-
Notifications
You must be signed in to change notification settings - Fork 169
/
run_predict.py
168 lines (147 loc) · 6.3 KB
/
run_predict.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import argparse
import paddle
from paddlenlp.generation import TextStreamer
from paddlemix.models.llava.constants import (
DEFAULT_IM_END_TOKEN,
DEFAULT_IM_START_TOKEN,
DEFAULT_IMAGE_TOKEN,
)
from paddlemix.models.llava.conversation import conv_templates
from paddlemix.models.llava.mm_utils import get_model_name_from_path, load_image
from paddlemix.utils.log import logger
import os
from paddlemix.models.llava.language_model.llava_llama import (
LlavaConfig,
LlavaLlamaForCausalLM,
)
from paddlemix.models.llava.language_model.tokenizer import LLavaTokenizer
from paddlemix.processors import LlavaProcessor
from paddlenlp.transformers import CLIPImageProcessor
def main(args):
paddle.seed(seed=0)
compute_dtype = "float16" if args.fp16 else "bfloat16"
if "npu" in paddle.get_device():
is_bfloat16_supported = True
else:
is_bfloat16_supported = paddle.amp.is_bfloat16_supported()
if compute_dtype == "bfloat16" and not is_bfloat16_supported:
logger.warning("bfloat16 is not supported on your device,change to float32")
compute_dtype = "float32"
model_name = get_model_name_from_path(args.model_path)
model_name_or_path = args.model_path
tokenizer = LLavaTokenizer.from_pretrained(model_name_or_path)
model_config = LlavaConfig.from_pretrained(model_name_or_path)
model = LlavaLlamaForCausalLM.from_pretrained(model_name_or_path, dtype=compute_dtype)
model.eval()
name_or_path = (os.path.join(model_name_or_path, "processor", "eval"))
image_processor = CLIPImageProcessor.from_pretrained(name_or_path)
processor = LlavaProcessor(
image_processor,
tokenizer,
max_length=args.max_new_tokens,
image_aspect_ratio=model_config.image_aspect_ratio
)
model.resize_token_embeddings(len(tokenizer))
vision_tower = model.get_vision_tower()
vision_tower.load_model()
if "llama-2" in model_name.lower():
conv_mode = "llava_llama_2"
elif "mistral" in model_name.lower():
conv_mode = "mistral_instruct"
elif "v1.6-34b" in model_name.lower():
conv_mode = "chatml_direct"
elif "v1" in model_name.lower():
conv_mode = "llava_v1"
elif "mpt" in model_name.lower():
conv_mode = "mpt"
else:
conv_mode = "llava_v0"
if args.conv_mode is not None and conv_mode != args.conv_mode:
print(
"[WARNING] the auto inferred conversation mode is {}, while `--conv-mode` is {}, using {}".format(
conv_mode, args.conv_mode, args.conv_mode
)
)
else:
args.conv_mode = conv_mode
conv = conv_templates[args.conv_mode].copy()
first_message = True
inp = args.question
if args.image_file is not None and first_message:
if model_config.mm_use_im_start_end:
inp = DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_TOKEN + DEFAULT_IM_END_TOKEN + "\n" + inp
else:
inp = DEFAULT_IMAGE_TOKEN + "\n" + inp
conv.append_message(conv.roles[0], inp)
first_message = False
else:
conv.append_message(conv.roles[0], inp)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()
record = {"image": args.image_file, "conversations": prompt}
image_size = load_image(args.image_file).size
data_dict = processor(record=record, image_aspect_ratio=model_config.image_aspect_ratio)
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
if args.benchmark:
import time
start = 0.0
total = 0.0
for i in range(20):
if i>10:
start = time.time()
with paddle.no_grad():
output_ids = model.generate(
input_ids=data_dict["input_ids"],
images=paddle.cast(data_dict["images"], compute_dtype),
image_sizes=[image_size],
decode_strategy="sampling" if args.temperature > 0 else "greedy_search",
temperature=args.temperature,
max_new_tokens=args.max_new_tokens,
streamer=streamer,
use_cache=True,
)
if i > 10:
total += time.time()-start
print("Time: ", total/10)
print("temperature: ", args.temperature)
print("compute_dtype:", compute_dtype)
else:
with paddle.no_grad():
output_ids = model.generate(
input_ids=data_dict["input_ids"],
images=paddle.cast(data_dict["images"], compute_dtype),
image_sizes=[image_size],
decode_strategy="sampling" if args.temperature > 0 else "greedy_search",
temperature=args.temperature,
max_new_tokens=args.max_new_tokens,
streamer=streamer,
use_cache=True,
)
outputs = tokenizer.decode(output_ids[0][0]).strip().split("<|im_end|>")[0].split("</s>")[0]
print("outputs:\n", outputs)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--model-path", type=str, default="liuhaotian/llava-v1.6-vicuna-7b")
parser.add_argument("--question", type=str, default="What is shown in this image?")
parser.add_argument("--image-file", type=str, required=True)
parser.add_argument("--conv-mode", type=str, default=None)
parser.add_argument("--temperature", type=float, default=0.2)
parser.add_argument("--max-new-tokens", type=int, default=512)
parser.add_argument("--fp16", action="store_true")
parser.add_argument("--debug", action="store_true")
parser.add_argument("--benchmark", action="store_true")
args = parser.parse_args()
main(args)