Update modeling_gpt2vision.py
Browse files- modeling_gpt2vision.py +0 -2
modeling_gpt2vision.py
CHANGED
|
@@ -67,7 +67,6 @@ class GPT2Vision(PreTrainedModel):
|
|
| 67 |
return_tensors="pt",
|
| 68 |
).to(device)
|
| 69 |
|
| 70 |
-
print("text_inputs",text_inputs)
|
| 71 |
# Adjust attention mask to account for image tokens and the extra <image> token
|
| 72 |
batch_size = text_inputs.input_ids.shape[0]
|
| 73 |
img_attention = torch.ones((batch_size, self.img_tokens + 1), dtype=torch.long, device=device)
|
|
@@ -100,7 +99,6 @@ class GPT2Vision(PreTrainedModel):
|
|
| 100 |
|
| 101 |
def generate(self, question, image, max_new_tokens=30, **kwargs):
|
| 102 |
prompt = f"\n\nQuestion:<image>{question}\n\nAnswer:"
|
| 103 |
-
print("prompt",prompt)
|
| 104 |
batch = {"image": [image], "text": prompt}
|
| 105 |
encoded_batch = self.tokenize_encode(batch, self.device)
|
| 106 |
inputs_embeds, attention_mask = self.preprocess_inputs(encoded_batch)
|
|
|
|
| 67 |
return_tensors="pt",
|
| 68 |
).to(device)
|
| 69 |
|
|
|
|
| 70 |
# Adjust attention mask to account for image tokens and the extra <image> token
|
| 71 |
batch_size = text_inputs.input_ids.shape[0]
|
| 72 |
img_attention = torch.ones((batch_size, self.img_tokens + 1), dtype=torch.long, device=device)
|
|
|
|
| 99 |
|
| 100 |
def generate(self, question, image, max_new_tokens=30, **kwargs):
|
| 101 |
prompt = f"\n\nQuestion:<image>{question}\n\nAnswer:"
|
|
|
|
| 102 |
batch = {"image": [image], "text": prompt}
|
| 103 |
encoded_batch = self.tokenize_encode(batch, self.device)
|
| 104 |
inputs_embeds, attention_mask = self.preprocess_inputs(encoded_batch)
|