Hugging Face's logo

NeoChen1024
/
llama-joycaption-beta-one-hf-llava-FP8-Dynamic

Safetensors
llava
captioning
compressed-tensors
Model card Files Community
NeoChen1024 commited on
Commit
ed68f46
·
verified ·
1 Parent(s): 8f51a10

Upload folder using huggingface_hub

Browse files
Files changed (9) hide show
  1. README.md +3 -2
  2. chat_template.jinja +43 -0
  3. config.json +12 -173
  4. generation_config.json +4 -1
  5. model-00001-of-00002.safetensors +3 -0
  6. model-00002-of-00002.safetensors +3 -0
  7. model.safetensors.index.json +0 -0
  8. recipe.yaml +1 -1
  9. tokenizer_config.json +1 -2
README.md CHANGED
@@ -1,10 +1,11 @@
1
  ---
2
  base_model:
3
- - fancyfeast/llama-joycaption-beta-one-hf-llava
 
4
  tags:
5
  - captioning
6
  ---
7
- # Model Card for Llama JoyCaption Beta One (FP8 Dynamic quantization)
8
 
9
  [Github](https://github.com/fpgaminer/joycaption)
10
 
 
1
  ---
2
  base_model:
3
+ - meta-llama/Llama-3.1-8B-Instruct
4
+ - google/siglip2-so400m-patch14-384
5
  tags:
6
  - captioning
7
  ---
8
+ # Model Card for Llama JoyCaption Beta One
9
 
10
  [Github](https://github.com/fpgaminer/joycaption)
11
 
chat_template.jinja ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if not date_string is defined %}
2
+ {%- set date_string = "26 July 2024" %}
3
+ {%- endif %}
4
+
5
+ {#- This block extracts the system message, so we can slot it into the right place. #}
6
+ {%- if messages[0]['role'] == 'system' %}
7
+ {%- set system_message = messages[0]['content'] %}
8
+ {%- set messages = messages[1:] %}
9
+ {%- else %}
10
+ {%- set system_message = "" %}
11
+ {%- endif %}
12
+
13
+ {#- System message + builtin tools #}
14
+ {{- "<|start_header_id|>system<|end_header_id|>
15
+
16
+ " }}
17
+ {{- "Cutting Knowledge Date: December 2023
18
+ " }}
19
+ {{- "Today Date: " + date_string + "
20
+
21
+ " }}
22
+ {{- system_message }}
23
+ {{- "<|eot_id|>" }}
24
+
25
+
26
+ {%- set first_user_message = True %}
27
+ {%- for message in messages %}
28
+ {%- if first_user_message and message['role'] == 'user' %}
29
+ {%- set first_user_message = False %}
30
+ {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>
31
+
32
+ <|reserved_special_token_70|><|reserved_special_token_69|><|reserved_special_token_71|>'+ message['content'].replace('<|reserved_special_token_69|>', '').lstrip() + '<|eot_id|>' }}
33
+ {%- else %}
34
+ {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>
35
+
36
+ '+ message['content'] + '<|eot_id|>' }}
37
+ {%- endif %}
38
+ {%- endfor %}
39
+ {%- if add_generation_prompt %}
40
+ {{- '<|start_header_id|>assistant<|end_header_id|>
41
+
42
+ ' }}
43
+ {%- endif %}
config.json CHANGED
@@ -2,6 +2,7 @@
2
  "architectures": [
3
  "LlavaForConditionalGeneration"
4
  ],
 
5
  "image_seq_length": 729,
6
  "image_token_index": 128077,
7
  "model_type": "llava",
@@ -10,6 +11,7 @@
10
  "quantization_config": {
11
  "config_groups": {
12
  "group_0": {
 
13
  "input_activations": {
14
  "actorder": null,
15
  "block_structure": null,
@@ -43,177 +45,14 @@
43
  "format": "float-quantized",
44
  "global_compression_ratio": null,
45
  "ignore": [
46
- "vision_tower.vision_model.encoder.layers.0.self_attn.k_proj",
47
- "vision_tower.vision_model.encoder.layers.0.self_attn.v_proj",
48
- "vision_tower.vision_model.encoder.layers.0.self_attn.q_proj",
49
- "vision_tower.vision_model.encoder.layers.0.self_attn.out_proj",
50
- "vision_tower.vision_model.encoder.layers.0.mlp.fc1",
51
- "vision_tower.vision_model.encoder.layers.0.mlp.fc2",
52
- "vision_tower.vision_model.encoder.layers.1.self_attn.k_proj",
53
- "vision_tower.vision_model.encoder.layers.1.self_attn.v_proj",
54
- "vision_tower.vision_model.encoder.layers.1.self_attn.q_proj",
55
- "vision_tower.vision_model.encoder.layers.1.self_attn.out_proj",
56
- "vision_tower.vision_model.encoder.layers.1.mlp.fc1",
57
- "vision_tower.vision_model.encoder.layers.1.mlp.fc2",
58
- "vision_tower.vision_model.encoder.layers.2.self_attn.k_proj",
59
- "vision_tower.vision_model.encoder.layers.2.self_attn.v_proj",
60
- "vision_tower.vision_model.encoder.layers.2.self_attn.q_proj",
61
- "vision_tower.vision_model.encoder.layers.2.self_attn.out_proj",
62
- "vision_tower.vision_model.encoder.layers.2.mlp.fc1",
63
- "vision_tower.vision_model.encoder.layers.2.mlp.fc2",
64
- "vision_tower.vision_model.encoder.layers.3.self_attn.k_proj",
65
- "vision_tower.vision_model.encoder.layers.3.self_attn.v_proj",
66
- "vision_tower.vision_model.encoder.layers.3.self_attn.q_proj",
67
- "vision_tower.vision_model.encoder.layers.3.self_attn.out_proj",
68
- "vision_tower.vision_model.encoder.layers.3.mlp.fc1",
69
- "vision_tower.vision_model.encoder.layers.3.mlp.fc2",
70
- "vision_tower.vision_model.encoder.layers.4.self_attn.k_proj",
71
- "vision_tower.vision_model.encoder.layers.4.self_attn.v_proj",
72
- "vision_tower.vision_model.encoder.layers.4.self_attn.q_proj",
73
- "vision_tower.vision_model.encoder.layers.4.self_attn.out_proj",
74
- "vision_tower.vision_model.encoder.layers.4.mlp.fc1",
75
- "vision_tower.vision_model.encoder.layers.4.mlp.fc2",
76
- "vision_tower.vision_model.encoder.layers.5.self_attn.k_proj",
77
- "vision_tower.vision_model.encoder.layers.5.self_attn.v_proj",
78
- "vision_tower.vision_model.encoder.layers.5.self_attn.q_proj",
79
- "vision_tower.vision_model.encoder.layers.5.self_attn.out_proj",
80
- "vision_tower.vision_model.encoder.layers.5.mlp.fc1",
81
- "vision_tower.vision_model.encoder.layers.5.mlp.fc2",
82
- "vision_tower.vision_model.encoder.layers.6.self_attn.k_proj",
83
- "vision_tower.vision_model.encoder.layers.6.self_attn.v_proj",
84
- "vision_tower.vision_model.encoder.layers.6.self_attn.q_proj",
85
- "vision_tower.vision_model.encoder.layers.6.self_attn.out_proj",
86
- "vision_tower.vision_model.encoder.layers.6.mlp.fc1",
87
- "vision_tower.vision_model.encoder.layers.6.mlp.fc2",
88
- "vision_tower.vision_model.encoder.layers.7.self_attn.k_proj",
89
- "vision_tower.vision_model.encoder.layers.7.self_attn.v_proj",
90
- "vision_tower.vision_model.encoder.layers.7.self_attn.q_proj",
91
- "vision_tower.vision_model.encoder.layers.7.self_attn.out_proj",
92
- "vision_tower.vision_model.encoder.layers.7.mlp.fc1",
93
- "vision_tower.vision_model.encoder.layers.7.mlp.fc2",
94
- "vision_tower.vision_model.encoder.layers.8.self_attn.k_proj",
95
- "vision_tower.vision_model.encoder.layers.8.self_attn.v_proj",
96
- "vision_tower.vision_model.encoder.layers.8.self_attn.q_proj",
97
- "vision_tower.vision_model.encoder.layers.8.self_attn.out_proj",
98
- "vision_tower.vision_model.encoder.layers.8.mlp.fc1",
99
- "vision_tower.vision_model.encoder.layers.8.mlp.fc2",
100
- "vision_tower.vision_model.encoder.layers.9.self_attn.k_proj",
101
- "vision_tower.vision_model.encoder.layers.9.self_attn.v_proj",
102
- "vision_tower.vision_model.encoder.layers.9.self_attn.q_proj",
103
- "vision_tower.vision_model.encoder.layers.9.self_attn.out_proj",
104
- "vision_tower.vision_model.encoder.layers.9.mlp.fc1",
105
- "vision_tower.vision_model.encoder.layers.9.mlp.fc2",
106
- "vision_tower.vision_model.encoder.layers.10.self_attn.k_proj",
107
- "vision_tower.vision_model.encoder.layers.10.self_attn.v_proj",
108
- "vision_tower.vision_model.encoder.layers.10.self_attn.q_proj",
109
- "vision_tower.vision_model.encoder.layers.10.self_attn.out_proj",
110
- "vision_tower.vision_model.encoder.layers.10.mlp.fc1",
111
- "vision_tower.vision_model.encoder.layers.10.mlp.fc2",
112
- "vision_tower.vision_model.encoder.layers.11.self_attn.k_proj",
113
- "vision_tower.vision_model.encoder.layers.11.self_attn.v_proj",
114
- "vision_tower.vision_model.encoder.layers.11.self_attn.q_proj",
115
- "vision_tower.vision_model.encoder.layers.11.self_attn.out_proj",
116
- "vision_tower.vision_model.encoder.layers.11.mlp.fc1",
117
- "vision_tower.vision_model.encoder.layers.11.mlp.fc2",
118
- "vision_tower.vision_model.encoder.layers.12.self_attn.k_proj",
119
- "vision_tower.vision_model.encoder.layers.12.self_attn.v_proj",
120
- "vision_tower.vision_model.encoder.layers.12.self_attn.q_proj",
121
- "vision_tower.vision_model.encoder.layers.12.self_attn.out_proj",
122
- "vision_tower.vision_model.encoder.layers.12.mlp.fc1",
123
- "vision_tower.vision_model.encoder.layers.12.mlp.fc2",
124
- "vision_tower.vision_model.encoder.layers.13.self_attn.k_proj",
125
- "vision_tower.vision_model.encoder.layers.13.self_attn.v_proj",
126
- "vision_tower.vision_model.encoder.layers.13.self_attn.q_proj",
127
- "vision_tower.vision_model.encoder.layers.13.self_attn.out_proj",
128
- "vision_tower.vision_model.encoder.layers.13.mlp.fc1",
129
- "vision_tower.vision_model.encoder.layers.13.mlp.fc2",
130
- "vision_tower.vision_model.encoder.layers.14.self_attn.k_proj",
131
- "vision_tower.vision_model.encoder.layers.14.self_attn.v_proj",
132
- "vision_tower.vision_model.encoder.layers.14.self_attn.q_proj",
133
- "vision_tower.vision_model.encoder.layers.14.self_attn.out_proj",
134
- "vision_tower.vision_model.encoder.layers.14.mlp.fc1",
135
- "vision_tower.vision_model.encoder.layers.14.mlp.fc2",
136
- "vision_tower.vision_model.encoder.layers.15.self_attn.k_proj",
137
- "vision_tower.vision_model.encoder.layers.15.self_attn.v_proj",
138
- "vision_tower.vision_model.encoder.layers.15.self_attn.q_proj",
139
- "vision_tower.vision_model.encoder.layers.15.self_attn.out_proj",
140
- "vision_tower.vision_model.encoder.layers.15.mlp.fc1",
141
- "vision_tower.vision_model.encoder.layers.15.mlp.fc2",
142
- "vision_tower.vision_model.encoder.layers.16.self_attn.k_proj",
143
- "vision_tower.vision_model.encoder.layers.16.self_attn.v_proj",
144
- "vision_tower.vision_model.encoder.layers.16.self_attn.q_proj",
145
- "vision_tower.vision_model.encoder.layers.16.self_attn.out_proj",
146
- "vision_tower.vision_model.encoder.layers.16.mlp.fc1",
147
- "vision_tower.vision_model.encoder.layers.16.mlp.fc2",
148
- "vision_tower.vision_model.encoder.layers.17.self_attn.k_proj",
149
- "vision_tower.vision_model.encoder.layers.17.self_attn.v_proj",
150
- "vision_tower.vision_model.encoder.layers.17.self_attn.q_proj",
151
- "vision_tower.vision_model.encoder.layers.17.self_attn.out_proj",
152
- "vision_tower.vision_model.encoder.layers.17.mlp.fc1",
153
- "vision_tower.vision_model.encoder.layers.17.mlp.fc2",
154
- "vision_tower.vision_model.encoder.layers.18.self_attn.k_proj",
155
- "vision_tower.vision_model.encoder.layers.18.self_attn.v_proj",
156
- "vision_tower.vision_model.encoder.layers.18.self_attn.q_proj",
157
- "vision_tower.vision_model.encoder.layers.18.self_attn.out_proj",
158
- "vision_tower.vision_model.encoder.layers.18.mlp.fc1",
159
- "vision_tower.vision_model.encoder.layers.18.mlp.fc2",
160
- "vision_tower.vision_model.encoder.layers.19.self_attn.k_proj",
161
- "vision_tower.vision_model.encoder.layers.19.self_attn.v_proj",
162
- "vision_tower.vision_model.encoder.layers.19.self_attn.q_proj",
163
- "vision_tower.vision_model.encoder.layers.19.self_attn.out_proj",
164
- "vision_tower.vision_model.encoder.layers.19.mlp.fc1",
165
- "vision_tower.vision_model.encoder.layers.19.mlp.fc2",
166
- "vision_tower.vision_model.encoder.layers.20.self_attn.k_proj",
167
- "vision_tower.vision_model.encoder.layers.20.self_attn.v_proj",
168
- "vision_tower.vision_model.encoder.layers.20.self_attn.q_proj",
169
- "vision_tower.vision_model.encoder.layers.20.self_attn.out_proj",
170
- "vision_tower.vision_model.encoder.layers.20.mlp.fc1",
171
- "vision_tower.vision_model.encoder.layers.20.mlp.fc2",
172
- "vision_tower.vision_model.encoder.layers.21.self_attn.k_proj",
173
- "vision_tower.vision_model.encoder.layers.21.self_attn.v_proj",
174
- "vision_tower.vision_model.encoder.layers.21.self_attn.q_proj",
175
- "vision_tower.vision_model.encoder.layers.21.self_attn.out_proj",
176
- "vision_tower.vision_model.encoder.layers.21.mlp.fc1",
177
- "vision_tower.vision_model.encoder.layers.21.mlp.fc2",
178
- "vision_tower.vision_model.encoder.layers.22.self_attn.k_proj",
179
- "vision_tower.vision_model.encoder.layers.22.self_attn.v_proj",
180
- "vision_tower.vision_model.encoder.layers.22.self_attn.q_proj",
181
- "vision_tower.vision_model.encoder.layers.22.self_attn.out_proj",
182
- "vision_tower.vision_model.encoder.layers.22.mlp.fc1",
183
- "vision_tower.vision_model.encoder.layers.22.mlp.fc2",
184
- "vision_tower.vision_model.encoder.layers.23.self_attn.k_proj",
185
- "vision_tower.vision_model.encoder.layers.23.self_attn.v_proj",
186
- "vision_tower.vision_model.encoder.layers.23.self_attn.q_proj",
187
- "vision_tower.vision_model.encoder.layers.23.self_attn.out_proj",
188
- "vision_tower.vision_model.encoder.layers.23.mlp.fc1",
189
- "vision_tower.vision_model.encoder.layers.23.mlp.fc2",
190
- "vision_tower.vision_model.encoder.layers.24.self_attn.k_proj",
191
- "vision_tower.vision_model.encoder.layers.24.self_attn.v_proj",
192
- "vision_tower.vision_model.encoder.layers.24.self_attn.q_proj",
193
- "vision_tower.vision_model.encoder.layers.24.self_attn.out_proj",
194
- "vision_tower.vision_model.encoder.layers.24.mlp.fc1",
195
- "vision_tower.vision_model.encoder.layers.24.mlp.fc2",
196
- "vision_tower.vision_model.encoder.layers.25.self_attn.k_proj",
197
- "vision_tower.vision_model.encoder.layers.25.self_attn.v_proj",
198
- "vision_tower.vision_model.encoder.layers.25.self_attn.q_proj",
199
- "vision_tower.vision_model.encoder.layers.25.self_attn.out_proj",
200
- "vision_tower.vision_model.encoder.layers.25.mlp.fc1",
201
- "vision_tower.vision_model.encoder.layers.25.mlp.fc2",
202
- "vision_tower.vision_model.encoder.layers.26.self_attn.k_proj",
203
- "vision_tower.vision_model.encoder.layers.26.self_attn.v_proj",
204
- "vision_tower.vision_model.encoder.layers.26.self_attn.q_proj",
205
- "vision_tower.vision_model.encoder.layers.26.self_attn.out_proj",
206
- "vision_tower.vision_model.encoder.layers.26.mlp.fc1",
207
- "vision_tower.vision_model.encoder.layers.26.mlp.fc2",
208
- "vision_tower.vision_model.head.mlp.fc1",
209
- "vision_tower.vision_model.head.mlp.fc2",
210
- "multi_modal_projector.linear_1",
211
- "multi_modal_projector.linear_2",
212
- "language_model.lm_head"
213
  ],
214
  "kv_cache_scheme": null,
215
  "quant_method": "compressed-tensors",
216
- "quantization_status": "compressed"
 
 
 
217
  },
218
  "text_config": {
219
  "_name_or_path": "meta-llama/Llama-3.1-8B-Instruct",
@@ -223,6 +62,7 @@
223
  "attention_bias": false,
224
  "attention_dropout": 0.0,
225
  "bos_token_id": 128000,
 
226
  "eos_token_id": [
227
  128001,
228
  128008,
@@ -249,18 +89,18 @@
249
  "rope_type": "llama3"
250
  },
251
  "rope_theta": 500000.0,
252
- "torch_dtype": "bfloat16",
253
  "use_cache": true,
254
  "vocab_size": 128256
255
  },
256
- "torch_dtype": "bfloat16",
257
- "transformers_version": "4.51.3",
258
  "vision_config": {
259
  "_name_or_path": "google/siglip2-so400m-patch14-384",
260
  "architectures": [
261
  "SiglipVisionModel"
262
  ],
263
  "attention_dropout": 0.0,
 
264
  "hidden_act": "gelu_pytorch_tanh",
265
  "hidden_size": 1152,
266
  "image_size": 384,
@@ -270,8 +110,7 @@
270
  "num_attention_heads": 16,
271
  "num_channels": 3,
272
  "num_hidden_layers": 27,
273
- "patch_size": 14,
274
- "torch_dtype": "bfloat16"
275
  },
276
  "vision_feature_layer": -2,
277
  "vision_feature_select_strategy": "full"
 
2
  "architectures": [
3
  "LlavaForConditionalGeneration"
4
  ],
5
+ "dtype": "bfloat16",
6
  "image_seq_length": 729,
7
  "image_token_index": 128077,
8
  "model_type": "llava",
 
11
  "quantization_config": {
12
  "config_groups": {
13
  "group_0": {
14
+ "format": "float-quantized",
15
  "input_activations": {
16
  "actorder": null,
17
  "block_structure": null,
 
45
  "format": "float-quantized",
46
  "global_compression_ratio": null,
47
  "ignore": [
48
+ "lm_head"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
49
  ],
50
  "kv_cache_scheme": null,
51
  "quant_method": "compressed-tensors",
52
+ "quantization_status": "compressed",
53
+ "sparsity_config": {},
54
+ "transform_config": {},
55
+ "version": "0.11.0"
56
  },
57
  "text_config": {
58
  "_name_or_path": "meta-llama/Llama-3.1-8B-Instruct",
 
62
  "attention_bias": false,
63
  "attention_dropout": 0.0,
64
  "bos_token_id": 128000,
65
+ "dtype": "bfloat16",
66
  "eos_token_id": [
67
  128001,
68
  128008,
 
89
  "rope_type": "llama3"
90
  },
91
  "rope_theta": 500000.0,
 
92
  "use_cache": true,
93
  "vocab_size": 128256
94
  },
95
+ "tie_word_embeddings": false,
96
+ "transformers_version": "4.56.1",
97
  "vision_config": {
98
  "_name_or_path": "google/siglip2-so400m-patch14-384",
99
  "architectures": [
100
  "SiglipVisionModel"
101
  ],
102
  "attention_dropout": 0.0,
103
+ "dtype": "bfloat16",
104
  "hidden_act": "gelu_pytorch_tanh",
105
  "hidden_size": 1152,
106
  "image_size": 384,
 
110
  "num_attention_heads": 16,
111
  "num_channels": 3,
112
  "num_hidden_layers": 27,
113
+ "patch_size": 14
 
114
  },
115
  "vision_feature_layer": -2,
116
  "vision_feature_select_strategy": "full"
generation_config.json CHANGED
@@ -1,10 +1,13 @@
1
  {
2
  "_from_model_config": true,
3
  "bos_token_id": 128000,
 
4
  "eos_token_id": [
5
  128001,
6
  128008,
7
  128009
8
  ],
9
- "transformers_version": "4.51.3"
 
 
10
  }
 
1
  {
2
  "_from_model_config": true,
3
  "bos_token_id": 128000,
4
+ "do_sample": true,
5
  "eos_token_id": [
6
  128001,
7
  128008,
8
  128009
9
  ],
10
+ "temperature": 0.6,
11
+ "top_p": 0.9,
12
+ "transformers_version": "4.56.1"
13
  }
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad17bb513ed7f4f788ade08ee2306627a549b3f281d4f0a73a75935e09485c5f
3
+ size 4999640960
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dd782e04b768e5c77e92b06fb207746a55a7c467b4a3b84b72b70e37fc3a71cc
3
+ size 4542015320
model.safetensors.index.json CHANGED
The diff for this file is too large to render. See raw diff
 
recipe.yaml CHANGED
@@ -1,6 +1,6 @@
1
  default_stage:
2
  default_modifiers:
3
  QuantizationModifier:
4
- ignore: ['re:.*lm_head', 're:multi_modal_projector.*', 're:vision_tower.*']
5
  targets: [Linear]
 
6
  scheme: FP8_DYNAMIC
 
1
  default_stage:
2
  default_modifiers:
3
  QuantizationModifier:
 
4
  targets: [Linear]
5
+ ignore: ['re:.*lm_head', 're:multi_modal_projector.*', 're:vision_tower.*']
6
  scheme: FP8_DYNAMIC
tokenizer_config.json CHANGED
@@ -2050,7 +2050,6 @@
2050
  }
2051
  },
2052
  "bos_token": "<|begin_of_text|>",
2053
- "chat_template": "{{- bos_token }}\n{%- if not date_string is defined %}\n {%- set date_string = \"26 July 2024\" %}\n{%- endif %}\n\n{#- This block extracts the system message, so we can slot it into the right place. #}\n{%- if messages[0]['role'] == 'system' %}\n {%- set system_message = messages[0]['content'] %}\n {%- set messages = messages[1:] %}\n{%- else %}\n {%- set system_message = \"\" %}\n{%- endif %}\n\n{#- System message + builtin tools #}\n{{- \"<|start_header_id|>system<|end_header_id|>\n\n\" }}\n{{- \"Cutting Knowledge Date: December 2023\n\" }}\n{{- \"Today Date: \" + date_string + \"\n\n\" }}\n{{- system_message }}\n{{- \"<|eot_id|>\" }}\n\n\n{%- set first_user_message = True %}\n{%- for message in messages %}\n {%- if first_user_message and message['role'] == 'user' %}\n\t\t{%- set first_user_message = False %}\n\t {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n<|reserved_special_token_70|><|reserved_special_token_69|><|reserved_special_token_71|>'+ message['content'] + '<|eot_id|>' }}\n\t{%- else %}\n {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] + '<|eot_id|>' }}\n\t{%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|start_header_id|>assistant<|end_header_id|>\n\n' }}\n{%- endif %}\n",
2054
  "clean_up_tokenization_spaces": true,
2055
  "eos_token": "<|eot_id|>",
2056
  "extra_special_tokens": {},
@@ -2060,5 +2059,5 @@
2060
  ],
2061
  "model_max_length": 131072,
2062
  "processor_class": "LlavaProcessor",
2063
- "tokenizer_class": "PreTrainedTokenizer"
2064
  }
 
2050
  }
2051
  },
2052
  "bos_token": "<|begin_of_text|>",
 
2053
  "clean_up_tokenization_spaces": true,
2054
  "eos_token": "<|eot_id|>",
2055
  "extra_special_tokens": {},
 
2059
  ],
2060
  "model_max_length": 131072,
2061
  "processor_class": "LlavaProcessor",
2062
+ "tokenizer_class": "PreTrainedTokenizerFast"
2063
  }