Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -22,7 +22,6 @@ def load_model():
|
|
22 |
|
23 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
|
24 |
|
25 |
-
# Create pipeline
|
26 |
pipe = pipeline(
|
27 |
"text-generation",
|
28 |
model=model,
|
@@ -36,24 +35,38 @@ pipe = load_model()
|
|
36 |
|
37 |
@spaces.GPU(duration=110)
|
38 |
def generate_response(prompt, max_length=1024):
|
39 |
-
#
|
40 |
messages = [
|
41 |
-
{"role": "system", "content": "You are a helpful AI
|
42 |
{"role": "user", "content": prompt}
|
43 |
]
|
44 |
|
45 |
# Generate response using pipeline
|
46 |
outputs = pipe(messages, max_new_tokens=max_length)
|
47 |
|
48 |
-
# Extract the generated text
|
49 |
response = outputs[0]["generated_text"]
|
50 |
|
51 |
-
#
|
52 |
-
|
53 |
-
|
|
|
|
|
|
|
54 |
|
55 |
return response_only
|
56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
demo = gr.Interface(
|
58 |
fn=generate_response,
|
59 |
inputs=[
|
@@ -69,7 +82,7 @@ demo = gr.Interface(
|
|
69 |
|
70 |
Model: [benhaotang/phi4-qwq-sky-t1]({MODEL_URL})""",
|
71 |
examples=[
|
72 |
-
[
|
73 |
]
|
74 |
)
|
75 |
|
|
|
22 |
|
23 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
|
24 |
|
|
|
25 |
pipe = pipeline(
|
26 |
"text-generation",
|
27 |
model=model,
|
|
|
35 |
|
36 |
@spaces.GPU(duration=110)
|
37 |
def generate_response(prompt, max_length=1024):
|
38 |
+
# Create messages with system prompt
|
39 |
messages = [
|
40 |
+
{"role": "system", "content": "You are a helpful AI assistant. You always think step by step."},
|
41 |
{"role": "user", "content": prompt}
|
42 |
]
|
43 |
|
44 |
# Generate response using pipeline
|
45 |
outputs = pipe(messages, max_new_tokens=max_length)
|
46 |
|
47 |
+
# Extract the generated text - outputs[0] is a dict with 'generated_text'
|
48 |
response = outputs[0]["generated_text"]
|
49 |
|
50 |
+
# Find the user's prompt in the response and get everything after it
|
51 |
+
try:
|
52 |
+
start_idx = response.find(prompt) + len(prompt)
|
53 |
+
response_only = response[start_idx:].strip()
|
54 |
+
except:
|
55 |
+
response_only = response # Fallback to full response if splitting fails
|
56 |
|
57 |
return response_only
|
58 |
|
59 |
+
# Example with proper line breaks
|
60 |
+
example_prompt = """For a scalar field theory with interaction Lagrangian $\mathcal{L}_{int} = g\phi^3 + \lambda\phi^4$:
|
61 |
+
|
62 |
+
1. Enumerate all possible 1-loop Feynman diagrams contributing to the scalar propagator
|
63 |
+
|
64 |
+
2. For each diagram, write down its loop contribution
|
65 |
+
|
66 |
+
3. Provide Mathematica code to calculate these loop amplitudes with dimensional regularization at $d=4-\epsilon$
|
67 |
+
|
68 |
+
Please explain your reasoning step by step."""
|
69 |
+
|
70 |
demo = gr.Interface(
|
71 |
fn=generate_response,
|
72 |
inputs=[
|
|
|
82 |
|
83 |
Model: [benhaotang/phi4-qwq-sky-t1]({MODEL_URL})""",
|
84 |
examples=[
|
85 |
+
[example_prompt] # Now using the formatted example
|
86 |
]
|
87 |
)
|
88 |
|