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
169
170
171
172
173
174
175
176
177
178
179
|
#!/usr/bin/env python3
import argparse
import datasets
import json
import os
import requests
import sys
import torch
from glob import glob
from PIL import Image
from transformers import AutoProcessor, AutoModelForImageTextToText
def clean_str(s):
return s.removeprefix("```json").removesuffix("```").replace("\n", "").strip()
def captioning_prompt(image):
return [
{
"role": "system",
"content": [{"type": "text", "text": "You are a professional English-German translator and also a renowned photography critic."}]
},
{
"role": "user",
"content": [
{"type": "image", "image": image},
{"type": "text", "text": "Write a detailed caption for this image in a single sentence. Translate the caption into German. The output needs to be JSON, the keys being 'English' and 'German' for the respective captions. Only output the JSON, nothing else."}
]
}
]
def captioning_prompt_with_source(image, source):
prompt = captioning_prompt(image)
prefix = json.dumps({"English": source}).removesuffix("}") + ', "German": "'
prompt.append({"role": "assistant", "content": [{"type": "text", "text": prefix}]})
return prompt
def translation_prompt(source):
return [
{
"role": "system",
"content": [{"type": "text", "text": "You are a professional English-German translator."}]
},
{
"role": "user",
"content": [
{"type": "text", "text": f"Translate the following caption into German. The output needs to be JSON, the only being 'Translation' for the translation. Only output the JSON, nothing else. Caption: {source}"}
]
}
]
def make_inputs(processor,
messages,
device):
return processor.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt"
).to(device, dtype=torch.bfloat16)
def generate_and_parse(model,
processor,
messages,
args,
example_id=None):
sys.stderr.write(f"Processing {example_id=}\n")
inputs = make_inputs(processor, messages, model.device)
input_len = inputs["input_ids"].shape[-1]
stop_token_ids = [processor.tokenizer.eos_token_id, processor.tokenizer.convert_tokens_to_ids("<end_of_turn>")]
with torch.inference_mode():
generation = model.generate(
**inputs,
max_new_tokens=args.max_new_tokens,
do_sample=not args.do_not_sample,
temperature=args.temperature,
top_p=args.top_p,
top_k=args.top_k,
eos_token_id=stop_token_ids,
disable_compile=True,
)
output_tokens = generation[0][input_len:]
output_text = clean_str(processor.decode(output_tokens, skip_special_tokens=True))
try:
return json.loads(output_text)
except Exception:
if example_id is not None:
sys.stderr.write(
f"Error loading JSON from string '{output_text}' for id={example_id}\n"
)
else:
sys.stderr.write(
f"Error loading JSON from string '{output_text}'\n"
)
return output_text
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--model", default="google/gemma-3-4b-it", type=str)
parser.add_argument("--attention-implementation", default="eager", type=str)
parser.add_argument("--lora-adapter", default=None, type=str)
parser.add_argument("--mode", choices=["from_scratch", "with_prefix", "translate"], type=str, required=True)
parser.add_argument("--dataset", default="asdf2k/caption_translation", type=str)
parser.add_argument("--data-subset", choices=["train", "dev", "test"], default="test", type=str)
parser.add_argument("--max-new-tokens", default=300, type=int)
parser.add_argument("--top-p", default=1.0, type=int)
parser.add_argument("--top-k", default=50, type=int)
parser.add_argument("--temperature", default=0.8, type=int)
parser.add_argument("--do-not-sample", action="store_true")
args = parser.parse_args()
model = AutoModelForImageTextToText.from_pretrained(
args.model,
device_map="cuda",
dtype=torch.bfloat16,
attn_implementation=args.attention_implementation,
).eval()
processor = AutoProcessor.from_pretrained(args.model, use_fast=True)
if args.lora_adapter:
from peft import PeftModel
model = PeftModel.from_pretrained(model, args.lora_adapter)
dataset = datasets.load_dataset(args.dataset)[args.data_subset]
if args.mode == "from_scratch": # Generate caption & translation from scratch
for x in dataset:
output = generate_and_parse(
model,
processor,
captioning_prompt(x["image"]),
args,
example_id=x["id"],
)
print(f"{x['id']}\t{json.dumps(output)}")
elif args.mode == "translate": # Generate German translation given English source
for x in dataset:
input_data = json.loads(x["assistant"])
output = generate_and_parse(
model,
processor,
translation_prompt(input_data["English"]),
args,
example_id=x["id"],
)
output = {"English": input_data["English"], "German": output["Translation"]}
print(f"{x['id']}\t{json.dumps(output)}")
elif args.mode == "with_prefix": # Generate German translation given English caption and image
for x in dataset:
assistant_output_as_input = json.loads(x["assistant"])
output = generate_and_parse(
model,
processor,
captioning_prompt_with_source(x["image"], assistant_output_as_input["English"]),
args,
example_id=x["id"],
)
print(f"{x['id']}\t{json.dumps(output)}")
else:
sys.stderr.write(f"Unkown mode '{args.mode}'")
if __name__ == "__main__":
main()
|
