Fix mdoel loading (#17)
Browse files- Fix mdoel loading (bd36ab0de70e8c918ebaef21ab674e45760e489b)
- Update README.md (333d6e0ec3210304dc43a4c7d75731378c4aa612)
- Update modeling_gme_qwen2vl.py (280b82673f3afce78487f0287755a7cc23199420)
- Update config.json (4174ed03d1015398958742e9dcc4c34277640a4f)
- Update README.md (dca92bddc7528a380b87c21f5383c8c856c27143)
- Update modeling_gme_qwen2vl.py (39953e5c545ac83822ad50cb3ef09214db7fbba5)
- Update README.md (62f5d69e260988e066a2e60c59180d25151d5e9b)
- README.md +12 -0
- config.json +7 -4
- modeling_gme_qwen2vl.py +39 -16
README.md
CHANGED
@@ -3696,7 +3696,19 @@ The `GME` models support three types of input: **text**, **image**, and **image-
|
|
3696 |
|
3697 |
**Transformers**
|
3698 |
|
|
|
|
|
3699 |
```python
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3700 |
t2i_prompt = 'Find an image that matches the given text.'
|
3701 |
texts = [
|
3702 |
"The Tesla Cybertruck is a battery electric pickup truck built by Tesla, Inc. since 2023.",
|
|
|
3696 |
|
3697 |
**Transformers**
|
3698 |
|
3699 |
+
The remote code has some issues with `transformers>=4.52.0`, please downgrade or use `sentence_transformers`
|
3700 |
+
|
3701 |
```python
|
3702 |
+
from transformers import AutoModel
|
3703 |
+
from transformers.utils.versions import require_version
|
3704 |
+
|
3705 |
+
|
3706 |
+
require_version(
|
3707 |
+
"transformers<4.52.0",
|
3708 |
+
"The remote code has some issues with transformers>=4.52.0, please downgrade: pip install transformers==4.51.3"
|
3709 |
+
)
|
3710 |
+
|
3711 |
+
|
3712 |
t2i_prompt = 'Find an image that matches the given text.'
|
3713 |
texts = [
|
3714 |
"The Tesla Cybertruck is a battery electric pickup truck built by Tesla, Inc. since 2023.",
|
config.json
CHANGED
@@ -1,9 +1,12 @@
|
|
1 |
{
|
2 |
"_name_or_path": "Alibaba-NLP/gme-Qwen2-VL-2B-Instruct",
|
3 |
-
"architectures": [
|
|
|
|
|
|
|
4 |
"auto_map": {
|
5 |
-
"
|
6 |
-
"
|
7 |
},
|
8 |
"attention_dropout": 0.0,
|
9 |
"bos_token_id": 151643,
|
@@ -15,7 +18,7 @@
|
|
15 |
"intermediate_size": 8960,
|
16 |
"max_position_embeddings": 32768,
|
17 |
"max_window_layers": 28,
|
18 |
-
"model_type": "
|
19 |
"num_attention_heads": 12,
|
20 |
"num_hidden_layers": 28,
|
21 |
"num_key_value_heads": 2,
|
|
|
1 |
{
|
2 |
"_name_or_path": "Alibaba-NLP/gme-Qwen2-VL-2B-Instruct",
|
3 |
+
"architectures": [
|
4 |
+
"Qwen2VLForConditionalGeneration",
|
5 |
+
"GmeQwen2VL"
|
6 |
+
],
|
7 |
"auto_map": {
|
8 |
+
"AutoConfig": "modeling_gme_qwen2vl.GmeQwen2VLConfig",
|
9 |
+
"AutoModel": "modeling_gme_qwen2vl.GmeQwen2VL"
|
10 |
},
|
11 |
"attention_dropout": 0.0,
|
12 |
"bos_token_id": 151643,
|
|
|
18 |
"intermediate_size": 8960,
|
19 |
"max_position_embeddings": 32768,
|
20 |
"max_window_layers": 28,
|
21 |
+
"model_type": "qwen2_vl",
|
22 |
"num_attention_heads": 12,
|
23 |
"num_hidden_layers": 28,
|
24 |
"num_key_value_heads": 2,
|
modeling_gme_qwen2vl.py
CHANGED
@@ -12,16 +12,25 @@ import torch
|
|
12 |
from PIL import Image
|
13 |
from torch.utils.data import DataLoader
|
14 |
from tqdm.autonotebook import tqdm
|
15 |
-
from transformers import
|
16 |
-
|
17 |
-
|
18 |
Qwen2VLConfig,
|
19 |
Qwen2VLForConditionalGeneration,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
)
|
21 |
-
import os
|
22 |
|
23 |
|
24 |
class GmeQwen2VLConfig(Qwen2VLConfig):
|
|
|
|
|
25 |
def __init__(
|
26 |
self,
|
27 |
min_image_tokens: int = 256,
|
@@ -35,14 +44,25 @@ class GmeQwen2VLConfig(Qwen2VLConfig):
|
|
35 |
self.max_length = max_length
|
36 |
|
37 |
|
38 |
-
class
|
39 |
config_class = GmeQwen2VLConfig
|
40 |
-
base_model_prefix
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
def __init__(self, config: GmeQwen2VLConfig, **kwargs: Any) -> None:
|
43 |
super().__init__(config)
|
44 |
-
self.
|
45 |
-
self.
|
|
|
|
|
|
|
46 |
|
47 |
min_pixels: int = config.min_image_tokens * 28 * 28
|
48 |
max_pixels: int = config.max_image_tokens * 28 * 28
|
@@ -55,6 +75,9 @@ class GmeQwen2VLForVision2Seq(PreTrainedModel):
|
|
55 |
self.default_instruction: str = "You are a helpful assistant."
|
56 |
self.sep: str = " "
|
57 |
|
|
|
|
|
|
|
58 |
def forward(
|
59 |
self,
|
60 |
input_ids: Optional[torch.LongTensor] = None,
|
@@ -70,21 +93,21 @@ class GmeQwen2VLForVision2Seq(PreTrainedModel):
|
|
70 |
**kwargs
|
71 |
) -> torch.Tensor:
|
72 |
if inputs_embeds is None:
|
73 |
-
inputs_embeds = self.
|
74 |
if pixel_values is not None:
|
75 |
-
pixel_values = pixel_values.type(self.
|
76 |
-
image_embeds = self.
|
77 |
-
image_mask = input_ids == self.
|
78 |
inputs_embeds[image_mask] = image_embeds
|
79 |
# if pixel_values_videos is not None:
|
80 |
-
# pixel_values_videos = pixel_values_videos.type(self.
|
81 |
-
# video_embeds = self.
|
82 |
-
# video_mask = input_ids == self.
|
83 |
# inputs_embeds[video_mask] = video_embeds
|
84 |
if attention_mask is not None:
|
85 |
attention_mask = attention_mask.to(inputs_embeds.device)
|
86 |
|
87 |
-
outputs = self.
|
88 |
input_ids=None,
|
89 |
position_ids=position_ids,
|
90 |
attention_mask=attention_mask,
|
|
|
12 |
from PIL import Image
|
13 |
from torch.utils.data import DataLoader
|
14 |
from tqdm.autonotebook import tqdm
|
15 |
+
from transformers import AutoProcessor, PreTrainedModel
|
16 |
+
from transformers.models.qwen2_vl.modeling_qwen2_vl import (
|
17 |
+
Qwen2VisionTransformerPretrainedModel,
|
18 |
Qwen2VLConfig,
|
19 |
Qwen2VLForConditionalGeneration,
|
20 |
+
Qwen2VLModel,
|
21 |
+
)
|
22 |
+
from transformers.utils.versions import require_version
|
23 |
+
|
24 |
+
|
25 |
+
require_version(
|
26 |
+
"transformers<4.52.0",
|
27 |
+
"This code has some issues with transformers>=4.52.0, please downgrade: pip install transformers==4.51.3"
|
28 |
)
|
|
|
29 |
|
30 |
|
31 |
class GmeQwen2VLConfig(Qwen2VLConfig):
|
32 |
+
# model_type = ''
|
33 |
+
|
34 |
def __init__(
|
35 |
self,
|
36 |
min_image_tokens: int = 256,
|
|
|
44 |
self.max_length = max_length
|
45 |
|
46 |
|
47 |
+
class GmeQwen2VL(PreTrainedModel):
|
48 |
config_class = GmeQwen2VLConfig
|
49 |
+
base_model_prefix = "model"
|
50 |
+
supports_gradient_checkpointing = True
|
51 |
+
_no_split_modules = ["Qwen2VLDecoderLayer", "Qwen2VLVisionBlock"]
|
52 |
+
# _skip_keys_device_placement = "past_key_values"
|
53 |
+
_supports_flash_attn_2 = True
|
54 |
+
_supports_sdpa = True
|
55 |
+
# _supports_cache_class = True
|
56 |
+
_supports_static_cache = False # TODO (joao): fix. torch.compile failing probably due to `cache_positions`
|
57 |
+
# _tied_weights_keys = ["lm_head.weight"]
|
58 |
|
59 |
def __init__(self, config: GmeQwen2VLConfig, **kwargs: Any) -> None:
|
60 |
super().__init__(config)
|
61 |
+
self.visual = Qwen2VisionTransformerPretrainedModel._from_config(config.vision_config)
|
62 |
+
self.model = Qwen2VLModel(config)
|
63 |
+
self.vocab_size = config.vocab_size
|
64 |
+
# self.lm_head = torch.nn.Linear(config.hidden_size, config.vocab_size, bias=False)
|
65 |
+
self.rope_deltas = None # cache rope_deltas here
|
66 |
|
67 |
min_pixels: int = config.min_image_tokens * 28 * 28
|
68 |
max_pixels: int = config.max_image_tokens * 28 * 28
|
|
|
75 |
self.default_instruction: str = "You are a helpful assistant."
|
76 |
self.sep: str = " "
|
77 |
|
78 |
+
# Initialize weights and apply final processing
|
79 |
+
self.post_init()
|
80 |
+
|
81 |
def forward(
|
82 |
self,
|
83 |
input_ids: Optional[torch.LongTensor] = None,
|
|
|
93 |
**kwargs
|
94 |
) -> torch.Tensor:
|
95 |
if inputs_embeds is None:
|
96 |
+
inputs_embeds = self.model.get_input_embeddings()(input_ids)
|
97 |
if pixel_values is not None:
|
98 |
+
pixel_values = pixel_values.type(self.visual.get_dtype())
|
99 |
+
image_embeds = self.visual(pixel_values, grid_thw=image_grid_thw).to(inputs_embeds.device)
|
100 |
+
image_mask = input_ids == self.config.image_token_id
|
101 |
inputs_embeds[image_mask] = image_embeds
|
102 |
# if pixel_values_videos is not None:
|
103 |
+
# pixel_values_videos = pixel_values_videos.type(self.visual.get_dtype())
|
104 |
+
# video_embeds = self.visual(pixel_values_videos, grid_thw=video_grid_thw).to(inputs_embeds.device)
|
105 |
+
# video_mask = input_ids == self.config.video_token_id
|
106 |
# inputs_embeds[video_mask] = video_embeds
|
107 |
if attention_mask is not None:
|
108 |
attention_mask = attention_mask.to(inputs_embeds.device)
|
109 |
|
110 |
+
outputs = self.model(
|
111 |
input_ids=None,
|
112 |
position_ids=position_ids,
|
113 |
attention_mask=attention_mask,
|