Spaces:
Sleeping
Sleeping
Update my_model/KBVQA.py
Browse files- my_model/KBVQA.py +240 -240
my_model/KBVQA.py
CHANGED
@@ -112,248 +112,248 @@ class KBVQA:
|
|
112 |
self.current_prompt_length = None
|
113 |
|
114 |
|
115 |
-
def create_bnb_config(self) -> BitsAndBytesConfig:
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
def load_caption_model(self) -> None:
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
def get_caption(self, img: Image.Image) -> str:
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
def load_detector(self, model: str) -> None:
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
def detect_objects(self, img: Image.Image) -> Tuple[Image.Image, str]:
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
def load_fine_tuned_model(self) -> None:
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
228 |
-
@property
|
229 |
-
def all_models_loaded(self) -> bool:
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
def format_prompt(self, current_query: str, history: Optional[str] = None, sys_prompt: Optional[str] = None,
|
241 |
-
|
242 |
-
|
243 |
-
|
244 |
-
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
|
277 |
-
|
278 |
-
|
279 |
-
|
280 |
-
|
281 |
-
|
|
|
|
|
282 |
else:
|
283 |
-
p = f"""{B_SENT}{B_INST} {
|
284 |
-
|
285 |
-
p
|
286 |
-
|
287 |
-
|
288 |
-
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
|
309 |
-
|
310 |
-
|
311 |
-
|
312 |
-
|
313 |
-
|
314 |
-
|
315 |
-
|
316 |
-
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
|
321 |
-
|
322 |
-
free_gpu_resources()
|
323 |
-
prompt = self.format_prompt(question, caption=caption, objects=detected_objects_str)
|
324 |
-
num_tokens = len(self.kbvqa_tokenizer.tokenize(prompt))
|
325 |
-
self.current_prompt_length = num_tokens
|
326 |
-
trim = False # flag used to check if prompt trim is required or no.
|
327 |
-
# max_context_window is set to 4,000 tokens, refer to the config file.
|
328 |
-
if self.current_prompt_length > self.max_context_window:
|
329 |
-
trim = True
|
330 |
-
st.warning(
|
331 |
-
f"Prompt length is {self.current_prompt_length} which is larger than the maximum context window of LLaMA-2,"
|
332 |
-
f" objects detected with low confidence will be removed one at a time until the prompt length is within the"
|
333 |
-
f" maximum context window ...")
|
334 |
-
# an object is trimmed from the bottom of the list until the overall prompt length is within the context window.
|
335 |
-
while self.current_prompt_length > self.max_context_window:
|
336 |
-
detected_objects_str = self.trim_objects(detected_objects_str)
|
337 |
prompt = self.format_prompt(question, caption=caption, objects=detected_objects_str)
|
338 |
-
|
339 |
-
|
340 |
-
if
|
341 |
-
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
-
|
346 |
-
|
347 |
-
|
348 |
-
|
349 |
-
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
357 |
|
358 |
def prepare_kbvqa_model(only_reload_detection_model: bool = False, force_reload: bool = False) -> KBVQA:
|
359 |
"""
|
|
|
112 |
self.current_prompt_length = None
|
113 |
|
114 |
|
115 |
+
def create_bnb_config(self) -> BitsAndBytesConfig:
|
116 |
+
"""
|
117 |
+
Creates a BitsAndBytes configuration based on the quantization setting.
|
118 |
+
Returns:
|
119 |
+
BitsAndBytesConfig: Configuration for BitsAndBytes optimized model.
|
120 |
+
"""
|
121 |
+
|
122 |
+
if self.quantization == '4bit':
|
123 |
+
return BitsAndBytesConfig(
|
124 |
+
load_in_4bit=True,
|
125 |
+
bnb_4bit_use_double_quant=True,
|
126 |
+
bnb_4bit_quant_type="nf4",
|
127 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
128 |
+
)
|
129 |
+
elif self.quantization == '8bit':
|
130 |
+
return BitsAndBytesConfig(
|
131 |
+
load_in_8bit=True,
|
132 |
+
bnb_8bit_use_double_quant=True,
|
133 |
+
bnb_8bit_quant_type="nf4",
|
134 |
+
bnb_8bit_compute_dtype=torch.bfloat16
|
135 |
+
)
|
136 |
+
|
137 |
+
|
138 |
+
def load_caption_model(self) -> None:
|
139 |
+
"""
|
140 |
+
Loads the image captioning model into the KBVQA instance.
|
141 |
+
|
142 |
+
Returns:
|
143 |
+
None
|
144 |
+
"""
|
145 |
+
|
146 |
+
self.captioner = ImageCaptioningModel()
|
147 |
+
self.captioner.load_model()
|
148 |
+
free_gpu_resources()
|
149 |
+
|
150 |
+
|
151 |
+
def get_caption(self, img: Image.Image) -> str:
|
152 |
+
"""
|
153 |
+
Generates a caption for a given image using the image captioning model.
|
154 |
+
|
155 |
+
Args:
|
156 |
+
img (PIL.Image.Image): The image for which to generate a caption.
|
157 |
+
|
158 |
+
Returns:
|
159 |
+
str: The generated caption for the image.
|
160 |
+
"""
|
161 |
+
caption = self.captioner.generate_caption(img)
|
162 |
+
free_gpu_resources()
|
163 |
+
return caption
|
164 |
+
|
165 |
+
|
166 |
+
def load_detector(self, model: str) -> None:
|
167 |
+
"""
|
168 |
+
Loads the object detection model.
|
169 |
+
|
170 |
+
Args:
|
171 |
+
model (str): The name of the object detection model to load.
|
172 |
+
|
173 |
+
Returns:
|
174 |
+
None
|
175 |
+
"""
|
176 |
+
|
177 |
+
self.detector = ObjectDetector()
|
178 |
+
self.detector.load_model(model)
|
179 |
+
free_gpu_resources()
|
180 |
+
|
181 |
+
|
182 |
+
def detect_objects(self, img: Image.Image) -> Tuple[Image.Image, str]:
|
183 |
+
"""
|
184 |
+
Detects objects in a given image using the loaded object detection model.
|
185 |
+
|
186 |
+
Args:
|
187 |
+
img (PIL.Image.Image): The image in which to detect objects.
|
188 |
+
|
189 |
+
Returns:
|
190 |
+
tuple: A tuple containing the image with detected objects drawn and a string representation of detected objects.
|
191 |
+
"""
|
192 |
+
|
193 |
+
image = self.detector.process_image(img)
|
194 |
+
free_gpu_resources()
|
195 |
+
detected_objects_string, detected_objects_list = self.detector.detect_objects(image, threshold=st.session_state[
|
196 |
+
'confidence_level'])
|
197 |
+
free_gpu_resources()
|
198 |
+
image_with_boxes = self.detector.draw_boxes(img, detected_objects_list)
|
199 |
+
free_gpu_resources()
|
200 |
+
return image_with_boxes, detected_objects_string
|
201 |
+
|
202 |
+
|
203 |
+
def load_fine_tuned_model(self) -> None:
|
204 |
+
"""
|
205 |
+
Loads the fine-tuned KBVQA model along with its tokenizer.
|
206 |
+
|
207 |
+
Returns:
|
208 |
+
None
|
209 |
+
"""
|
210 |
+
|
211 |
+
self.kbvqa_model = AutoModelForCausalLM.from_pretrained(self.kbvqa_model_name,
|
212 |
+
device_map="auto",
|
213 |
+
low_cpu_mem_usage=True,
|
214 |
+
quantization_config=self.bnb_config,
|
215 |
+
token=self.access_token)
|
216 |
+
|
217 |
+
free_gpu_resources()
|
218 |
+
|
219 |
+
self.kbvqa_tokenizer = AutoTokenizer.from_pretrained(self.kbvqa_model_name,
|
220 |
+
use_fast=self.use_fast,
|
221 |
+
low_cpu_mem_usage=True,
|
222 |
+
trust_remote_code=self.trust_remote,
|
223 |
+
add_eos_token=self.add_eos_token,
|
224 |
+
token=self.access_token)
|
225 |
+
free_gpu_resources()
|
226 |
+
|
227 |
+
|
228 |
+
@property
|
229 |
+
def all_models_loaded(self) -> bool:
|
230 |
+
"""
|
231 |
+
Checks if all the required models (KBVQA, captioner, detector) are loaded.
|
232 |
+
|
233 |
+
Returns:
|
234 |
+
bool: True if all models are loaded, False otherwise.
|
235 |
+
"""
|
236 |
+
|
237 |
+
return self.kbvqa_model is not None and self.captioner is not None and self.detector is not None
|
238 |
+
|
239 |
+
|
240 |
+
def format_prompt(self, current_query: str, history: Optional[str] = None, sys_prompt: Optional[str] = None,
|
241 |
+
caption: str = None, objects: Optional[str] = None) -> str:
|
242 |
+
"""
|
243 |
+
Formats the prompt for the KBVQA model based on the provided parameters.
|
244 |
+
|
245 |
+
This implements the Prompt Engineering Module of the Overall KB-VQA Archetecture.
|
246 |
+
|
247 |
+
Args:
|
248 |
+
current_query (str): The current question to be answered.
|
249 |
+
history (str, optional): The history of previous interactions.
|
250 |
+
sys_prompt (str, optional): The system prompt or instructions for the model.
|
251 |
+
caption (str, optional): The caption of the image.
|
252 |
+
objects (str, optional): The detected objects in the image.
|
253 |
+
|
254 |
+
Returns:
|
255 |
+
str: The formatted prompt for the KBVQA model.
|
256 |
+
"""
|
257 |
+
|
258 |
+
# These are the special tokens designed for the model to be fine-tuned on.
|
259 |
+
B_CAP = '[CAP]'
|
260 |
+
E_CAP = '[/CAP]'
|
261 |
+
B_QES = '[QES]'
|
262 |
+
E_QES = '[/QES]'
|
263 |
+
B_OBJ = '[OBJ]'
|
264 |
+
E_OBJ = '[/OBJ]'
|
265 |
+
|
266 |
+
# These are the default special tokens of LLaMA-2 Chat Model.
|
267 |
+
B_SENT = '<s>'
|
268 |
+
E_SENT = '</s>'
|
269 |
+
B_INST = '[INST]'
|
270 |
+
E_INST = '[/INST]'
|
271 |
+
B_SYS = '<<SYS>>\n'
|
272 |
+
E_SYS = '\n<</SYS>>\n\n'
|
273 |
+
|
274 |
+
current_query = current_query.strip()
|
275 |
+
if sys_prompt is None:
|
276 |
+
sys_prompt = config.SYSTEM_PROMPT.strip()
|
277 |
+
|
278 |
+
# History can be used to facilitate multi turn chat, not used for the Run Inference tool within the demo app.
|
279 |
+
if history is None:
|
280 |
+
if objects is None:
|
281 |
+
p = f"""{B_SENT}{B_INST} {B_SYS}{sys_prompt}{E_SYS}{B_CAP}{caption}{E_CAP}{B_QES}{current_query}{E_QES}{E_INST}"""
|
282 |
+
else:
|
283 |
+
p = f"""{B_SENT}{B_INST} {B_SYS}{sys_prompt}{E_SYS}{B_CAP}{caption}{E_CAP}{B_OBJ}{objects}{E_OBJ}{B_QES}taking into consideration the objects with high certainty, {current_query}{E_QES}{E_INST}"""
|
284 |
else:
|
285 |
+
p = f"""{history}\n{B_SENT}{B_INST} {B_QES}{current_query}{E_QES}{E_INST}"""
|
286 |
+
|
287 |
+
return p
|
288 |
+
|
289 |
+
|
290 |
+
@staticmethod
|
291 |
+
def trim_objects(detected_objects_str: str) -> str:
|
292 |
+
"""
|
293 |
+
Trim the last object from the detected objects string.
|
294 |
+
This is implemented to ensure that the prompt length is within the context window, threshold set to 4,000 tokens.
|
295 |
+
|
296 |
+
Args:
|
297 |
+
detected_objects_str (str): String containing detected objects.
|
298 |
+
|
299 |
+
Returns:
|
300 |
+
str: The string with the last object removed.
|
301 |
+
"""
|
302 |
+
|
303 |
+
objects = detected_objects_str.strip().split("\n")
|
304 |
+
if len(objects) >= 1:
|
305 |
+
return "\n".join(objects[:-1])
|
306 |
+
return ""
|
307 |
+
|
308 |
+
|
309 |
+
def generate_answer(self, question: str, caption: str, detected_objects_str: str) -> str:
|
310 |
+
"""
|
311 |
+
Generates an answer to a given question using the KBVQA model.
|
312 |
+
|
313 |
+
Args:
|
314 |
+
question (str): The question to be answered.
|
315 |
+
caption (str): The caption of the image related to the question.
|
316 |
+
detected_objects_str (str): The string representation of detected objects in the image.
|
317 |
+
|
318 |
+
Returns:
|
319 |
+
str: The generated answer to the question.
|
320 |
+
"""
|
321 |
+
|
322 |
+
free_gpu_resources()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
323 |
prompt = self.format_prompt(question, caption=caption, objects=detected_objects_str)
|
324 |
+
num_tokens = len(self.kbvqa_tokenizer.tokenize(prompt))
|
325 |
+
self.current_prompt_length = num_tokens
|
326 |
+
trim = False # flag used to check if prompt trim is required or no.
|
327 |
+
# max_context_window is set to 4,000 tokens, refer to the config file.
|
328 |
+
if self.current_prompt_length > self.max_context_window:
|
329 |
+
trim = True
|
330 |
+
st.warning(
|
331 |
+
f"Prompt length is {self.current_prompt_length} which is larger than the maximum context window of LLaMA-2,"
|
332 |
+
f" objects detected with low confidence will be removed one at a time until the prompt length is within the"
|
333 |
+
f" maximum context window ...")
|
334 |
+
# an object is trimmed from the bottom of the list until the overall prompt length is within the context window.
|
335 |
+
while self.current_prompt_length > self.max_context_window:
|
336 |
+
detected_objects_str = self.trim_objects(detected_objects_str)
|
337 |
+
prompt = self.format_prompt(question, caption=caption, objects=detected_objects_str)
|
338 |
+
self.current_prompt_length = len(self.kbvqa_tokenizer.tokenize(prompt))
|
339 |
+
|
340 |
+
if detected_objects_str == "":
|
341 |
+
break # Break if no objects are left
|
342 |
+
if trim:
|
343 |
+
st.warning(f"New prompt length is: {self.current_prompt_length}")
|
344 |
+
trim = False
|
345 |
+
|
346 |
+
model_inputs = self.kbvqa_tokenizer(prompt, add_special_tokens=False, return_tensors="pt").to('cuda')
|
347 |
+
free_gpu_resources()
|
348 |
+
input_ids = model_inputs["input_ids"]
|
349 |
+
output_ids = self.kbvqa_model.generate(input_ids)
|
350 |
+
free_gpu_resources()
|
351 |
+
index = input_ids.shape[1] # needed to avoid printing the input prompt
|
352 |
+
history = self.kbvqa_tokenizer.decode(output_ids[0], skip_special_tokens=False)
|
353 |
+
output_text = self.kbvqa_tokenizer.decode(output_ids[0][index:], skip_special_tokens=True)
|
354 |
+
|
355 |
+
return output_text.capitalize()
|
356 |
+
|
357 |
|
358 |
def prepare_kbvqa_model(only_reload_detection_model: bool = False, force_reload: bool = False) -> KBVQA:
|
359 |
"""
|