Spaces:
Running
on
Zero
Running
on
Zero
update app
Browse files
app.py
CHANGED
|
@@ -84,7 +84,6 @@ logger.info(f"Loading model 1: {MODEL_ID_1}")
|
|
| 84 |
processor_1 = AutoProcessor.from_pretrained(MODEL_ID_1, trust_remote_code=True)
|
| 85 |
model_1 = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 86 |
MODEL_ID_1,
|
| 87 |
-
attn_implementation="flash_attention_2",
|
| 88 |
trust_remote_code=True,
|
| 89 |
torch_dtype=torch.float16 if device == "cuda" else torch.float32
|
| 90 |
).to(device).eval()
|
|
@@ -96,7 +95,6 @@ logger.info(f"Loading model 2: {MODEL_ID_2}")
|
|
| 96 |
processor_2 = AutoProcessor.from_pretrained(MODEL_ID_2, trust_remote_code=True)
|
| 97 |
model_2 = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 98 |
MODEL_ID_2,
|
| 99 |
-
attn_implementation="flash_attention_2",
|
| 100 |
trust_remote_code=True,
|
| 101 |
torch_dtype=torch.float16 if device == "cuda" else torch.float32
|
| 102 |
).to(device).eval()
|
|
@@ -108,7 +106,6 @@ logger.info(f"Loading model 3: {MODEL_ID_3}")
|
|
| 108 |
processor_3 = AutoProcessor.from_pretrained(MODEL_ID_3, trust_remote_code=True)
|
| 109 |
model_3 = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 110 |
MODEL_ID_3,
|
| 111 |
-
attn_implementation="flash_attention_2",
|
| 112 |
trust_remote_code=True,
|
| 113 |
torch_dtype=torch.float16 if device == "cuda" else torch.float32
|
| 114 |
).to(device).eval()
|
|
|
|
| 84 |
processor_1 = AutoProcessor.from_pretrained(MODEL_ID_1, trust_remote_code=True)
|
| 85 |
model_1 = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 86 |
MODEL_ID_1,
|
|
|
|
| 87 |
trust_remote_code=True,
|
| 88 |
torch_dtype=torch.float16 if device == "cuda" else torch.float32
|
| 89 |
).to(device).eval()
|
|
|
|
| 95 |
processor_2 = AutoProcessor.from_pretrained(MODEL_ID_2, trust_remote_code=True)
|
| 96 |
model_2 = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 97 |
MODEL_ID_2,
|
|
|
|
| 98 |
trust_remote_code=True,
|
| 99 |
torch_dtype=torch.float16 if device == "cuda" else torch.float32
|
| 100 |
).to(device).eval()
|
|
|
|
| 106 |
processor_3 = AutoProcessor.from_pretrained(MODEL_ID_3, trust_remote_code=True)
|
| 107 |
model_3 = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 108 |
MODEL_ID_3,
|
|
|
|
| 109 |
trust_remote_code=True,
|
| 110 |
torch_dtype=torch.float16 if device == "cuda" else torch.float32
|
| 111 |
).to(device).eval()
|