Spaces:
Running
on
Zero
Running
on
Zero
Update
Browse files
app.py
CHANGED
@@ -18,10 +18,12 @@ if not torch.cuda.is_available():
|
|
18 |
|
19 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
20 |
|
|
|
21 |
MODEL_ID_OPT_6_7B = "Salesforce/blip2-opt-6.7b"
|
|
|
22 |
MODEL_ID_FLAN_T5_XXL = "Salesforce/blip2-flan-t5-xxl"
|
23 |
MODEL_ID = os.getenv("MODEL_ID", MODEL_ID_FLAN_T5_XXL)
|
24 |
-
assert MODEL_ID in [MODEL_ID_OPT_6_7B, MODEL_ID_FLAN_T5_XXL]
|
25 |
|
26 |
if torch.cuda.is_available():
|
27 |
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
@@ -31,10 +33,14 @@ if torch.cuda.is_available():
|
|
31 |
@spaces.GPU
|
32 |
def generate_caption(
|
33 |
image: PIL.Image.Image,
|
34 |
-
decoding_method: str,
|
35 |
-
temperature: float,
|
36 |
-
length_penalty: float,
|
37 |
-
repetition_penalty: float,
|
|
|
|
|
|
|
|
|
38 |
) -> str:
|
39 |
inputs = processor(images=image, return_tensors="pt").to(device, torch.float16)
|
40 |
generated_ids = model.generate(
|
@@ -43,10 +49,10 @@ def generate_caption(
|
|
43 |
temperature=temperature,
|
44 |
length_penalty=length_penalty,
|
45 |
repetition_penalty=repetition_penalty,
|
46 |
-
max_length=
|
47 |
-
min_length=
|
48 |
-
num_beams=
|
49 |
-
top_p=
|
50 |
)
|
51 |
result = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
52 |
return result
|
@@ -55,23 +61,27 @@ def generate_caption(
|
|
55 |
@spaces.GPU
|
56 |
def answer_question(
|
57 |
image: PIL.Image.Image,
|
58 |
-
|
59 |
-
decoding_method: str,
|
60 |
-
temperature: float,
|
61 |
-
length_penalty: float,
|
62 |
-
repetition_penalty: float,
|
|
|
|
|
|
|
|
|
63 |
) -> str:
|
64 |
-
inputs = processor(images=image, text=
|
65 |
generated_ids = model.generate(
|
66 |
**inputs,
|
67 |
do_sample=decoding_method == "Nucleus sampling",
|
68 |
temperature=temperature,
|
69 |
length_penalty=length_penalty,
|
70 |
repetition_penalty=repetition_penalty,
|
71 |
-
max_length=
|
72 |
-
min_length=
|
73 |
-
num_beams=
|
74 |
-
top_p=
|
75 |
)
|
76 |
result = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
77 |
return result
|
@@ -86,10 +96,14 @@ def postprocess_output(output: str) -> str:
|
|
86 |
def chat(
|
87 |
image: PIL.Image.Image,
|
88 |
text: str,
|
89 |
-
decoding_method: str,
|
90 |
-
temperature: float,
|
91 |
-
length_penalty: float,
|
92 |
-
repetition_penalty: float,
|
|
|
|
|
|
|
|
|
93 |
history_orig: list[str] = [],
|
94 |
history_qa: list[str] = [],
|
95 |
) -> tuple[list[tuple[str, str]], list[str], list[str]]:
|
@@ -99,12 +113,16 @@ def chat(
|
|
99 |
prompt = " ".join(history_qa)
|
100 |
|
101 |
output = answer_question(
|
102 |
-
image,
|
103 |
-
prompt,
|
104 |
-
decoding_method,
|
105 |
-
temperature,
|
106 |
-
length_penalty,
|
107 |
-
repetition_penalty,
|
|
|
|
|
|
|
|
|
108 |
)
|
109 |
output = postprocess_output(output)
|
110 |
history_orig.append(output)
|
@@ -160,7 +178,7 @@ with gr.Blocks(css="style.css") as demo:
|
|
160 |
clear_chat_button = gr.Button("Clear")
|
161 |
chat_button = gr.Button("Submit", variant="primary")
|
162 |
with gr.Accordion(label="Advanced settings", open=False):
|
163 |
-
|
164 |
label="Text Decoding Method",
|
165 |
choices=["Beam search", "Nucleus sampling"],
|
166 |
value="Nucleus sampling",
|
@@ -170,24 +188,53 @@ with gr.Blocks(css="style.css") as demo:
|
|
170 |
info="Used with nucleus sampling.",
|
171 |
minimum=0.5,
|
172 |
maximum=1.0,
|
173 |
-
value=1.0,
|
174 |
step=0.1,
|
|
|
175 |
)
|
176 |
length_penalty = gr.Slider(
|
177 |
label="Length Penalty",
|
178 |
info="Set to larger for longer sequence, used with beam search.",
|
179 |
minimum=-1.0,
|
180 |
maximum=2.0,
|
181 |
-
value=1.0,
|
182 |
step=0.2,
|
|
|
183 |
)
|
184 |
-
|
185 |
-
label="
|
186 |
info="Larger value prevents repetition.",
|
187 |
minimum=1.0,
|
188 |
maximum=5.0,
|
189 |
-
value=1.5,
|
190 |
step=0.5,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
191 |
)
|
192 |
|
193 |
gr.Examples(
|
@@ -199,10 +246,14 @@ with gr.Blocks(css="style.css") as demo:
|
|
199 |
fn=generate_caption,
|
200 |
inputs=[
|
201 |
image,
|
202 |
-
|
203 |
temperature,
|
204 |
length_penalty,
|
205 |
-
|
|
|
|
|
|
|
|
|
206 |
],
|
207 |
outputs=caption_output,
|
208 |
api_name="caption",
|
@@ -211,10 +262,14 @@ with gr.Blocks(css="style.css") as demo:
|
|
211 |
chat_inputs = [
|
212 |
image,
|
213 |
vqa_input,
|
214 |
-
|
215 |
temperature,
|
216 |
length_penalty,
|
217 |
-
|
|
|
|
|
|
|
|
|
218 |
history_orig,
|
219 |
history_qa,
|
220 |
]
|
|
|
18 |
|
19 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
20 |
|
21 |
+
MODEL_ID_OPT_2_7B = "Salesforce/blip2-opt-2.7b"
|
22 |
MODEL_ID_OPT_6_7B = "Salesforce/blip2-opt-6.7b"
|
23 |
+
MODEL_ID_FLAN_T5_XL = "Salesforce/blip2-flan-t5-xl"
|
24 |
MODEL_ID_FLAN_T5_XXL = "Salesforce/blip2-flan-t5-xxl"
|
25 |
MODEL_ID = os.getenv("MODEL_ID", MODEL_ID_FLAN_T5_XXL)
|
26 |
+
assert MODEL_ID in [MODEL_ID_OPT_2_7B, MODEL_ID_OPT_6_7B, MODEL_ID_FLAN_T5_XL, MODEL_ID_FLAN_T5_XXL]
|
27 |
|
28 |
if torch.cuda.is_available():
|
29 |
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
|
|
33 |
@spaces.GPU
|
34 |
def generate_caption(
|
35 |
image: PIL.Image.Image,
|
36 |
+
decoding_method: str = "Nucleus sampling",
|
37 |
+
temperature: float = 1.0,
|
38 |
+
length_penalty: float = 1.0,
|
39 |
+
repetition_penalty: float = 1.5,
|
40 |
+
max_length: int = 50,
|
41 |
+
min_length: int = 1,
|
42 |
+
num_beams: int = 5,
|
43 |
+
top_p: float = 0.9,
|
44 |
) -> str:
|
45 |
inputs = processor(images=image, return_tensors="pt").to(device, torch.float16)
|
46 |
generated_ids = model.generate(
|
|
|
49 |
temperature=temperature,
|
50 |
length_penalty=length_penalty,
|
51 |
repetition_penalty=repetition_penalty,
|
52 |
+
max_length=max_length,
|
53 |
+
min_length=min_length,
|
54 |
+
num_beams=num_beams,
|
55 |
+
top_p=top_p,
|
56 |
)
|
57 |
result = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
58 |
return result
|
|
|
61 |
@spaces.GPU
|
62 |
def answer_question(
|
63 |
image: PIL.Image.Image,
|
64 |
+
prompt: str,
|
65 |
+
decoding_method: str = "Nucleus sampling",
|
66 |
+
temperature: float = 1.0,
|
67 |
+
length_penalty: float = 1.0,
|
68 |
+
repetition_penalty: float = 1.5,
|
69 |
+
max_length: int = 50,
|
70 |
+
min_length: int = 1,
|
71 |
+
num_beams: int = 5,
|
72 |
+
top_p: float = 0.9,
|
73 |
) -> str:
|
74 |
+
inputs = processor(images=image, text=prompt, return_tensors="pt").to(device, torch.float16)
|
75 |
generated_ids = model.generate(
|
76 |
**inputs,
|
77 |
do_sample=decoding_method == "Nucleus sampling",
|
78 |
temperature=temperature,
|
79 |
length_penalty=length_penalty,
|
80 |
repetition_penalty=repetition_penalty,
|
81 |
+
max_length=max_length,
|
82 |
+
min_length=min_length,
|
83 |
+
num_beams=num_beams,
|
84 |
+
top_p=top_p,
|
85 |
)
|
86 |
result = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
87 |
return result
|
|
|
96 |
def chat(
|
97 |
image: PIL.Image.Image,
|
98 |
text: str,
|
99 |
+
decoding_method: str = "Nucleus sampling",
|
100 |
+
temperature: float = 1.0,
|
101 |
+
length_penalty: float = 1.0,
|
102 |
+
repetition_penalty: float = 1.5,
|
103 |
+
max_length: int = 50,
|
104 |
+
min_length: int = 1,
|
105 |
+
num_beams: int = 5,
|
106 |
+
top_p: float = 0.9,
|
107 |
history_orig: list[str] = [],
|
108 |
history_qa: list[str] = [],
|
109 |
) -> tuple[list[tuple[str, str]], list[str], list[str]]:
|
|
|
113 |
prompt = " ".join(history_qa)
|
114 |
|
115 |
output = answer_question(
|
116 |
+
image=image,
|
117 |
+
prompt=prompt,
|
118 |
+
decoding_method=decoding_method,
|
119 |
+
temperature=temperature,
|
120 |
+
length_penalty=length_penalty,
|
121 |
+
repetition_penalty=repetition_penalty,
|
122 |
+
max_length=max_length,
|
123 |
+
min_length=min_length,
|
124 |
+
num_beams=num_beams,
|
125 |
+
top_p=top_p,
|
126 |
)
|
127 |
output = postprocess_output(output)
|
128 |
history_orig.append(output)
|
|
|
178 |
clear_chat_button = gr.Button("Clear")
|
179 |
chat_button = gr.Button("Submit", variant="primary")
|
180 |
with gr.Accordion(label="Advanced settings", open=False):
|
181 |
+
text_decoding_method = gr.Radio(
|
182 |
label="Text Decoding Method",
|
183 |
choices=["Beam search", "Nucleus sampling"],
|
184 |
value="Nucleus sampling",
|
|
|
188 |
info="Used with nucleus sampling.",
|
189 |
minimum=0.5,
|
190 |
maximum=1.0,
|
|
|
191 |
step=0.1,
|
192 |
+
value=1.0,
|
193 |
)
|
194 |
length_penalty = gr.Slider(
|
195 |
label="Length Penalty",
|
196 |
info="Set to larger for longer sequence, used with beam search.",
|
197 |
minimum=-1.0,
|
198 |
maximum=2.0,
|
|
|
199 |
step=0.2,
|
200 |
+
value=1.0,
|
201 |
)
|
202 |
+
repetition_penalty = gr.Slider(
|
203 |
+
label="Repetition Penalty",
|
204 |
info="Larger value prevents repetition.",
|
205 |
minimum=1.0,
|
206 |
maximum=5.0,
|
|
|
207 |
step=0.5,
|
208 |
+
value=1.5,
|
209 |
+
)
|
210 |
+
max_length = gr.Slider(
|
211 |
+
label="Max Length",
|
212 |
+
minimum=1,
|
213 |
+
maximum=512,
|
214 |
+
step=1,
|
215 |
+
value=50,
|
216 |
+
)
|
217 |
+
min_length = gr.Slider(
|
218 |
+
label="Minimum Length",
|
219 |
+
minimum=1,
|
220 |
+
maximum=100,
|
221 |
+
step=1,
|
222 |
+
value=1,
|
223 |
+
)
|
224 |
+
num_beams = gr.Slider(
|
225 |
+
label="Number of Beams",
|
226 |
+
minimum=1,
|
227 |
+
maximum=10,
|
228 |
+
step=1,
|
229 |
+
value=5,
|
230 |
+
)
|
231 |
+
top_p = gr.Slider(
|
232 |
+
label="Top P",
|
233 |
+
info="Used with nucleus sampling.",
|
234 |
+
minimum=0.5,
|
235 |
+
maximum=1.0,
|
236 |
+
step=0.1,
|
237 |
+
value=0.9,
|
238 |
)
|
239 |
|
240 |
gr.Examples(
|
|
|
246 |
fn=generate_caption,
|
247 |
inputs=[
|
248 |
image,
|
249 |
+
text_decoding_method,
|
250 |
temperature,
|
251 |
length_penalty,
|
252 |
+
repetition_penalty,
|
253 |
+
max_length,
|
254 |
+
min_length,
|
255 |
+
num_beams,
|
256 |
+
top_p,
|
257 |
],
|
258 |
outputs=caption_output,
|
259 |
api_name="caption",
|
|
|
262 |
chat_inputs = [
|
263 |
image,
|
264 |
vqa_input,
|
265 |
+
text_decoding_method,
|
266 |
temperature,
|
267 |
length_penalty,
|
268 |
+
repetition_penalty,
|
269 |
+
max_length,
|
270 |
+
min_length,
|
271 |
+
num_beams,
|
272 |
+
top_p,
|
273 |
history_orig,
|
274 |
history_qa,
|
275 |
]
|