mshukor
commited on
Commit
•
902be23
1
Parent(s):
78ad2cd
vqa
Browse files
app.py
CHANGED
@@ -80,13 +80,9 @@ checkpoint = torch.load(checkpoint_path, map_location='cpu')
|
|
80 |
state_dict = checkpoint['model']
|
81 |
msg = model_caption.load_state_dict(state_dict,strict=False)
|
82 |
|
83 |
-
|
84 |
-
|
85 |
|
86 |
###### VQA
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
config = 'configs/image/ePALM_vqa.yaml'
|
91 |
config = yaml.load(open(config, 'r'))
|
92 |
|
@@ -112,6 +108,28 @@ state_dict = checkpoint['model']
|
|
112 |
msg = model_vqa.load_state_dict(state_dict,strict=False)
|
113 |
|
114 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
115 |
|
116 |
|
117 |
|
@@ -148,8 +166,7 @@ num_beams=3
|
|
148 |
max_length=30
|
149 |
|
150 |
|
151 |
-
|
152 |
-
model_vqa.bfloat16()
|
153 |
|
154 |
|
155 |
def inference(image, audio, video, task_type, instruction):
|
@@ -157,11 +174,26 @@ def inference(image, audio, video, task_type, instruction):
|
|
157 |
if task_type == 'Image Captioning':
|
158 |
text = ['']
|
159 |
text_input = tokenizer(text, padding='longest', return_tensors="pt").to(device)
|
160 |
-
model = model_caption
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
161 |
elif task_type == 'Visual Question Answering':
|
162 |
question = instruction+'?'+special_answer_token
|
163 |
text_input = tokenizer(question, padding='longest', return_tensors="pt").to(device)
|
164 |
-
|
|
|
165 |
else:
|
166 |
raise NotImplemented
|
167 |
|
|
|
80 |
state_dict = checkpoint['model']
|
81 |
msg = model_caption.load_state_dict(state_dict,strict=False)
|
82 |
|
83 |
+
model_caption.bfloat16()
|
|
|
84 |
|
85 |
###### VQA
|
|
|
|
|
|
|
86 |
config = 'configs/image/ePALM_vqa.yaml'
|
87 |
config = yaml.load(open(config, 'r'))
|
88 |
|
|
|
108 |
msg = model_vqa.load_state_dict(state_dict,strict=False)
|
109 |
|
110 |
|
111 |
+
model_vqa.bfloat16()
|
112 |
+
|
113 |
+
|
114 |
+
|
115 |
+
# Video Captioning
|
116 |
+
checkpoint_path = 'checkpoints/float32/ePALM_video_caption_msrvtt/checkpoint_best.pth'
|
117 |
+
# checkpoint_path = '/data/mshukor/logs/eplam/models/accelerate/ePALM_pt_L_acc_caption/checkpoint_best.pth'
|
118 |
+
checkpoint = torch.load(checkpoint_path, map_location='cpu')
|
119 |
+
state_dict_video_caption = checkpoint['model']
|
120 |
+
|
121 |
+
# Video QA
|
122 |
+
checkpoint_path = 'checkpoints/float32/ePALM_video_qa_msrvtt/checkpoint_best.pth'
|
123 |
+
# checkpoint_path = '/data/mshukor/logs/eplam/models/accelerate/ePALM_pt_L_acc_caption/checkpoint_best.pth'
|
124 |
+
checkpoint = torch.load(checkpoint_path, map_location='cpu')
|
125 |
+
state_dict_video_qa = checkpoint['model']
|
126 |
+
|
127 |
+
|
128 |
+
# Audio Captioning
|
129 |
+
checkpoint_path = 'checkpoints/float32/ePALM_audio_caption/checkpoint_best.pth'
|
130 |
+
# checkpoint_path = '/data/mshukor/logs/eplam/models/accelerate/ePALM_pt_L_acc_caption/checkpoint_best.pth'
|
131 |
+
checkpoint = torch.load(checkpoint_path, map_location='cpu')
|
132 |
+
state_dict_audio_caption = checkpoint['model']
|
133 |
|
134 |
|
135 |
|
|
|
166 |
max_length=30
|
167 |
|
168 |
|
169 |
+
|
|
|
170 |
|
171 |
|
172 |
def inference(image, audio, video, task_type, instruction):
|
|
|
174 |
if task_type == 'Image Captioning':
|
175 |
text = ['']
|
176 |
text_input = tokenizer(text, padding='longest', return_tensors="pt").to(device)
|
177 |
+
model = model_caption.clone()
|
178 |
+
elif task_type == 'Video Captioning':
|
179 |
+
text = ['']
|
180 |
+
text_input = tokenizer(text, padding='longest', return_tensors="pt").to(device)
|
181 |
+
model_caption = model_caption.load_state_dict(state_dict_video_caption,strict=False)
|
182 |
+
model = model_caption.clone()
|
183 |
+
elif task_type == 'Audio Captioning':
|
184 |
+
text = ['']
|
185 |
+
text_input = tokenizer(text, padding='longest', return_tensors="pt").to(device)
|
186 |
+
model_caption = model_caption.load_state_dict(state_dict_audio_caption,strict=False)
|
187 |
+
model = model_caption.clone()
|
188 |
+
elif task_type == 'Visual Question Answering':
|
189 |
+
question = instruction+'?'+special_answer_token
|
190 |
+
text_input = tokenizer(question, padding='longest', return_tensors="pt").to(device)
|
191 |
+
model = model_vqa.clone()
|
192 |
elif task_type == 'Visual Question Answering':
|
193 |
question = instruction+'?'+special_answer_token
|
194 |
text_input = tokenizer(question, padding='longest', return_tensors="pt").to(device)
|
195 |
+
model_vqa = model_vqa.load_state_dict(state_dict_video_qa,strict=False)
|
196 |
+
model = model_vqa.clone()
|
197 |
else:
|
198 |
raise NotImplemented
|
199 |
|