Spaces:
Runtime error
Runtime error
Upload folder using huggingface_hub
Browse files- models/model_loader.py +2 -2
models/model_loader.py
CHANGED
|
@@ -19,7 +19,7 @@ def load_embed_model(model_path: str = "nvidia/llama-nemotron-embed-vl-1b-v2"):
|
|
| 19 |
model_path,
|
| 20 |
config=config,
|
| 21 |
torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
|
| 22 |
-
trust_remote_code=
|
| 23 |
low_cpu_mem_usage=True, # ✅ CPU optimization
|
| 24 |
).to(device).eval()
|
| 25 |
|
|
@@ -44,7 +44,7 @@ def load_rerank_model(model_path: str = "nvidia/llama-nemotron-rerank-vl-1b-v2")
|
|
| 44 |
|
| 45 |
processor = AutoProcessor.from_pretrained(
|
| 46 |
model_path,
|
| 47 |
-
trust_remote_code=
|
| 48 |
max_input_tiles=6,
|
| 49 |
use_thumbnail=True,
|
| 50 |
rerank_max_length=2048
|
|
|
|
| 19 |
model_path,
|
| 20 |
config=config,
|
| 21 |
torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
|
| 22 |
+
trust_remote_code=True,
|
| 23 |
low_cpu_mem_usage=True, # ✅ CPU optimization
|
| 24 |
).to(device).eval()
|
| 25 |
|
|
|
|
| 44 |
|
| 45 |
processor = AutoProcessor.from_pretrained(
|
| 46 |
model_path,
|
| 47 |
+
trust_remote_code=True,
|
| 48 |
max_input_tiles=6,
|
| 49 |
use_thumbnail=True,
|
| 50 |
rerank_max_length=2048
|