Update app.py
Browse files
app.py
CHANGED
|
@@ -31,11 +31,6 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
| 31 |
model.config.sliding_window = 4096
|
| 32 |
model.eval()
|
| 33 |
|
| 34 |
-
terminators = [
|
| 35 |
-
pipeline.tokenizer.eos_token_id,
|
| 36 |
-
pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
|
| 37 |
-
]
|
| 38 |
-
|
| 39 |
@spaces.GPU(duration=90)
|
| 40 |
def generate(
|
| 41 |
message: str,
|
|
@@ -66,7 +61,6 @@ def generate(
|
|
| 66 |
temperature=temperature,
|
| 67 |
num_beams=1,
|
| 68 |
repetition_penalty=repetition_penalty,
|
| 69 |
-
eos_token_id=terminators,
|
| 70 |
)
|
| 71 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
| 72 |
t.start()
|
|
|
|
| 31 |
model.config.sliding_window = 4096
|
| 32 |
model.eval()
|
| 33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
@spaces.GPU(duration=90)
|
| 35 |
def generate(
|
| 36 |
message: str,
|
|
|
|
| 61 |
temperature=temperature,
|
| 62 |
num_beams=1,
|
| 63 |
repetition_penalty=repetition_penalty,
|
|
|
|
| 64 |
)
|
| 65 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
| 66 |
t.start()
|