Spaces:
Runtime error
Runtime error
Commit
·
994c940
1
Parent(s):
1fdb555
try to avoid cuda OO error
Browse files
app.py
CHANGED
@@ -90,18 +90,18 @@ def evaluate(
|
|
90 |
):
|
91 |
content = process_webpage(url=url)
|
92 |
# avoid GPU memory overflow
|
93 |
-
torch.cuda.empty_cache()
|
94 |
-
prompt = generate_prompt(instruction, content)
|
95 |
-
inputs = tokenizer(prompt, return_tensors="pt")
|
96 |
-
input_ids = inputs["input_ids"].to(device)
|
97 |
-
generation_config = GenerationConfig(
|
98 |
-
temperature=temperature,
|
99 |
-
top_p=top_p,
|
100 |
-
top_k=top_k,
|
101 |
-
num_beams=num_beams,
|
102 |
-
**kwargs,
|
103 |
-
)
|
104 |
with torch.no_grad():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
generation_output = model.generate(
|
106 |
input_ids=input_ids,
|
107 |
generation_config=generation_config,
|
@@ -109,8 +109,8 @@ def evaluate(
|
|
109 |
output_scores=True,
|
110 |
max_new_tokens=max_new_tokens,
|
111 |
)
|
112 |
-
|
113 |
-
|
114 |
# avoid GPU memory overflow
|
115 |
torch.cuda.empty_cache()
|
116 |
return output.split("### Response:")[1].strip()
|
|
|
90 |
):
|
91 |
content = process_webpage(url=url)
|
92 |
# avoid GPU memory overflow
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
with torch.no_grad():
|
94 |
+
torch.cuda.empty_cache()
|
95 |
+
prompt = generate_prompt(instruction, content)
|
96 |
+
inputs = tokenizer(prompt, return_tensors="pt")
|
97 |
+
input_ids = inputs["input_ids"].to(device)
|
98 |
+
generation_config = GenerationConfig(
|
99 |
+
temperature=temperature,
|
100 |
+
top_p=top_p,
|
101 |
+
top_k=top_k,
|
102 |
+
num_beams=num_beams,
|
103 |
+
**kwargs,
|
104 |
+
)
|
105 |
generation_output = model.generate(
|
106 |
input_ids=input_ids,
|
107 |
generation_config=generation_config,
|
|
|
109 |
output_scores=True,
|
110 |
max_new_tokens=max_new_tokens,
|
111 |
)
|
112 |
+
s = generation_output.sequences[0]
|
113 |
+
output = tokenizer.decode(s)
|
114 |
# avoid GPU memory overflow
|
115 |
torch.cuda.empty_cache()
|
116 |
return output.split("### Response:")[1].strip()
|