prince-canuma commited on
Commit
43121de
1 Parent(s): cb37780

Upload folder using huggingface_hub

Browse files
added_tokens.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "<|endoftext|>": 151643,
3
+ "<|im_end|>": 151645,
4
+ "<|im_start|>": 151644
5
+ }
config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "BunnyQwenForCausalLM"
4
+ ],
5
+ "attention_dropout": 0.0,
6
+ "auto_map": {
7
+ "AutoConfig": "configuration_llava_qwen2.LlavaQwen2Config",
8
+ "AutoModelForCausalLM": "modeling_llava_qwen2.LlavaQwen2ForCausalLM"
9
+ },
10
+ "eos_token_id": 151645,
11
+ "freeze_mm_mlp_adapter": false,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 8192,
14
+ "image_aspect_ratio": "pad",
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 29568,
17
+ "language_model": "cognitivecomputations/dolphin-2.9.2-qwen2-72b",
18
+ "max_position_embeddings": 131072,
19
+ "max_window_layers": 28,
20
+ "mm_hidden_size": 1152,
21
+ "mm_projector_lr": null,
22
+ "mm_projector_type": "mlp2x_gelu",
23
+ "mm_vision_tower": "google/siglip-so400m-patch14-384",
24
+ "model_type": "llava-qwen2",
25
+ "num_attention_heads": 64,
26
+ "num_hidden_layers": 80,
27
+ "num_key_value_heads": 8,
28
+ "quantization": {
29
+ "group_size": 64,
30
+ "bits": 4
31
+ },
32
+ "rms_norm_eps": 1e-05,
33
+ "rope_theta": 1000000.0,
34
+ "sliding_window": 4096,
35
+ "tie_word_embeddings": false,
36
+ "tokenizer_model_max_length": 4096,
37
+ "tokenizer_padding_side": "right",
38
+ "torch_dtype": "bfloat16",
39
+ "transformers_version": "4.41.2",
40
+ "tune_mm_mlp_adapter": false,
41
+ "use_cache": true,
42
+ "use_mm_proj": true,
43
+ "use_sliding_window": false,
44
+ "vision_config": {
45
+ "intermediate_size": 4352
46
+ },
47
+ "vocab_size": 152064
48
+ }
configuration_llava_qwen2.py ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2024 The Qwen team, Alibaba Group and the HuggingFace Inc. team. All rights reserved.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+ """ Qwen2 model configuration"""
16
+
17
+ from transformers.configuration_utils import PretrainedConfig
18
+ from transformers.utils import logging
19
+
20
+
21
+ logger = logging.get_logger(__name__)
22
+
23
+ QWEN2_PRETRAINED_CONFIG_ARCHIVE_MAP = {
24
+ "Qwen/Qwen2-7B-beta": "https://huggingface.co/Qwen/Qwen2-7B-beta/resolve/main/config.json",
25
+ }
26
+
27
+
28
+ class Qwen2Config(PretrainedConfig):
29
+ r"""
30
+ This is the configuration class to store the configuration of a [`Qwen2Model`]. It is used to instantiate a
31
+ Qwen2 model according to the specified arguments, defining the model architecture. Instantiating a configuration
32
+ with the defaults will yield a similar configuration to that of
33
+ Qwen2-7B-beta [Qwen/Qwen2-7B-beta](https://huggingface.co/Qwen/Qwen2-7B-beta).
34
+
35
+ Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
36
+ documentation from [`PretrainedConfig`] for more information.
37
+
38
+
39
+ Args:
40
+ vocab_size (`int`, *optional*, defaults to 151936):
41
+ Vocabulary size of the Qwen2 model. Defines the number of different tokens that can be represented by the
42
+ `inputs_ids` passed when calling [`Qwen2Model`]
43
+ hidden_size (`int`, *optional*, defaults to 4096):
44
+ Dimension of the hidden representations.
45
+ intermediate_size (`int`, *optional*, defaults to 22016):
46
+ Dimension of the MLP representations.
47
+ num_hidden_layers (`int`, *optional*, defaults to 32):
48
+ Number of hidden layers in the Transformer encoder.
49
+ num_attention_heads (`int`, *optional*, defaults to 32):
50
+ Number of attention heads for each attention layer in the Transformer encoder.
51
+ num_key_value_heads (`int`, *optional*, defaults to 32):
52
+ This is the number of key_value heads that should be used to implement Grouped Query Attention. If
53
+ `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
54
+ `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
55
+ converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
56
+ by meanpooling all the original heads within that group. For more details checkout [this
57
+ paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `32`.
58
+ hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
59
+ The non-linear activation function (function or string) in the decoder.
60
+ max_position_embeddings (`int`, *optional*, defaults to 32768):
61
+ The maximum sequence length that this model might ever be used with.
62
+ initializer_range (`float`, *optional*, defaults to 0.02):
63
+ The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
64
+ rms_norm_eps (`float`, *optional*, defaults to 1e-06):
65
+ The epsilon used by the rms normalization layers.
66
+ use_cache (`bool`, *optional*, defaults to `True`):
67
+ Whether or not the model should return the last key/values attentions (not used by all models). Only
68
+ relevant if `config.is_decoder=True`.
69
+ tie_word_embeddings (`bool`, *optional*, defaults to `False`):
70
+ Whether the model's input and output word embeddings should be tied.
71
+ rope_theta (`float`, *optional*, defaults to 10000.0):
72
+ The base period of the RoPE embeddings.
73
+ use_sliding_window (`bool`, *optional*, defaults to `False`):
74
+ Whether to use sliding window attention.
75
+ sliding_window (`int`, *optional*, defaults to 4096):
76
+ Sliding window attention (SWA) window size. If not specified, will default to `4096`.
77
+ max_window_layers (`int`, *optional*, defaults to 28):
78
+ The number of layers that use SWA (Sliding Window Attention). The bottom layers use SWA while the top use full attention.
79
+ attention_dropout (`float`, *optional*, defaults to 0.0):
80
+ The dropout ratio for the attention probabilities.
81
+
82
+ ```python
83
+ >>> from transformers import Qwen2Model, Qwen2Config
84
+
85
+ >>> # Initializing a Qwen2 style configuration
86
+ >>> configuration = Qwen2Config()
87
+
88
+ >>> # Initializing a model from the Qwen2-7B style configuration
89
+ >>> model = Qwen2Model(configuration)
90
+
91
+ >>> # Accessing the model configuration
92
+ >>> configuration = model.config
93
+ ```"""
94
+
95
+ model_type = "qwen2"
96
+ keys_to_ignore_at_inference = ["past_key_values"]
97
+
98
+ def __init__(
99
+ self,
100
+ vocab_size=151936,
101
+ hidden_size=4096,
102
+ intermediate_size=22016,
103
+ num_hidden_layers=32,
104
+ num_attention_heads=32,
105
+ num_key_value_heads=32,
106
+ hidden_act="silu",
107
+ max_position_embeddings=32768,
108
+ initializer_range=0.02,
109
+ rms_norm_eps=1e-6,
110
+ use_cache=True,
111
+ tie_word_embeddings=False,
112
+ rope_theta=10000.0,
113
+ use_sliding_window=False,
114
+ sliding_window=4096,
115
+ max_window_layers=28,
116
+ attention_dropout=0.0,
117
+ **kwargs,
118
+ ):
119
+ self.vocab_size = vocab_size
120
+ self.max_position_embeddings = max_position_embeddings
121
+ self.hidden_size = hidden_size
122
+ self.intermediate_size = intermediate_size
123
+ self.num_hidden_layers = num_hidden_layers
124
+ self.num_attention_heads = num_attention_heads
125
+ self.use_sliding_window = use_sliding_window
126
+ self.sliding_window = sliding_window
127
+ self.max_window_layers = max_window_layers
128
+
129
+ # for backward compatibility
130
+ if num_key_value_heads is None:
131
+ num_key_value_heads = num_attention_heads
132
+
133
+ self.num_key_value_heads = num_key_value_heads
134
+ self.hidden_act = hidden_act
135
+ self.initializer_range = initializer_range
136
+ self.rms_norm_eps = rms_norm_eps
137
+ self.use_cache = use_cache
138
+ self.rope_theta = rope_theta
139
+ self.attention_dropout = attention_dropout
140
+
141
+ super().__init__(
142
+ tie_word_embeddings=tie_word_embeddings,
143
+ **kwargs,
144
+ )
145
+
146
+ from typing import Union
147
+ from transformers import PretrainedConfig
148
+ import os
149
+
150
+
151
+ class SigLipVisionConfig(PretrainedConfig):
152
+ model_type = "siglip_vision_model"
153
+
154
+ def __init__(
155
+ self,
156
+ hidden_size=1152,
157
+ image_mean=(0.5, 0.5, 0.5),
158
+ intermediate_size=4304,
159
+ num_hidden_layers=27,
160
+ num_attention_heads=16,
161
+ num_channels=3,
162
+ image_size=384,
163
+ patch_size=14,
164
+ hidden_act="gelu_pytorch_tanh",
165
+ layer_norm_eps=1e-6,
166
+ attention_dropout=0.0,
167
+ **kwargs,
168
+ ):
169
+ super().__init__(**kwargs)
170
+
171
+ self.hidden_size = hidden_size
172
+ self.intermediate_size = intermediate_size
173
+ self.num_hidden_layers = num_hidden_layers
174
+ self.num_attention_heads = num_attention_heads
175
+ self.num_channels = num_channels
176
+ self.patch_size = patch_size
177
+ self.image_size = image_size
178
+ self.attention_dropout = attention_dropout
179
+ self.layer_norm_eps = layer_norm_eps
180
+ self.hidden_act = hidden_act
181
+ self.image_mean = image_mean
182
+
183
+ @classmethod
184
+ def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
185
+ cls._set_token_in_kwargs(kwargs)
186
+
187
+ config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
188
+
189
+ # get the vision config dict if we are loading from SigLipConfig
190
+ if config_dict.get("model_type") == "siglip":
191
+ config_dict = config_dict["vision_config"]
192
+
193
+ if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
194
+ logger.warning(
195
+ f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
196
+ f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
197
+ )
198
+
199
+ return cls.from_dict(config_dict, **kwargs)
200
+
201
+ class LlavaQwen2Config(Qwen2Config):
202
+ model_type = "llava-qwen2"
dolphin_vision_streamlit.py ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import streamlit as st
2
+ import torch
3
+ import transformers
4
+ from transformers import AutoModelForCausalLM, AutoTokenizer
5
+ from PIL import Image
6
+ import warnings
7
+
8
+ # Disable warnings and progress bars
9
+ transformers.logging.set_verbosity_error()
10
+ transformers.logging.disable_progress_bar()
11
+ warnings.filterwarnings('ignore')
12
+
13
+ # Set device
14
+ device = 'cuda' if torch.cuda.is_available() else 'cpu'
15
+ torch.set_default_device(device)
16
+
17
+ @st.cache_resource
18
+ def load_model():
19
+ model_name = 'cognitivecomputations/dolphin-vision-72b'
20
+ model = AutoModelForCausalLM.from_pretrained(
21
+ model_name,
22
+ torch_dtype=torch.float16,
23
+ device_map='auto',
24
+ trust_remote_code=True
25
+ )
26
+ tokenizer = AutoTokenizer.from_pretrained(
27
+ model_name,
28
+ trust_remote_code=True
29
+ )
30
+ return model, tokenizer
31
+
32
+ def generate_response(model, tokenizer, prompt, image=None):
33
+ messages = [
34
+ {"role": "user", "content": f'<image>\n{prompt}' if image else prompt}
35
+ ]
36
+ text = tokenizer.apply_chat_template(
37
+ messages,
38
+ tokenize=False,
39
+ add_generation_prompt=True
40
+ )
41
+ text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
42
+ input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0)
43
+
44
+ if image:
45
+ image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)
46
+ else:
47
+ image_tensor = None
48
+
49
+ output_ids = model.generate(
50
+ input_ids,
51
+ images=image_tensor,
52
+ max_new_tokens=2048,
53
+ use_cache=True
54
+ )[0]
55
+
56
+ return tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip()
57
+
58
+ st.title("Chat with DolphinVision 🐬")
59
+
60
+ model, tokenizer = load_model()
61
+
62
+ uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])
63
+ image = None
64
+ if uploaded_file is not None:
65
+ image = Image.open(uploaded_file)
66
+ st.image(image, caption='Uploaded Image', use_column_width=True)
67
+
68
+ user_input = st.text_input("You:", "")
69
+
70
+ if st.button("Send"):
71
+ if user_input:
72
+ with st.spinner("Generating response..."):
73
+ response = generate_response(model, tokenizer, user_input, image)
74
+ st.text_area("DolphinVision:", value=response, height=200)
75
+ else:
76
+ st.warning("Please enter a message.")
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model-00001-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a21190033bcd6b3fb0377facb2beaea216f1fe4ef6816efbfe1ab5c9e42d3465
3
+ size 5270244362
model-00002-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:383cf6d631d066548d4ba726dbb222baccedebc1447745dbc97250ac99479e92
3
+ size 5294882509
model-00003-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d2b7b3a94a8146cecfb0b06062df77c0bf76622d1e1addbae15827a22d45aabd
3
+ size 5294882501
model-00004-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf1e0e21fde14fad2ef7260ddc21349ea345aa5fc2e79f9e75c8c64e9e66645b
3
+ size 5346175142
model-00005-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f94af4ad340040ac0b3f040c3a31c3f85b1aa9b691a8e310bfbf68ccb9589bb4
3
+ size 5294849492
model-00006-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee0c60ee7af2d37b5aaa62bdb6baa2324319b7b5e48dbcb335fbf1c1963b3d15
3
+ size 5294882532
model-00007-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:acd93cc776a5db9af43593423573f8333ecf17480d72779cbceb9831681f34a3
3
+ size 5294882491
model-00008-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:95a38df43e2a5464e422eda53c86bdbe83f5b356b34a4df1a82583e8a7193c0a
3
+ size 4115036455
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
modeling_llava_qwen2.py ADDED
The diff for this file is too large to render. See raw diff
 
special_tokens_map.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>"
5
+ ],
6
+ "eos_token": {
7
+ "content": "<|im_end|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false
12
+ },
13
+ "pad_token": {
14
+ "content": "<|endoftext|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false
19
+ }
20
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "151643": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "151644": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "151645": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ }
28
+ },
29
+ "additional_special_tokens": [
30
+ "<|im_start|>",
31
+ "<|im_end|>"
32
+ ],
33
+ "bos_token": null,
34
+ "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nAnswer the questions.<|im_end|>' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
35
+ "clean_up_tokenization_spaces": false,
36
+ "eos_token": "<|im_end|>",
37
+ "errors": "replace",
38
+ "model_max_length": 4096,
39
+ "pad_token": "<|endoftext|>",
40
+ "padding_side": "right",
41
+ "split_special_tokens": false,
42
+ "tokenizer_class": "Qwen2Tokenizer",
43
+ "unk_token": null
44
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff