.layers.10.self_attn.q_proj.weight": "model.safetensors",
34
+ "language_model.layers.10.self_attn.v_proj.weight": "model.safetensors",
35
+ "language_model.layers.11.input_layernorm.weight": "model.safetensors",
36
+ "language_model.layers.11.mlp.down_proj.weight": "model.safetensors",
37
+ "language_model.layers.11.mlp.gate_proj.weight": "model.safetensors",
38
+ "language_model.layers.11.mlp.up_proj.weight": "model.safetensors",
39
+ "language_model.layers.11.post_attention_layernorm.weight": "model.safetensors",
40
+ "language_model.layers.11.self_attn.k_proj.weight": "model.safetensors",
41
+ "language_model.layers.11.self_attn.o_proj.weight": "model.safetensors",
42
+ "language_model.layers.11.self_attn.q_proj.weight": "model.safetensors",
43
+ "language_model.layers.11.self_attn.v_proj.weight": "model.safetensors",
44
+ "language_model.layers.12.input_layernorm.weight": "model.safetensors",
45
+ "language_model.layers.12.mlp.down_proj.weight": "model.safetensors",
46
+ "language_model.layers.12.mlp.gate_proj.weight": "model.safetensors",
47
+ "language_model.layers.12.mlp.up_proj.weight": "model.safetensors",
48
+ "language_model.layers.12.post_attention_layernorm.weight": "model.safetensors",
49
+ "language_model.layers.12.self_attn.k_proj.weight": "model.safetensors",
50
+ "language_model.layers.12.self_attn.o_proj.weight": "model.safetensors",
51
+ "language_model.layers.12.self_attn.q_proj.weight": "model.safetensors",
52
+ "language_model.layers.12.self_attn.v_proj.weight": "model.safetensors",
53
+ "language_model.layers.13.input_layernorm.weight": "model.safetensors",
54
+ "language_model.layers.13.mlp.down_proj.weight": "model.safetensors",
55
+ "language_model.layers.13.mlp.gate_proj.weight": "model.safetensors",
56
+ "language_model.layers.13.mlp.up_proj.weight": "model.safetensors",
57
+ "language_model.layers.13.post_attention_layernorm.weight": "model.safetensors",
58
+ "language_model.layers.13.self_attn.k_proj.weight": "model.safetensors",
59
+ "language_model.layers.13.self_attn.o_proj.weight": "model.safetensors",
60
+ "language_model.layers.13.self_attn.q_proj.weight": "model.safetensors",
61
+ "language_model.layers.13.self_attn.v_proj.weight": "model.safetensors",
62
+ "language_model.layers.14.input_layernorm.weight": "model.safetensors",
63
+ "language_model.layers.14.mlp.down_proj.weight": "model.safetensors",
64
+ "language_model.layers.14.mlp.gate_proj.weight": "model.safetensors",
65
+ "language_model.layers.14.mlp.up_proj.weight": "model.safetensors",
66
+ "language_model.layers.14.post_attention_layernorm.weight": "model.safetensors",
67
+ "language_model.layers.14.self_attn.k_proj.weight": "model.safetensors",
68
+ "language_model.layers.14.self_attn.o_proj.weight": "model.safetensors",
69
+ "language_model.layers.14.self_attn.q_proj.weight": "model.safetensors",
70
+ "language_model.layers.14.self_attn.v_proj.weight": "model.safetensors",
71
+ "language_model.layers.15.input_layernorm.weight": "model.safetensors",
72
+ "language_model.layers.15.mlp.down_proj.weight": "model.safetensors",
73
+ "language_model.layers.15.mlp.gate_proj.weight": "model.safetensors",
74
+ "language_model.layers.15.mlp.up_proj.weight": "model.safetensors",
75
+ "language_model.layers.15.post_attention_layernorm.weight": "model.safetensors",
76
+ "language_model.layers.15.self_attn.k_proj.weight": "model.safetensors",
77
+ "language_model.layers.15.self_attn.o_proj.weight": "model.safetensors",
78
+ "language_model.layers.15.self_attn.q_proj.weight": "model.safetensors",
79
+ "language_model.layers.15.self_attn.v_proj.weight": "model.safetensors",
80
+ "language_model.layers.16.input_layernorm.weight": "model.safetensors",
81
+ "language_model.layers.16.mlp.down_proj.weight": "model.safetensors",
82
+ "language_model.layers.16.mlp.gate_proj.weight": "model.safetensors",
83
+ "language_model.layers.16.mlp.up_proj.weight": "model.safetensors",
84
+ "language_model.layers.16.post_attention_layernorm.weight": "model.safetensors",
85
+ "language_model.layers.16.self_attn.k_proj.weight": "model.safetensors",
86
+ "language_model.layers.16.self_attn.o_proj.weight": "model.safetensors",
87
+ "language_model.layers.16.self_attn.q_proj.weight": "model.safetensors",
88
+ "language_model.layers.16.self_attn.v_proj.weight": "model.safetensors",
89
+ "language_model.layers.17.input_layernorm.weight": "model.safetensors",
90
+ "language_model.layers.17.mlp.down_proj.weight": "model.safetensors",
91
+ "language_model.layers.17.mlp.gate_proj.weight": "model.safetensors",
92
+ "language_model.layers.17.mlp.up_proj.weight": "model.safetensors",
93
+ "language_model.layers.17.post_attention_layernorm.weight": "model.safetensors",
94
+ "language_model.layers.17.self_attn.k_proj.weight": "model.safetensors",
95
+ "language_model.layers.17.self_attn.o_proj.weight": "model.safetensors",
96
+ "language_model.layers.17.self_attn.q_proj.weight": "model.safetensors",
97
+ "language_model.layers.17.self_attn.v_proj.weight": "model.safetensors",
98
+ "language_model.layers.18.input_layernorm.weight": "model.safetensors",
99
+ "language_model.layers.18.mlp.down_proj.weight": "model.safetensors",
100
+ "language_model.layers.18.mlp.gate_proj.weight": "model.safetensors",
101
+ "language_model.layers.18.mlp.up_proj.weight": "model.safetensors",
102
+ "language_model.layers.18.post_attention_layernorm.weight": "model.safetensors",
103
+ "language_model.layers.18.self_attn.k_proj.weight": "model.safetensors",
104
+ "language_model.layers.18.self_attn.o_proj.weight": "model.safetensors",
105
+ "language_model.layers.18.self_attn.q_proj.weight": "model.safetensors",
106
+ "language_model.layers.18.self_attn.v_proj.weight": "model.safetensors",
107
+ "language_model.layers.19.input_layernorm.weight": "model.safetensors",
108
+ "language_model.layers.19.mlp.down_proj.weight": "model.safetensors",
109
+ "language_model.layers.19.mlp.gate_proj.weight": "model.safetensors",
110
+ "language_model.layers.19.mlp.up_proj.weight": "model.safetensors",
111
+ "language_model.layers.19.post_attention_layernorm.weight": "model.safetensors",
112
+ "language_model.layers.19.self_attn.k_proj.weight": "model.safetensors",
113
+ "language_model.layers.19.self_attn.o_proj.weight": "model.safetensors",
114
+ "language_model.layers.19.self_attn.q_proj.weight": "model.safetensors",
115
+ "language_model.layers.19.self_attn.v_proj.weight": "model.safetensors",
116
+ "language_model.layers.2.input_layernorm.weight": "model.safetensors",
117
+ "language_model.layers.2.mlp.down_proj.weight": "model.safetensors",
118
+ "language_model.layers.2.mlp.gate_proj.weight": "model.safetensors",
119
+ "language_model.layers.2.mlp.up_proj.weight": "model.safetensors",
120
+ "language_model.layers.2.post_attention_layernorm.weight": "model.safetensors",
121
+ "language_model.layers.2.self_attn.k_proj.weight": "model.safetensors",
122
+ "language_model.layers.2.self_attn.o_proj.weight": "model.safetensors",
123
+ "language_model.layers.2.self_attn.q_proj.weight": "model.safetensors",
124
+ "language_model.layers.2.self_attn.v_proj.weight": "model.safetensors",
125
+ "language_model.layers.20.input_layernorm.weight": "model.safetensors",
126
+ "language_model.layers.20.mlp.down_proj.weight": "model.safetensors",
127
+ "language_model.layers.20.mlp.gate_proj.weight": "model.safetensors",
128
+ "language_model.layers.20.mlp.up_proj.weight": "model.safetensors",
129
+ "language_model.layers.20.post_attention_layernorm.weight": "model.safetensors",
130
+ "language_model.layers.20.self_attn.k_proj.weight": "model.safetensors",
131
+ "language_model.layers.20.self_attn.o_proj.weight": "model.safetensors",
132
+ "language_model.layers.20.self_attn.q_proj.weight": "model.safetensors",
133
+ "language_model.layers.20.self_attn.v_proj.weight": "model.safetensors",
134
+ "language_model.layers.21.input_layernorm.weight": "model.safetensors",
135
+ "language_model.layers.21.mlp.down_proj.weight": "model.safetensors",
136
+ "language_model.layers.21.mlp.gate_proj.weight": "model.safetensors",
137
+ "language_model.layers.21.mlp.up_proj.weight": "model.safetensors",
138
+ "language_model.layers.21.post_attention_layernorm.weight": "model.safetensors",
139
+ "language_model.layers.21.self_attn.k_proj.weight": "model.safetensors",
140
+ "language_model.layers.21.self_attn.o_proj.weight": "model.safetensors",
141
+ "language_model.layers.21.self_attn.q_proj.weight": "model.safetensors",
142
+ "language_model.layers.21.self_attn.v_proj.weight": "model.safetensors",
143
+ "language_model.layers.22.input_layernorm.weight": "model.safetensors",
144
+ "language_model.layers.22.mlp.down_proj.weight": "model.safetensors",
145
+ "language_model.layers.22.mlp.gate_proj.weight": "model.safetensors",
146
+ "language_model.layers.22.mlp.up_proj.weight": "model.safetensors",
147
+ "language_model.layers.22.post_attention_layernorm.weight": "model.safetensors",
148
+ "language_model.layers.22.self_attn.k_proj.weight": "model.safetensors",
149
+ "language_model.layers.22.self_attn.o_proj.weight": "model.safetensors",
150
+ "language_model.layers.22.self_attn.q_proj.weight": "model.safetensors",
151
+ "language_model.layers.22.self_attn.v_proj.weight": "model.safetensors",
152
+ "language_model.layers.23.input_layernorm.weight": "model.safetensors",
153
+ "language_model.layers.23.mlp.down_proj.weight": "model.safetensors",
154
+ "language_model.layers.23.mlp.gate_proj.weight": "model.safetensors",
155
+ "language_model.layers.23.mlp.up_proj.weight": "model.safetensors",
156
+ "language_model.layers.23.post_attention_layernorm.weight": "model.safetensors",
157
+ "language_model.layers.23.self_attn.k_proj.weight": "model.safetensors",
158
+ "language_model.layers.23.self_attn.o_proj.weight": "model.safetensors",
159
+ "language_model.layers.23.self_attn.q_proj.weight": "model.safetensors",
160
+ "language_model.layers.23.self_attn.v_proj.weight": "model.safetensors",
161
+ "language_model.layers.24.input_layernorm.weight": "model.safetensors",
162
+ "language_model.layers.24.mlp.down_proj.weight": "model.safetensors",
163
+ "language_model.layers.24.mlp.gate_proj.weight": "model.safetensors",
164
+ "language_model.layers.24.mlp.up_proj.weight": "model.safetensors",
165
+ "language_model.layers.24.post_attention_layernorm.weight": "model.safetensors",
166
+ "language_model.layers.24.self_attn.k_proj.weight": "model.safetensors",
167
+ "language_model.layers.24.self_attn.o_proj.weight": "model.safetensors",
168
+ "language_model.layers.24.self_attn.q_proj.weight": "model.safetensors",
169
+ "language_model.layers.24.self_attn.v_proj.weight": "model.safetensors",
170
+ "language_model.layers.25.input_layernorm.weight": "model.safetensors",
171
+ "language_model.layers.25.mlp.down_proj.weight": "model.safetensors",
172
+ "language_model.layers.25.mlp.gate_proj.weight": "model.safetensors",
173
+ "language_model.layers.25.mlp.up_proj.weight": "model.safetensors",
174
+ "language_model.layers.25.post_attention_layernorm.weight": "model.safetensors",
175
+ "language_model.layers.25.self_attn.k_proj.weight": "model.safetensors",
176
+ "language_model.layers.25.self_attn.o_proj.weight": "model.safetensors",
177
+ "language_model.layers.25.self_attn.q_proj.weight": "model.safetensors",
178
+ "language_model.layers.25.self_attn.v_proj.weight": "model.safetensors",
179
+ "language_model.layers.26.input_layernorm.weight": "model.safetensors",
180
+ "language_model.layers.26.mlp.down_proj.weight": "model.safetensors",
181
+ "language_model.layers.26.mlp.gate_proj.weight": "model.safetensors",
182
+ "language_model.layers.26.mlp.up_proj.weight": "model.safetensors",
183
+ "language_model.layers.26.post_attention_layernorm.weight": "model.safetensors",
184
+ "language_model.layers.26.self_attn.k_proj.weight": "model.safetensors",
185
+ "language_model.layers.26.self_attn.o_proj.weight": "model.safetensors",
186
+ "language_model.layers.26.self_attn.q_proj.weight": "model.safetensors",
187
+ "language_model.layers.26.self_attn.v_proj.weight": "model.safetensors",
188
+ "language_model.layers.27.input_layernorm.weight": "model.safetensors",
189
+ "language_model.layers.27.mlp.down_proj.weight": "model.safetensors",
190
+ "language_model.layers.27.mlp.gate_proj.weight": "model.safetensors",
191
+ "language_model.layers.27.mlp.up_proj.weight": "model.safetensors",
192
+ "language_model.layers.27.post_attention_layernorm.weight": "model.safetensors",
193
+ "language_model.layers.27.self_attn.k_proj.weight": "model.safetensors",
194
+ "language_model.layers.27.self_attn.o_proj.weight": "model.safetensors",
195
+ "language_model.layers.27.self_attn.q_proj.weight": "model.safetensors",
196
+ "language_model.layers.27.self_attn.v_proj.weight": "model.safetensors",
197
+ "language_model.layers.28.input_layernorm.weight": "model.safetensors",
198
+ "language_model.layers.28.mlp.down_proj.weight": "model.safetensors",
199
+ "language_model.layers.28.mlp.gate_proj.weight": "model.safetensors",
200
+ "language_model.layers.28.mlp.up_proj.weight": "model.safetensors",
201
+ "language_model.layers.28.post_attention_layernorm.weight": "model.safetensors",
202
+ "language_model.layers.28.self_attn.k_proj.weight": "model.safetensors",
203
+ "language_model.layers.28.self_attn.o_proj.weight": "model.safetensors",
204
+ "language_model.layers.28.self_attn.q_proj.weight": "model.safetensors",
205
+ "language_model.layers.28.self_attn.v_proj.weight": "model.safetensors",
206
+ "language_model.layers.29.input_layernorm.weight": "model.safetensors",
207
+ "language_model.layers.29.mlp.down_proj.weight": "model.safetensors",
208
+ "language_model.layers.29.mlp.gate_proj.weight": "model.safetensors",
209
+ "language_model.layers.29.mlp.up_proj.weight": "model.safetensors",
210
+ "language_model.layers.29.post_attention_layernorm.weight": "model.safetensors",
211
+ "language_model.layers.29.self_attn.k_proj.weight": "model.safetensors",
212
+ "language_model.layers.29.self_attn.o_proj.weight": "model.safetensors",
213
+ "language_model.layers.29.self_attn.q_proj.weight": "model.safetensors",
214
+ "language_model.layers.29.self_attn.v_proj.weight": "model.safetensors",
215
+ "language_model.layers.3.input_layernorm.weight": "model.safetensors",
216
+ "language_model.layers.3.mlp.down_proj.weight": "model.safetensors",
217
+ "language_model.layers.3.mlp.gate_proj.weight": "model.safetensors",
218
+ "language_model.layers.3.mlp.up_proj.weight": "model.safetensors",
219
+ "language_model.layers.3.post_attention_layernorm.weight": "model.safetensors",
220
+ "language_model.layers.3.self_attn.k_proj.weight": "model.safetensors",
221
+ "language_model.layers.3.self_attn.o_proj.weight": "model.safetensors",
222
+ "language_model.layers.3.self_attn.q_proj.weight": "model.safetensors",
223
+ "language_model.layers.3.self_attn.v_proj.weight": "model.safetensors",
224
+ "language_model.layers.4.input_layernorm.weight": "model.safetensors",
225
+ "language_model.layers.4.mlp.down_proj.weight": "model.safetensors",
226
+ "language_model.layers.4.mlp.gate_proj.weight": "model.safetensors",
227
+ "language_model.layers.4.mlp.up_proj.weight": "model.safetensors",
228
+ "language_model.layers.4.post_attention_layernorm.weight": "model.safetensors",
229
+ "language_model.layers.4.self_attn.k_proj.weight": "model.safetensors",
230
+ "language_model.layers.4.self_attn.o_proj.weight": "model.safetensors",
231
+ "language_model.layers.4.self_attn.q_proj.weight": "model.safetensors",
232
+ "language_model.layers.4.self_attn.v_proj.weight": "model.safetensors",
233
+ "language_model.layers.5.input_layernorm.weight": "model.safetensors",
234
+ "language_model.layers.5.mlp.down_proj.weight": "model.safetensors",
235
+ "language_model.layers.5.mlp.gate_proj.weight": "model.safetensors",
236
+ "language_model.layers.5.mlp.up_proj.weight": "model.safetensors",
237
+ "language_model.layers.5.post_attention_layernorm.weight": "model.safetensors",
238
+ "language_model.layers.5.self_attn.k_proj.weight": "model.safetensors",
239
+ "language_model.layers.5.self_attn.o_proj.weight": "model.safetensors",
240
+ "language_model.layers.5.self_attn.q_proj.weight": "model.safetensors",
241
+ "language_model.layers.5.self_attn.v_proj.weight": "model.safetensors",
242
+ "language_model.layers.6.input_layernorm.weight": "model.safetensors",
243
+ "language_model.layers.6.mlp.down_proj.weight": "model.safetensors",
244
+ "language_model.layers.6.mlp.gate_proj.weight": "model.safetensors",
245
+ "language_model.layers.6.mlp.up_proj.weight": "model.safetensors",
246
+ "language_model.layers.6.post_attention_layernorm.weight": "model.safetensors",
247
+ "language_model.layers.6.self_attn.k_proj.weight": "model.safetensors",
248
+ "language_model.layers.6.self_attn.o_proj.weight": "model.safetensors",
249
+ "language_model.layers.6.self_attn.q_proj.weight": "model.safetensors",
250
+ "language_model.layers.6.self_attn.v_proj.weight": "model.safetensors",
251
+ "language_model.layers.7.input_layernorm.weight": "model.safetensors",
252
+ "language_model.layers.7.mlp.down_proj.weight": "model.safetensors",
253
+ "language_model.layers.7.mlp.gate_proj.weight": "model.safetensors",
254
+ "language_model.layers.7.mlp.up_proj.weight": "model.safetensors",
255
+ "language_model.layers.7.post_attention_layernorm.weight": "model.safetensors",
256
+ "language_model.layers.7.self_attn.k_proj.weight": "model.safetensors",
257
+ "language_model.layers.7.self_attn.o_proj.weight": "model.safetensors",
258
+ "language_model.layers.7.self_attn.q_proj.weight": "model.safetensors",
259
+ "language_model.layers.7.self_attn.v_proj.weight": "model.safetensors",
260
+ "language_model.layers.8.input_layernorm.weight": "model.safetensors",
261
+ "language_model.layers.8.mlp.down_proj.weight": "model.safetensors",
262
+ "language_model.layers.8.mlp.gate_proj.weight": "model.safetensors",
263
+ "language_model.layers.8.mlp.up_proj.weight": "model.safetensors",
264
+ "language_model.layers.8.post_attention_layernorm.weight": "model.safetensors",
265
+ "language_model.layers.8.self_attn.k_proj.weight": "model.safetensors",
266
+ "language_model.layers.8.self_attn.o_proj.weight": "model.safetensors",
267
+ "language_model.layers.8.self_attn.q_proj.weight": "model.safetensors",
268
+ "language_model.layers.8.self_attn.v_proj.weight": "model.safetensors",
269
+ "language_model.layers.9.input_layernorm.weight": "model.safetensors",
270
+ "language_model.layers.9.mlp.down_proj.weight": "model.safetensors",
271
+ "language_model.layers.9.mlp.gate_proj.weight": "model.safetensors",
272
+ "language_model.layers.9.mlp.up_proj.weight": "model.safetensors",
273
+ "language_model.layers.9.post_attention_layernorm.weight": "model.safetensors",
274
+ "language_model.layers.9.self_attn.k_proj.weight": "model.safetensors",
275
+ "language_model.layers.9.self_attn.o_proj.weight": "model.safetensors",
276
+ "language_model.layers.9.self_attn.q_proj.weight": "model.safetensors",
277
+ "language_model.layers.9.self_attn.v_proj.weight": "model.safetensors",
278
+ "language_model.lm_head.weight": "model.safetensors",
279
+ "language_model.norm.weight": "model.safetensors",
280
+ "vision_model.embeddings.patch_embedding.bias": "model.safetensors",
281
+ "vision_model.embeddings.patch_embedding.weight": "model.safetensors",
282
+ "vision_model.embeddings.position_embedding.weight": "model.safetensors",
283
+ "vision_model.encoder.layers.0.layer_norm1.bias": "model.safetensors",
284
+ "vision_model.encoder.layers.0.layer_norm1.weight": "model.safetensors",
285
+ "vision_model.encoder.layers.0.layer_norm2.bias": "model.safetensors",
286
+ "vision_model.encoder.layers.0.layer_norm2.weight": "model.safetensors",
287
+ "vision_model.encoder.layers.0.mlp.fc1.bias": "model.safetensors",
288
+ "vision_model.encoder.layers.0.mlp.fc1.weight": "model.safetensors",
289
+ "vision_model.encoder.layers.0.mlp.fc2.bias": "model.safetensors",
290
+ "vision_model.encoder.layers.0.mlp.fc2.weight": "model.safetensors",
291
+ "vision_model.encoder.layers.0.self_attn.k_proj.bias": "model.safetensors",
292
+ "vision_model.encoder.layers.0.self_attn.k_proj.weight": "model.safetensors",
293
+ "vision_model.encoder.layers.0.self_attn.out_proj.bias": "model.safetensors",
294
+ "vision_model.encoder.layers.0.self_attn.out_proj.weight": "model.safetensors",
295
+ "vision_model.encoder.layers.0.self_attn.q_proj.bias": "model.safetensors",
296
+ "vision_model.encoder.layers.0.self_attn.q_proj.weight": "model.safetensors",
297
+ "vision_model.encoder.layers.0.self_attn.v_proj.bias": "model.safetensors",
298
+ "vision_model.encoder.layers.0.self_attn.v_proj.weight": "model.safetensors",
299
+ "vision_model.encoder.layers.1.layer_norm1.bias": "model.safetensors",
300
+ "vision_model.encoder.layers.1.layer_norm1.weight": "model.safetensors",
301
+ "vision_model.encoder.layers.1.layer_norm2.bias": "model.safetensors",
302
+ "vision_model.encoder.layers.1.layer_norm2.weight": "model.safetensors",
303
+ "vision_model.encoder.layers.1.mlp.fc1.bias": "model.safetensors",
304
+ "vision_model.encoder.layers.1.mlp.fc1.weight": "model.safetensors",
305
+ "vision_model.encoder.layers.1.mlp.fc2.bias": "model.safetensors",
306
+ "vision_model.encoder.layers.1.mlp.fc2.weight": "model.safetensors",
307
+ "vision_model.encoder.layers.1.self_attn.k_proj.bias": "model.safetensors",
308
+ "vision_model.encoder.layers.1.self_attn.k_proj.weight": "model.safetensors",
309
+ "vision_model.encoder.layers.1.self_attn.out_proj.bias": "model.safetensors",
310
+ "vision_model.encoder.layers.1.self_attn.out_proj.weight": "model.safetensors",
311
+ "vision_model.encoder.layers.1.self_attn.q_proj.bias": "model.safetensors",
312
+ "vision_model.encoder.layers.1.self_attn.q_proj.weight": "model.safetensors",
313
+ "vision_model.encoder.layers.1.self_attn.v_proj.bias": "model.safetensors",
314
+ "vision_model.encoder.layers.1.self_attn.v_proj.weight": "model.safetensors",
315
+ "vision_model.encoder.layers.10.layer_norm1.bias": "model.safetensors",
316
+ "vision_model.encoder.layers.10.layer_norm1.weight": "model.safetensors",
317
+ "vision_model.encoder.layers.10.layer_norm2.bias": "model.safetensors",
318
+ "vision_model.encoder.layers.10.layer_norm2.weight": "model.safetensors",
319
+ "vision_model.encoder.layers.10.mlp.fc1.bias": "model.safetensors",
320
+ "vision_model.encoder.layers.10.mlp.fc1.weight": "model.safetensors",
321
+ "vision_model.encoder.layers.10.mlp.fc2.bias": "model.safetensors",
322
+ "vision_model.encoder.layers.10.mlp.fc2.weight": "model.safetensors",
323
+ "vision_model.encoder.layers.10.self_attn.k_proj.bias": "model.safetensors",
324
+ "vision_model.encoder.layers.10.self_attn.k_proj.weight": "model.safetensors",
325
+ "vision_model.encoder.layers.10.self_attn.out_proj.bias": "model.safetensors",
326
+ "vision_model.encoder.layers.10.self_attn.out_proj.weight": "model.safetensors",
327
+ "vision_model.encoder.layers.10.self_attn.q_proj.bias": "model.safetensors",
328
+ "vision_model.encoder.layers.10.self_attn.q_proj.weight": "model.safetensors",
329
+ "vision_model.encoder.layers.10.self_attn.v_proj.bias": "model.safetensors",
330
+ "vision_model.encoder.layers.10.self_attn.v_proj.weight": "model.safetensors",
331
+ "vision_model.encoder.layers.11.layer_norm1.bias": "model.safetensors",
332
+ "vision_model.encoder.layers.11.layer_norm1.weight": "model.safetensors",
333
+ "vision_model.encoder.layers.11.layer_norm2.bias": "model.safetensors",
334
+ "vision_model.encoder.layers.11.layer_norm2.weight": "model.safetensors",
335
+ "vision_model.encoder.layers.11.mlp.fc1.bias": "model.safetensors",
336
+ "vision_model.encoder.layers.11.mlp.fc1.weight": "model.safetensors",
337
+ "vision_model.encoder.layers.11.mlp.fc2.bias": "model.safetensors",
338
+ "vision_model.encoder.layers.11.mlp.fc2.weight": "model.safetensors",
339
+ "vision_model.encoder.layers.11.self_attn.k_proj.bias": "model.safetensors",
340
+ "vision_model.encoder.layers.11.self_attn.k_proj.weight": "model.safetensors",
341
+ "vision_model.encoder.layers.11.self_attn.out_proj.bias": "model.safetensors",
342
+ "vision_model.encoder.layers.11.self_attn.out_proj.weight": "model.safetensors",
343
+ "vision_model.encoder.layers.11.self_attn.q_proj.bias": "model.safetensors",
344
+ "vision_model.encoder.layers.11.self_attn.q_proj.weight": "model.safetensors",
345
+ "vision_model.encoder.layers.11.self_attn.v_proj.bias": "model.safetensors",
346
+ "vision_model.encoder.layers.11.self_attn.v_proj.weight": "model.safetensors",
347
+ "vision_model.encoder.layers.2.layer_norm1.bias": "model.safetensors",
348
+ "vision_model.encoder.layers.2.layer_norm1.weight": "model.safetensors",
349
+ "vision_model.encoder.layers.2.layer_norm2.bias": "model.safetensors",
350
+ "vision_model.encoder.layers.2.layer_norm2.weight": "model.safetensors",
351
+ "vision_model.encoder.layers.2.mlp.fc1.bias": "model.safetensors",
352
+ "vision_model.encoder.layers.2.mlp.fc1.weight": "model.safetensors",
353
+ "vision_model.encoder.layers.2.mlp.fc2.bias": "model.safetensors",
354
+ "vision_model.encoder.layers.2.mlp.fc2.weight": "model.safetensors",
355
+ "vision_model.encoder.layers.2.self_attn.k_proj.bias": "model.safetensors",
356
+ "vision_model.encoder.layers.2.self_attn.k_proj.weight": "model.safetensors",
357
+ "vision_model.encoder.layers.2.self_attn.out_proj.bias": "model.safetensors",
358
+ "vision_model.encoder.layers.2.self_attn.out_proj.weight": "model.safetensors",
359
+ "vision_model.encoder.layers.2.self_attn.q_proj.bias": "model.safetensors",
360
+ "vision_model.encoder.layers.2.self_attn.q_proj.weight": "model.safetensors",
361
+ "vision_model.encoder.layers.2.self_attn.v_proj.bias": "model.safetensors",
362
+ "vision_model.encoder.layers.2.self_attn.v_proj.weight": "model.safetensors",
363
+ "vision_model.encoder.layers.3.layer_norm1.bias": "model.safetensors",
364
+ "vision_model.encoder.layers.3.layer_norm1.weight": "model.safetensors",
365
+ "vision_model.encoder.layers.3.layer_norm2.bias": "model.safetensors",
366
+ "vision_model.encoder.layers.3.layer_norm2.weight": "model.safetensors",
367
+ "vision_model.encoder.layers.3.mlp.fc1.bias": "model.safetensors",
368
+ "vision_model.encoder.layers.3.mlp.fc1.weight": "model.safetensors",
369
+ "vision_model.encoder.layers.3.mlp.fc2.bias": "model.safetensors",
370
+ "vision_model.encoder.layers.3.mlp.fc2.weight": "model.safetensors",
371
+ "vision_model.encoder.layers.3.self_attn.k_proj.bias": "model.safetensors",
372
+ "vision_model.encoder.layers.3.self_attn.k_proj.weight": "model.safetensors",
373
+ "vision_model.encoder.layers.3.self_attn.out_proj.bias": "model.safetensors",
374
+ "vision_model.encoder.layers.3.self_attn.out_proj.weight": "model.safetensors",
375
+ "vision_model.encoder.layers.3.self_attn.q_proj.bias": "model.safetensors",
376
+ "vision_model.encoder.layers.3.self_attn.q_proj.weight": "model.safetensors",
377
+ "vision_model.encoder.layers.3.self_attn.v_proj.bias": "model.safetensors",
378
+ "vision_model.encoder.layers.3.self_attn.v_proj.weight": "model.safetensors",
379
+ "vision_model.encoder.layers.4.layer_norm1.bias": "model.safetensors",
380
+ "vision_model.encoder.layers.4.layer_norm1.weight": "model.safetensors",
381
+ "vision_model.encoder.layers.4.layer_norm2.bias": "model.safetensors",
382
+ "vision_model.encoder.layers.4.layer_norm2.weight": "model.safetensors",
383
+ "vision_model.encoder.layers.4.mlp.fc1.bias": "model.safetensors",
384
+ "vision_model.encoder.layers.4.mlp.fc1.weight": "model.safetensors",
385
+ "vision_model.encoder.layers.4.mlp.fc2.bias": "model.safetensors",
386
+ "vision_model.encoder.layers.4.mlp.fc2.weight": "model.safetensors",
387
+ "vision_model.encoder.layers.4.self_attn.k_proj.bias": "model.safetensors",
388
+ "vision_model.encoder.layers.4.self_attn.k_proj.weight": "model.safetensors",
389
+ "vision_model.encoder.layers.4.self_attn.out_proj.bias": "model.safetensors",
390
+ "vision_model.encoder.layers.4.self_attn.out_proj.weight": "model.safetensors",
391
+ "vision_model.encoder.layers.4.self_attn.q_proj.bias": "model.safetensors",
392
+ "vision_model.encoder.layers.4.self_attn.q_proj.weight": "model.safetensors",
393
+ "vision_model.encoder.layers.4.self_attn.v_proj.bias": "model.safetensors",
394
+ "vision_model.encoder.layers.4.self_attn.v_proj.weight": "model.safetensors",
395
+ "vision_model.encoder.layers.5.layer_norm1.bias": "model.safetensors",
396
+ "vision_model.encoder.layers.5.layer_norm1.weight": "model.safetensors",
397
+ "vision_model.encoder.layers.5.layer_norm2.bias": "model.safetensors",
398
+ "vision_model.encoder.layers.5.layer_norm2.weight": "model.safetensors",
399
+ "vision_model.encoder.layers.5.mlp.fc1.bias": "model.safetensors",
400
+ "vision_model.encoder.layers.5.mlp.fc1.weight": "model.safetensors",
401
+ "vision_model.encoder.layers.5.mlp.fc2.bias": "model.safetensors",
402
+ "vision_model.encoder.layers.5.mlp.fc2.weight": "model.safetensors",
403
+ "vision_model.encoder.layers.5.self_attn.k_proj.bias": "model.safetensors",
404
+ "vision_model.encoder.layers.5.self_attn.k_proj.weight": "model.safetensors",
405
+ "vision_model.encoder.layers.5.self_attn.out_proj.bias": "model.safetensors",
406
+ "vision_model.encoder.layers.5.self_attn.out_proj.weight": "model.safetensors",
407
+ "vision_model.encoder.layers.5.self_attn.q_proj.bias": "model.safetensors",
408
+ "vision_model.encoder.layers.5.self_attn.q_proj.weight": "model.safetensors",
409
+ "vision_model.encoder.layers.5.self_attn.v_proj.bias": "model.safetensors",
410
+ "vision_model.encoder.layers.5.self_attn.v_proj.weight": "model.safetensors",
411
+ "vision_model.encoder.layers.6.layer_norm1.bias": "model.safetensors",
412
+ "vision_model.encoder.layers.6.layer_norm1.weight": "model.safetensors",
413
+ "vision_model.encoder.layers.6.layer_norm2.bias": "model.safetensors",
414
+ "vision_model.encoder.layers.6.layer_norm2.weight": "model.safetensors",
415
+ "vision_model.encoder.layers.6.mlp.fc1.bias": "model.safetensors",
416
+ "vision_model.encoder.layers.6.mlp.fc1.weight": "model.safetensors",
417
+ "vision_model.encoder.layers.6.mlp.fc2.bias": "model.safetensors",
418
+ "vision_model.encoder.layers.6.mlp.fc2.weight": "model.safetensors",
419
+ "vision_model.encoder.layers.6.self_attn.k_proj.bias": "model.safetensors",
420
+ "vision_model.encoder.layers.6.self_attn.k_proj.weight": "model.safetensors",
421
+ "vision_model.encoder.layers.6.self_attn.out_proj.bias": "model.safetensors",
422
+ "vision_model.encoder.layers.6.self_attn.out_proj.weight": "model.safetensors",
423
+ "vision_model.encoder.layers.6.self_attn.q_proj.bias": "model.safetensors",
424
+ "vision_model.encoder.layers.6.self_attn.q_proj.weight": "model.safetensors",
425
+ "vision_model.encoder.layers.6.self_attn.v_proj.bias": "model.safetensors",
426
+ "vision_model.encoder.layers.6.self_attn.v_proj.weight": "model.safetensors",
427
+ "vision_model.encoder.layers.7.layer_norm1.bias": "model.safetensors",
428
+ "vision_model.encoder.layers.7.layer_norm1.weight": "model.safetensors",
429
+ "vision_model.encoder.layers.7.layer_norm2.bias": "model.safetensors",
430
+ "vision_model.encoder.layers.7.layer_norm2.weight": "model.safetensors",
431
+ "vision_model.encoder.layers.7.mlp.fc1.bias": "model.safetensors",
432
+ "vision_model.encoder.layers.7.mlp.fc1.weight": "model.safetensors",
433
+ "vision_model.encoder.layers.7.mlp.fc2.bias": "model.safetensors",
434
+ "vision_model.encoder.layers.7.mlp.fc2.weight": "model.safetensors",
435
+ "vision_model.encoder.layers.7.self_attn.k_proj.bias": "model.safetensors",
436
+ "vision_model.encoder.layers.7.self_attn.k_proj.weight": "model.safetensors",
437
+ "vision_model.encoder.layers.7.self_attn.out_proj.bias": "model.safetensors",
438
+ "vision_model.encoder.layers.7.self_attn.out_proj.weight": "model.safetensors",
439
+ "vision_model.encoder.layers.7.self_attn.q_proj.bias": "model.safetensors",
440
+ "vision_model.encoder.layers.7.self_attn.q_proj.weight": "model.safetensors",
441
+ "vision_model.encoder.layers.7.self_attn.v_proj.bias": "model.safetensors",
442
+ "vision_model.encoder.layers.7.self_attn.v_proj.weight": "model.safetensors",
443
+ "vision_model.encoder.layers.8.layer_norm1.bias": "model.safetensors",
444
+ "vision_model.encoder.layers.8.layer_norm1.weight": "model.safetensors",
445
+ "vision_model.encoder.layers.8.layer_norm2.bias": "model.safetensors",
446
+ "vision_model.encoder.layers.8.layer_norm2.weight": "model.safetensors",
447
+ "vision_model.encoder.layers.8.mlp.fc1.bias": "model.safetensors",
448
+ "vision_model.encoder.layers.8.mlp.fc1.weight": "model.safetensors",
449
+ "vision_model.encoder.layers.8.mlp.fc2.bias": "model.safetensors",
450
+ "vision_model.encoder.layers.8.mlp.fc2.weight": "model.safetensors",
451
+ "vision_model.encoder.layers.8.self_attn.k_proj.bias": "model.safetensors",
452
+ "vision_model.encoder.layers.8.self_attn.k_proj.weight": "model.safetensors",
453
+ "vision_model.encoder.layers.8.self_attn.out_proj.bias": "model.safetensors",
454
+ "vision_model.encoder.layers.8.self_attn.out_proj.weight": "model.safetensors",
455
+ "vision_model.encoder.layers.8.self_attn.q_proj.bias": "model.safetensors",
456
+ "vision_model.encoder.layers.8.self_attn.q_proj.weight": "model.safetensors",
457
+ "vision_model.encoder.layers.8.self_attn.v_proj.bias": "model.safetensors",
458
+ "vision_model.encoder.layers.8.self_attn.v_proj.weight": "model.safetensors",
459
+ "vision_model.encoder.layers.9.layer_norm1.bias": "model.safetensors",
460
+ "vision_model.encoder.layers.9.layer_norm1.weight": "model.safetensors",
461
+ "vision_model.encoder.layers.9.layer_norm2.bias": "model.safetensors",
462
+ "vision_model.encoder.layers.9.layer_norm2.weight": "model.safetensors",
463
+ "vision_model.encoder.layers.9.mlp.fc1.bias": "model.safetensors",
464
+ "vision_model.encoder.layers.9.mlp.fc1.weight": "model.safetensors",
465
+ "vision_model.encoder.layers.9.mlp.fc2.bias": "model.safetensors",
466
+ "vision_model.encoder.layers.9.mlp.fc2.weight": "model.safetensors",
467
+ "vision_model.encoder.layers.9.self_attn.k_proj.bias": "model.safetensors",
468
+ "vision_model.encoder.layers.9.self_attn.k_proj.weight": "model.safetensors",
469
+ "vision_model.encoder.layers.9.self_attn.out_proj.bias": "model.safetensors",
470
+ "vision_model.encoder.layers.9.self_attn.out_proj.weight": "model.safetensors",
471
+ "vision_model.encoder.layers.9.self_attn.q_proj.bias": "model.safetensors",
472
+ "vision_model.encoder.layers.9.self_attn.q_proj.weight": "model.safetensors",
473
+ "vision_model.encoder.layers.9.self_attn.v_proj.bias": "model.safetensors",
474
+ "vision_model.encoder.layers.9.self_attn.v_proj.weight": "model.safetensors",
475
+ "vision_model.post_layernorm.bias": "model.safetensors",
476
+ "vision_model.post_layernorm.weight": "model.safetensors"
477
+ }
478
+ }
preprocessor_config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_convert_rgb": true,
3
+ "do_image_splitting": true,
4
+ "do_normalize": true,
5
+ "do_pad": true,
6
+ "do_rescale": true,
7
+ "do_resize": true,
8
+ "image_mean": [
9
+ 0.5,
10
+ 0.5,
11
+ 0.5
12
+ ],
13
+ "image_processor_type": "Idefics3ImageProcessor",
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "max_image_size": {
20
+ "longest_edge": 512
21
+ },
22
+ "processor_class": "Idefics3Processor",
23
+ "resample": 1,
24
+ "rescale_factor": 0.00392156862745098,
25
+ "size": {
26
+ "longest_edge": 2048
27
+ }
28
+ }
processor_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "image_seq_len": 64,
3
+ "processor_class": "Idefics3Processor"
4
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ {
4
+ "content": "<fake_token_around_image>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ {
11
+ "content": "<image>",
12
+ "lstrip": false,
13
+ "normalized": false,
14
+ "rstrip": false,
15
+ "single_word": false
16
+ },
17
+ {
18
+ "content": "<end_of_utterance>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ }
24
+ ],
25
+ "bos_token": {
26
+ "content": "<|im_start|>",
27
+ "lstrip": false,
28
+ "normalized": false,
29
+ "rstrip": false,
30
+ "single_word": false
31
+ },
32
+ "eos_token": {
33
+ "content": "<|im_end|>",
34
+ "lstrip": false,
35
+ "normalized": false,
36
+ "rstrip": false,
37
+ "single_word": false
38
+ },
39
+ "pad_token": {
40
+ "content": "<|im_end|>",
41
+ "lstrip": false,
42
+ "normalized": false,
43
+ "rstrip": false,
44
+ "single_word": false
45
+ },
46
+ "unk_token": {
47
+ "content": "<|endoftext|>",
48
+ "lstrip": false,
49
+ "normalized": false,
50
+ "rstrip": false,
51
+ "single_word": false
52
+ }
53
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,1189 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "0": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "1": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "2": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "3": {
29
+ "content": "<repo_name>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "4": {
37
+ "content": "<reponame>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "5": {
45
+ "content": "<file_sep>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "6": {
53
+ "content": "<filename>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "7": {
61
+ "content": "<gh_stars>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "8": {
69
+ "content": "<issue_start>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "9": {
77
+ "content": "<issue_comment>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "10": {
85
+ "content": "<issue_closed>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "11": {
93
+ "content": "<jupyter_start>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "12": {
101
+ "content": "<jupyter_text>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "13": {
109
+ "content": "<jupyter_code>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ },
116
+ "14": {
117
+ "content": "<jupyter_output>",
118
+ "lstrip": false,
119
+ "normalized": false,
120
+ "rstrip": false,
121
+ "single_word": false,
122
+ "special": true
123
+ },
124
+ "15": {
125
+ "content": "<jupyter_script>",
126
+ "lstrip": false,
127
+ "normalized": false,
128
+ "rstrip": false,
129
+ "single_word": false,
130
+ "special": true
131
+ },
132
+ "16": {
133
+ "content": "<empty_output>",
134
+ "lstrip": false,
135
+ "normalized": false,
136
+ "rstrip": false,
137
+ "single_word": false,
138
+ "special": true
139
+ },
140
+ "49152": {
141
+ "content": "<global-img>",
142
+ "lstrip": false,
143
+ "normalized": false,
144
+ "rstrip": false,
145
+ "single_word": false,
146
+ "special": true
147
+ },
148
+ "49153": {
149
+ "content": "<row_1_col_1>",
150
+ "lstrip": false,
151
+ "normalized": false,
152
+ "rstrip": false,
153
+ "single_word": false,
154
+ "special": true
155
+ },
156
+ "49154": {
157
+ "content": "<row_1_col_2>",
158
+ "lstrip": false,
159
+ "normalized": false,
160
+ "rstrip": false,
161
+ "single_word": false,
162
+ "special": true
163
+ },
164
+ "49155": {
165
+ "content": "<row_1_col_3>",
166
+ "lstrip": false,
167
+ "normalized": false,
168
+ "rstrip": false,
169
+ "single_word": false,
170
+ "special": true
171
+ },
172
+ "49156": {
173
+ "content": "<row_1_col_4>",
174
+ "lstrip": false,
175
+ "normalized": false,
176
+ "rstrip": false,
177
+ "single_word": false,
178
+ "special": true
179
+ },
180
+ "49157": {
181
+ "content": "<row_1_col_5>",
182
+ "lstrip": false,
183
+ "normalized": false,
184
+ "rstrip": false,
185
+ "single_word": false,
186
+ "special": true
187
+ },
188
+ "49158": {
189
+ "content": "<row_1_col_6>",
190
+ "lstrip": false,
191
+ "normalized": false,
192
+ "rstrip": false,
193
+ "single_word": false,
194
+ "special": true
195
+ },
196
+ "49159": {
197
+ "content": "<row_2_col_1>",
198
+ "lstrip": false,
199
+ "normalized": false,
200
+ "rstrip": false,
201
+ "single_word": false,
202
+ "special": true
203
+ },
204
+ "49160": {
205
+ "content": "<row_2_col_2>",
206
+ "lstrip": false,
207
+ "normalized": false,
208
+ "rstrip": false,
209
+ "single_word": false,
210
+ "special": true
211
+ },
212
+ "49161": {
213
+ "content": "<row_2_col_3>",
214
+ "lstrip": false,
215
+ "normalized": false,
216
+ "rstrip": false,
217
+ "single_word": false,
218
+ "special": true
219
+ },
220
+ "49162": {
221
+ "content": "<row_2_col_4>",
222
+ "lstrip": false,
223
+ "normalized": false,
224
+ "rstrip": false,
225
+ "single_word": false,
226
+ "special": true
227
+ },
228
+ "49163": {
229
+ "content": "<row_2_col_5>",
230
+ "lstrip": false,
231
+ "normalized": false,
232
+ "rstrip": false,
233
+ "single_word": false,
234
+ "special": true
235
+ },
236
+ "49164": {
237
+ "content": "<row_2_col_6>",
238
+ "lstrip": false,
239
+ "normalized": false,
240
+ "rstrip": false,
241
+ "single_word": false,
242
+ "special": true
243
+ },
244
+ "49165": {
245
+ "content": "<row_3_col_1>",
246
+ "lstrip": false,
247
+ "normalized": false,
248
+ "rstrip": false,
249
+ "single_word": false,
250
+ "special": true
251
+ },
252
+ "49166": {
253
+ "content": "<row_3_col_2>",
254
+ "lstrip": false,
255
+ "normalized": false,
256
+ "rstrip": false,
257
+ "single_word": false,
258
+ "special": true
259
+ },
260
+ "49167": {
261
+ "content": "<row_3_col_3>",
262
+ "lstrip": false,
263
+ "normalized": false,
264
+ "rstrip": false,
265
+ "single_word": false,
266
+ "special": true
267
+ },
268
+ "49168": {
269
+ "content": "<row_3_col_4>",
270
+ "lstrip": false,
271
+ "normalized": false,
272
+ "rstrip": false,
273
+ "single_word": false,
274
+ "special": true
275
+ },
276
+ "49169": {
277
+ "content": "<row_3_col_5>",
278
+ "lstrip": false,
279
+ "normalized": false,
280
+ "rstrip": false,
281
+ "single_word": false,
282
+ "special": true
283
+ },
284
+ "49170": {
285
+ "content": "<row_3_col_6>",
286
+ "lstrip": false,
287
+ "normalized": false,
288
+ "rstrip": false,
289
+ "single_word": false,
290
+ "special": true
291
+ },
292
+ "49171": {
293
+ "content": "<row_4_col_1>",
294
+ "lstrip": false,
295
+ "normalized": false,
296
+ "rstrip": false,
297
+ "single_word": false,
298
+ "special": true
299
+ },
300
+ "49172": {
301
+ "content": "<row_4_col_2>",
302
+ "lstrip": false,
303
+ "normalized": false,
304
+ "rstrip": false,
305
+ "single_word": false,
306
+ "special": true
307
+ },
308
+ "49173": {
309
+ "content": "<row_4_col_3>",
310
+ "lstrip": false,
311
+ "normalized": false,
312
+ "rstrip": false,
313
+ "single_word": false,
314
+ "special": true
315
+ },
316
+ "49174": {
317
+ "content": "<row_4_col_4>",
318
+ "lstrip": false,
319
+ "normalized": false,
320
+ "rstrip": false,
321
+ "single_word": false,
322
+ "special": true
323
+ },
324
+ "49175": {
325
+ "content": "<row_4_col_5>",
326
+ "lstrip": false,
327
+ "normalized": false,
328
+ "rstrip": false,
329
+ "single_word": false,
330
+ "special": true
331
+ },
332
+ "49176": {
333
+ "content": "<row_4_col_6>",
334
+ "lstrip": false,
335
+ "normalized": false,
336
+ "rstrip": false,
337
+ "single_word": false,
338
+ "special": true
339
+ },
340
+ "49177": {
341
+ "content": "<row_5_col_1>",
342
+ "lstrip": false,
343
+ "normalized": false,
344
+ "rstrip": false,
345
+ "single_word": false,
346
+ "special": true
347
+ },
348
+ "49178": {
349
+ "content": "<row_5_col_2>",
350
+ "lstrip": false,
351
+ "normalized": false,
352
+ "rstrip": false,
353
+ "single_word": false,
354
+ "special": true
355
+ },
356
+ "49179": {
357
+ "content": "<row_5_col_3>",
358
+ "lstrip": false,
359
+ "normalized": false,
360
+ "rstrip": false,
361
+ "single_word": false,
362
+ "special": true
363
+ },
364
+ "49180": {
365
+ "content": "<row_5_col_4>",
366
+ "lstrip": false,
367
+ "normalized": false,
368
+ "rstrip": false,
369
+ "single_word": false,
370
+ "special": true
371
+ },
372
+ "49181": {
373
+ "content": "<row_5_col_5>",
374
+ "lstrip": false,
375
+ "normalized": false,
376
+ "rstrip": false,
377
+ "single_word": false,
378
+ "special": true
379
+ },
380
+ "49182": {
381
+ "content": "<row_5_col_6>",
382
+ "lstrip": false,
383
+ "normalized": false,
384
+ "rstrip": false,
385
+ "single_word": false,
386
+ "special": true
387
+ },
388
+ "49183": {
389
+ "content": "<row_6_col_1>",
390
+ "lstrip": false,
391
+ "normalized": false,
392
+ "rstrip": false,
393
+ "single_word": false,
394
+ "special": true
395
+ },
396
+ "49184": {
397
+ "content": "<row_6_col_2>",
398
+ "lstrip": false,
399
+ "normalized": false,
400
+ "rstrip": false,
401
+ "single_word": false,
402
+ "special": true
403
+ },
404
+ "49185": {
405
+ "content": "<row_6_col_3>",
406
+ "lstrip": false,
407
+ "normalized": false,
408
+ "rstrip": false,
409
+ "single_word": false,
410
+ "special": true
411
+ },
412
+ "49186": {
413
+ "content": "<row_6_col_4>",
414
+ "lstrip": false,
415
+ "normalized": false,
416
+ "rstrip": false,
417
+ "single_word": false,
418
+ "special": true
419
+ },
420
+ "49187": {
421
+ "content": "<row_6_col_5>",
422
+ "lstrip": false,
423
+ "normalized": false,
424
+ "rstrip": false,
425
+ "single_word": false,
426
+ "special": true
427
+ },
428
+ "49188": {
429
+ "content": "<row_6_col_6>",
430
+ "lstrip": false,
431
+ "normalized": false,
432
+ "rstrip": false,
433
+ "single_word": false,
434
+ "special": true
435
+ },
436
+ "49189": {
437
+ "content": "<fake_token_around_image>",
438
+ "lstrip": false,
439
+ "normalized": false,
440
+ "rstrip": false,
441
+ "single_word": false,
442
+ "special": true
443
+ },
444
+ "49190": {
445
+ "content": "<image>",
446
+ "lstrip": false,
447
+ "normalized": false,
448
+ "rstrip": false,
449
+ "single_word": false,
450
+ "special": true
451
+ },
452
+ "49191": {
453
+ "content": "<caption>",
454
+ "lstrip": false,
455
+ "normalized": false,
456
+ "rstrip": false,
457
+ "single_word": false,
458
+ "special": true
459
+ },
460
+ "49192": {
461
+ "content": "</caption>",
462
+ "lstrip": false,
463
+ "normalized": false,
464
+ "rstrip": false,
465
+ "single_word": false,
466
+ "special": true
467
+ },
468
+ "49193": {
469
+ "content": "<footnote>",
470
+ "lstrip": false,
471
+ "normalized": false,
472
+ "rstrip": false,
473
+ "single_word": false,
474
+ "special": true
475
+ },
476
+ "49194": {
477
+ "content": "<|reserved_special_token_3|>",
478
+ "lstrip": false,
479
+ "normalized": false,
480
+ "rstrip": false,
481
+ "single_word": false,
482
+ "special": true
483
+ },
484
+ "49195": {
485
+ "content": "</footnote>",
486
+ "lstrip": false,
487
+ "normalized": false,
488
+ "rstrip": false,
489
+ "single_word": false,
490
+ "special": true
491
+ },
492
+ "49196": {
493
+ "content": "<formula>",
494
+ "lstrip": false,
495
+ "normalized": false,
496
+ "rstrip": false,
497
+ "single_word": false,
498
+ "special": true
499
+ },
500
+ "49197": {
501
+ "content": "</formula>",
502
+ "lstrip": false,
503
+ "normalized": false,
504
+ "rstrip": false,
505
+ "single_word": false,
506
+ "special": true
507
+ },
508
+ "49198": {
509
+ "content": "<list_item>",
510
+ "lstrip": false,
511
+ "normalized": false,
512
+ "rstrip": false,
513
+ "single_word": false,
514
+ "special": true
515
+ },
516
+ "49199": {
517
+ "content": "</list_item>",
518
+ "lstrip": false,
519
+ "normalized": false,
520
+ "rstrip": false,
521
+ "single_word": false,
522
+ "special": true
523
+ },
524
+ "49200": {
525
+ "content": "<page_footer>",
526
+ "lstrip": false,
527
+ "normalized": false,
528
+ "rstrip": false,
529
+ "single_word": false,
530
+ "special": true
531
+ },
532
+ "49201": {
533
+ "content": "</page_footer>",
534
+ "lstrip": false,
535
+ "normalized": false,
536
+ "rstrip": false,
537
+ "single_word": false,
538
+ "special": true
539
+ },
540
+ "49202": {
541
+ "content": "<page_header>",
542
+ "lstrip": false,
543
+ "normalized": false,
544
+ "rstrip": false,
545
+ "single_word": false,
546
+ "special": true
547
+ },
548
+ "49203": {
549
+ "content": "</page_header>",
550
+ "lstrip": false,
551
+ "normalized": false,
552
+ "rstrip": false,
553
+ "single_word": false,
554
+ "special": true
555
+ },
556
+ "49204": {
557
+ "content": "<picture>",
558
+ "lstrip": false,
559
+ "normalized": false,
560
+ "rstrip": false,
561
+ "single_word": false,
562
+ "special": true
563
+ },
564
+ "49205": {
565
+ "content": "</picture>",
566
+ "lstrip": false,
567
+ "normalized": false,
568
+ "rstrip": false,
569
+ "single_word": false,
570
+ "special": true
571
+ },
572
+ "49206": {
573
+ "content": "<section_header_level_",
574
+ "lstrip": false,
575
+ "normalized": false,
576
+ "rstrip": false,
577
+ "single_word": false,
578
+ "special": true
579
+ },
580
+ "49207": {
581
+ "content": "</section_header_level_",
582
+ "lstrip": false,
583
+ "normalized": false,
584
+ "rstrip": false,
585
+ "single_word": false,
586
+ "special": true
587
+ },
588
+ "49208": {
589
+ "content": "<otsl>",
590
+ "lstrip": false,
591
+ "normalized": false,
592
+ "rstrip": false,
593
+ "single_word": false,
594
+ "special": true
595
+ },
596
+ "49209": {
597
+ "content": "</otsl>",
598
+ "lstrip": false,
599
+ "normalized": false,
600
+ "rstrip": false,
601
+ "single_word": false,
602
+ "special": true
603
+ },
604
+ "49210": {
605
+ "content": "<checkbox_selected>",
606
+ "lstrip": false,
607
+ "normalized": false,
608
+ "rstrip": false,
609
+ "single_word": false,
610
+ "special": true
611
+ },
612
+ "49211": {
613
+ "content": "</checkbox_selected>",
614
+ "lstrip": false,
615
+ "normalized": false,
616
+ "rstrip": false,
617
+ "single_word": false,
618
+ "special": true
619
+ },
620
+ "49212": {
621
+ "content": "<checkbox_unselected>",
622
+ "lstrip": false,
623
+ "normalized": false,
624
+ "rstrip": false,
625
+ "single_word": false,
626
+ "special": true
627
+ },
628
+ "49213": {
629
+ "content": "</checkbox_unselected>",
630
+ "lstrip": false,
631
+ "normalized": false,
632
+ "rstrip": false,
633
+ "single_word": false,
634
+ "special": true
635
+ },
636
+ "49214": {
637
+ "content": "<form>",
638
+ "lstrip": false,
639
+ "normalized": false,
640
+ "rstrip": false,
641
+ "single_word": false,
642
+ "special": true
643
+ },
644
+ "49215": {
645
+ "content": "</form>",
646
+ "lstrip": false,
647
+ "normalized": false,
648
+ "rstrip": false,
649
+ "single_word": false,
650
+ "special": true
651
+ },
652
+ "49216": {
653
+ "content": "<key_value_region>",
654
+ "lstrip": false,
655
+ "normalized": false,
656
+ "rstrip": false,
657
+ "single_word": false,
658
+ "special": true
659
+ },
660
+ "49217": {
661
+ "content": "</key_value_region>",
662
+ "lstrip": false,
663
+ "normalized": false,
664
+ "rstrip": false,
665
+ "single_word": false,
666
+ "special": true
667
+ },
668
+ "49218": {
669
+ "content": "<loc_",
670
+ "lstrip": false,
671
+ "normalized": false,
672
+ "rstrip": false,
673
+ "single_word": false,
674
+ "special": true
675
+ },
676
+ "49219": {
677
+ "content": "<paragraph>",
678
+ "lstrip": false,
679
+ "normalized": false,
680
+ "rstrip": false,
681
+ "single_word": false,
682
+ "special": true
683
+ },
684
+ "49220": {
685
+ "content": "</paragraph>",
686
+ "lstrip": false,
687
+ "normalized": false,
688
+ "rstrip": false,
689
+ "single_word": false,
690
+ "special": true
691
+ },
692
+ "49221": {
693
+ "content": "<reference>",
694
+ "lstrip": false,
695
+ "normalized": false,
696
+ "rstrip": false,
697
+ "single_word": false,
698
+ "special": true
699
+ },
700
+ "49222": {
701
+ "content": "</reference>",
702
+ "lstrip": false,
703
+ "normalized": false,
704
+ "rstrip": false,
705
+ "single_word": false,
706
+ "special": true
707
+ },
708
+ "49223": {
709
+ "content": "<ordered_list>",
710
+ "lstrip": false,
711
+ "normalized": false,
712
+ "rstrip": false,
713
+ "single_word": false,
714
+ "special": true
715
+ },
716
+ "49224": {
717
+ "content": "</ordered_list>",
718
+ "lstrip": false,
719
+ "normalized": false,
720
+ "rstrip": false,
721
+ "single_word": false,
722
+ "special": true
723
+ },
724
+ "49225": {
725
+ "content": "<unordered_list>",
726
+ "lstrip": false,
727
+ "normalized": false,
728
+ "rstrip": false,
729
+ "single_word": false,
730
+ "special": true
731
+ },
732
+ "49226": {
733
+ "content": "</unordered_list>",
734
+ "lstrip": false,
735
+ "normalized": false,
736
+ "rstrip": false,
737
+ "single_word": false,
738
+ "special": true
739
+ },
740
+ "49227": {
741
+ "content": "<group>",
742
+ "lstrip": false,
743
+ "normalized": false,
744
+ "rstrip": false,
745
+ "single_word": false,
746
+ "special": true
747
+ },
748
+ "49228": {
749
+ "content": "</group>",
750
+ "lstrip": false,
751
+ "normalized": false,
752
+ "rstrip": false,
753
+ "single_word": false,
754
+ "special": true
755
+ },
756
+ "49229": {
757
+ "content": "<doctag>",
758
+ "lstrip": false,
759
+ "normalized": false,
760
+ "rstrip": false,
761
+ "single_word": false,
762
+ "special": true
763
+ },
764
+ "49230": {
765
+ "content": "</doctag>",
766
+ "lstrip": false,
767
+ "normalized": false,
768
+ "rstrip": false,
769
+ "single_word": false,
770
+ "special": true
771
+ },
772
+ "49231": {
773
+ "content": "<page_",
774
+ "lstrip": false,
775
+ "normalized": false,
776
+ "rstrip": false,
777
+ "single_word": false,
778
+ "special": true
779
+ },
780
+ "49232": {
781
+ "content": "<text_break>",
782
+ "lstrip": false,
783
+ "normalized": false,
784
+ "rstrip": false,
785
+ "single_word": false,
786
+ "special": true
787
+ },
788
+ "49233": {
789
+ "content": "<fcel>",
790
+ "lstrip": false,
791
+ "normalized": false,
792
+ "rstrip": false,
793
+ "single_word": false,
794
+ "special": true
795
+ },
796
+ "49234": {
797
+ "content": "<ecel>",
798
+ "lstrip": false,
799
+ "normalized": false,
800
+ "rstrip": false,
801
+ "single_word": false,
802
+ "special": true
803
+ },
804
+ "49235": {
805
+ "content": "<lcel>",
806
+ "lstrip": false,
807
+ "normalized": false,
808
+ "rstrip": false,
809
+ "single_word": false,
810
+ "special": true
811
+ },
812
+ "49236": {
813
+ "content": "<ucel>",
814
+ "lstrip": false,
815
+ "normalized": false,
816
+ "rstrip": false,
817
+ "single_word": false,
818
+ "special": true
819
+ },
820
+ "49237": {
821
+ "content": "<xcel>",
822
+ "lstrip": false,
823
+ "normalized": false,
824
+ "rstrip": false,
825
+ "single_word": false,
826
+ "special": true
827
+ },
828
+ "49238": {
829
+ "content": "<nl>",
830
+ "lstrip": false,
831
+ "normalized": false,
832
+ "rstrip": false,
833
+ "single_word": false,
834
+ "special": true
835
+ },
836
+ "49239": {
837
+ "content": "<ched>",
838
+ "lstrip": false,
839
+ "normalized": false,
840
+ "rstrip": false,
841
+ "single_word": false,
842
+ "special": true
843
+ },
844
+ "49240": {
845
+ "content": "<rhed>",
846
+ "lstrip": false,
847
+ "normalized": false,
848
+ "rstrip": false,
849
+ "single_word": false,
850
+ "special": true
851
+ },
852
+ "49241": {
853
+ "content": "<|reserved_special_token_50|>",
854
+ "lstrip": false,
855
+ "normalized": false,
856
+ "rstrip": false,
857
+ "single_word": false,
858
+ "special": true
859
+ },
860
+ "49242": {
861
+ "content": "<key_",
862
+ "lstrip": false,
863
+ "normalized": false,
864
+ "rstrip": false,
865
+ "single_word": false,
866
+ "special": true
867
+ },
868
+ "49243": {
869
+ "content": "</key_",
870
+ "lstrip": false,
871
+ "normalized": false,
872
+ "rstrip": false,
873
+ "single_word": false,
874
+ "special": true
875
+ },
876
+ "49244": {
877
+ "content": "<value_",
878
+ "lstrip": false,
879
+ "normalized": false,
880
+ "rstrip": false,
881
+ "single_word": false,
882
+ "special": true
883
+ },
884
+ "49245": {
885
+ "content": "</value_",
886
+ "lstrip": false,
887
+ "normalized": false,
888
+ "rstrip": false,
889
+ "single_word": false,
890
+ "special": true
891
+ },
892
+ "49246": {
893
+ "content": "<link_",
894
+ "lstrip": false,
895
+ "normalized": false,
896
+ "rstrip": false,
897
+ "single_word": false,
898
+ "special": true
899
+ },
900
+ "49247": {
901
+ "content": "<chart>",
902
+ "lstrip": false,
903
+ "normalized": false,
904
+ "rstrip": false,
905
+ "single_word": false,
906
+ "special": true
907
+ },
908
+ "49248": {
909
+ "content": "</chart>",
910
+ "lstrip": false,
911
+ "normalized": false,
912
+ "rstrip": false,
913
+ "single_word": false,
914
+ "special": true
915
+ },
916
+ "49249": {
917
+ "content": "<page_break>",
918
+ "lstrip": false,
919
+ "normalized": false,
920
+ "rstrip": false,
921
+ "single_word": false,
922
+ "special": true
923
+ },
924
+ "49250": {
925
+ "content": "<smiles>",
926
+ "lstrip": false,
927
+ "normalized": false,
928
+ "rstrip": false,
929
+ "single_word": false,
930
+ "special": true
931
+ },
932
+ "49251": {
933
+ "content": "</smiles>",
934
+ "lstrip": false,
935
+ "normalized": false,
936
+ "rstrip": false,
937
+ "single_word": false,
938
+ "special": true
939
+ },
940
+ "49252": {
941
+ "content": "<|reserved_special_token_61|>",
942
+ "lstrip": false,
943
+ "normalized": false,
944
+ "rstrip": false,
945
+ "single_word": false,
946
+ "special": true
947
+ },
948
+ "49253": {
949
+ "content": "<|reserved_special_token_62|>",
950
+ "lstrip": false,
951
+ "normalized": false,
952
+ "rstrip": false,
953
+ "single_word": false,
954
+ "special": true
955
+ },
956
+ "49254": {
957
+ "content": "<|reserved_special_token_63|>",
958
+ "lstrip": false,
959
+ "normalized": false,
960
+ "rstrip": false,
961
+ "single_word": false,
962
+ "special": true
963
+ },
964
+ "49255": {
965
+ "content": "<|reserved_special_token_64|>",
966
+ "lstrip": false,
967
+ "normalized": false,
968
+ "rstrip": false,
969
+ "single_word": false,
970
+ "special": true
971
+ },
972
+ "49256": {
973
+ "content": "<|reserved_special_token_65|>",
974
+ "lstrip": false,
975
+ "normalized": false,
976
+ "rstrip": false,
977
+ "single_word": false,
978
+ "special": true
979
+ },
980
+ "49257": {
981
+ "content": "<|reserved_special_token_66|>",
982
+ "lstrip": false,
983
+ "normalized": false,
984
+ "rstrip": false,
985
+ "single_word": false,
986
+ "special": true
987
+ },
988
+ "49258": {
989
+ "content": "<|reserved_special_token_67|>",
990
+ "lstrip": false,
991
+ "normalized": false,
992
+ "rstrip": false,
993
+ "single_word": false,
994
+ "special": true
995
+ },
996
+ "49259": {
997
+ "content": "<|reserved_special_token_68|>",
998
+ "lstrip": false,
999
+ "normalized": false,
1000
+ "rstrip": false,
1001
+ "single_word": false,
1002
+ "special": true
1003
+ },
1004
+ "49260": {
1005
+ "content": "<|reserved_special_token_69|>",
1006
+ "lstrip": false,
1007
+ "normalized": false,
1008
+ "rstrip": false,
1009
+ "single_word": false,
1010
+ "special": true
1011
+ },
1012
+ "49261": {
1013
+ "content": "<|reserved_special_token_70|>",
1014
+ "lstrip": false,
1015
+ "normalized": false,
1016
+ "rstrip": false,
1017
+ "single_word": false,
1018
+ "special": true
1019
+ },
1020
+ "49262": {
1021
+ "content": "<|reserved_special_token_71|>",
1022
+ "lstrip": false,
1023
+ "normalized": false,
1024
+ "rstrip": false,
1025
+ "single_word": false,
1026
+ "special": true
1027
+ },
1028
+ "49263": {
1029
+ "content": "<|reserved_special_token_72|>",
1030
+ "lstrip": false,
1031
+ "normalized": false,
1032
+ "rstrip": false,
1033
+ "single_word": false,
1034
+ "special": true
1035
+ },
1036
+ "49264": {
1037
+ "content": "<|reserved_special_token_73|>",
1038
+ "lstrip": false,
1039
+ "normalized": false,
1040
+ "rstrip": false,
1041
+ "single_word": false,
1042
+ "special": true
1043
+ },
1044
+ "49265": {
1045
+ "content": "<|reserved_special_token_74|>",
1046
+ "lstrip": false,
1047
+ "normalized": false,
1048
+ "rstrip": false,
1049
+ "single_word": false,
1050
+ "special": true
1051
+ },
1052
+ "49266": {
1053
+ "content": "<|reserved_special_token_75|>",
1054
+ "lstrip": false,
1055
+ "normalized": false,
1056
+ "rstrip": false,
1057
+ "single_word": false,
1058
+ "special": true
1059
+ },
1060
+ "49267": {
1061
+ "content": "<|reserved_special_token_76|>",
1062
+ "lstrip": false,
1063
+ "normalized": false,
1064
+ "rstrip": false,
1065
+ "single_word": false,
1066
+ "special": true
1067
+ },
1068
+ "49268": {
1069
+ "content": "<|reserved_special_token_77|>",
1070
+ "lstrip": false,
1071
+ "normalized": false,
1072
+ "rstrip": false,
1073
+ "single_word": false,
1074
+ "special": true
1075
+ },
1076
+ "49269": {
1077
+ "content": "<|reserved_special_token_78|>",
1078
+ "lstrip": false,
1079
+ "normalized": false,
1080
+ "rstrip": false,
1081
+ "single_word": false,
1082
+ "special": true
1083
+ },
1084
+ "49270": {
1085
+ "content": "<|reserved_special_token_79|>",
1086
+ "lstrip": false,
1087
+ "normalized": false,
1088
+ "rstrip": false,
1089
+ "single_word": false,
1090
+ "special": true
1091
+ },
1092
+ "49271": {
1093
+ "content": "<|reserved_special_token_80|>",
1094
+ "lstrip": false,
1095
+ "normalized": false,
1096
+ "rstrip": false,
1097
+ "single_word": false,
1098
+ "special": true
1099
+ },
1100
+ "49272": {
1101
+ "content": "<|reserved_special_token_81|>",
1102
+ "lstrip": false,
1103
+ "normalized": false,
1104
+ "rstrip": false,
1105
+ "single_word": false,
1106
+ "special": true
1107
+ },
1108
+ "49273": {
1109
+ "content": "<|reserved_special_token_82|>",
1110
+ "lstrip": false,
1111
+ "normalized": false,
1112
+ "rstrip": false,
1113
+ "single_word": false,
1114
+ "special": true
1115
+ },
1116
+ "49274": {
1117
+ "content": "<|reserved_special_token_83|>",
1118
+ "lstrip": false,
1119
+ "normalized": false,
1120
+ "rstrip": false,
1121
+ "single_word": false,
1122
+ "special": true
1123
+ },
1124
+ "49275": {
1125
+ "content": "<|reserved_special_token_84|>",
1126
+ "lstrip": false,
1127
+ "normalized": false,
1128
+ "rstrip": false,
1129
+ "single_word": false,
1130
+ "special": true
1131
+ },
1132
+ "49276": {
1133
+ "content": "<|reserved_special_token_85|>",
1134
+ "lstrip": false,
1135
+ "normalized": false,
1136
+ "rstrip": false,
1137
+ "single_word": false,
1138
+ "special": true
1139
+ },
1140
+ "49277": {
1141
+ "content": "<|reserved_special_token_86|>",
1142
+ "lstrip": false,
1143
+ "normalized": false,
1144
+ "rstrip": false,
1145
+ "single_word": false,
1146
+ "special": true
1147
+ },
1148
+ "49278": {
1149
+ "content": "<|reserved_special_token_87|>",
1150
+ "lstrip": false,
1151
+ "normalized": false,
1152
+ "rstrip": false,
1153
+ "single_word": false,
1154
+ "special": true
1155
+ },
1156
+ "49279": {
1157
+ "content": "<end_of_utterance>",
1158
+ "lstrip": false,
1159
+ "normalized": false,
1160
+ "rstrip": false,
1161
+ "single_word": false,
1162
+ "special": true
1163
+ }
1164
+ },
1165
+ "additional_special_tokens": [
1166
+ "<fake_token_around_image>",
1167
+ "<image>",
1168
+ "<end_of_utterance>"
1169
+ ],
1170
+ "bos_token": "<|im_start|>",
1171
+ "chat_template": "<|im_start|>{% for message in messages %}{{message['role'] | capitalize}}{% if message['content'][0]['type'] == 'image' %}{{':'}}{% else %}{{': '}}{% endif %}{% for line in message['content'] %}{% if line['type'] == 'text' %}{{line['text']}}{% elif line['type'] == 'image' %}{{ '<image>' }}{% endif %}{% endfor %}<end_of_utterance>\n{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}",
1172
+ "clean_up_tokenization_spaces": false,
1173
+ "eos_token": "<|im_end|>",
1174
+ "extra_special_tokens": {},
1175
+ "legacy": false,
1176
+ "max_length": 8192,
1177
+ "model_max_length": 8192,
1178
+ "pad_to_multiple_of": null,
1179
+ "pad_token": "<|im_end|>",
1180
+ "pad_token_type_id": 0,
1181
+ "padding_side": "right",
1182
+ "processor_class": "Idefics3Processor",
1183
+ "stride": 0,
1184
+ "tokenizer_class": "GPT2Tokenizer",
1185
+ "truncation_side": "right",
1186
+ "truncation_strategy": "longest_first",
1187
+ "unk_token": "<|endoftext|>",
1188
+ "vocab_size": 49152
1189
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info(f"Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info(f"Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)