Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -40,15 +40,6 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
| 40 |
low_cpu_mem_usage=True,
|
| 41 |
device_map="auto")
|
| 42 |
|
| 43 |
-
pipeline = transformers.pipeline(
|
| 44 |
-
"text-generation",
|
| 45 |
-
model=model_id,
|
| 46 |
-
model_kwargs={"torch_dtype": torch.bfloat16},
|
| 47 |
-
device_map="auto",
|
| 48 |
-
)
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
@spaces.GPU()
|
| 53 |
def stream_chat(
|
| 54 |
message: str,
|
|
|
|
| 40 |
low_cpu_mem_usage=True,
|
| 41 |
device_map="auto")
|
| 42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
@spaces.GPU()
|
| 44 |
def stream_chat(
|
| 45 |
message: str,
|