Spaces:
Configuration error
Configuration error
fix gradio history format bug
Browse files
app.py
CHANGED
|
@@ -1,32 +1,52 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import pipeline
|
| 3 |
|
| 4 |
-
# Load
|
| 5 |
model = pipeline(
|
| 6 |
"text-generation",
|
| 7 |
model="microsoft/Phi-3-mini-4k-instruct",
|
| 8 |
device_map="auto"
|
| 9 |
)
|
| 10 |
|
| 11 |
-
# Build
|
| 12 |
def build_prompt(history, user_input):
|
| 13 |
prompt = ""
|
| 14 |
|
| 15 |
-
|
| 16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
prompt += f"<|user|>\n{user_input}\n<|assistant|>\n"
|
| 19 |
|
| 20 |
return prompt
|
| 21 |
|
| 22 |
|
| 23 |
-
#
|
| 24 |
def chat(message, history):
|
|
|
|
| 25 |
try:
|
| 26 |
-
# Create formatted prompt
|
| 27 |
prompt = build_prompt(history, message)
|
| 28 |
|
| 29 |
-
# Generate response
|
| 30 |
response = model(
|
| 31 |
prompt,
|
| 32 |
max_new_tokens=150,
|
|
@@ -37,10 +57,9 @@ def chat(message, history):
|
|
| 37 |
return_full_text=False
|
| 38 |
)
|
| 39 |
|
| 40 |
-
# Extract text safely
|
| 41 |
answer = response[0]["generated_text"].strip()
|
| 42 |
|
| 43 |
-
# Remove unwanted continuation
|
| 44 |
stop_tokens = [
|
| 45 |
"<|user|>",
|
| 46 |
"<|assistant|>",
|
|
@@ -52,7 +71,6 @@ def chat(message, history):
|
|
| 52 |
if token in answer:
|
| 53 |
answer = answer.split(token)[0].strip()
|
| 54 |
|
| 55 |
-
# Safety fallback
|
| 56 |
if not answer:
|
| 57 |
answer = "I could not generate a response."
|
| 58 |
|
|
@@ -62,11 +80,11 @@ def chat(message, history):
|
|
| 62 |
return f"Error: {str(e)}"
|
| 63 |
|
| 64 |
|
| 65 |
-
# Create
|
| 66 |
demo = gr.ChatInterface(
|
| 67 |
fn=chat,
|
| 68 |
title="Symbolic Alpha AI Agent",
|
| 69 |
-
description="Phi-3 Mini
|
| 70 |
)
|
| 71 |
|
| 72 |
# Launch app
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import pipeline
|
| 3 |
|
| 4 |
+
# Load model
|
| 5 |
model = pipeline(
|
| 6 |
"text-generation",
|
| 7 |
model="microsoft/Phi-3-mini-4k-instruct",
|
| 8 |
device_map="auto"
|
| 9 |
)
|
| 10 |
|
| 11 |
+
# Build conversation prompt safely
|
| 12 |
def build_prompt(history, user_input):
|
| 13 |
prompt = ""
|
| 14 |
|
| 15 |
+
try:
|
| 16 |
+
for item in history:
|
| 17 |
+
|
| 18 |
+
# Handle new Gradio message format
|
| 19 |
+
if isinstance(item, dict):
|
| 20 |
+
role = item.get("role", "")
|
| 21 |
+
content = item.get("content", "")
|
| 22 |
+
|
| 23 |
+
if role == "user":
|
| 24 |
+
prompt += f"<|user|>\n{content}\n"
|
| 25 |
+
|
| 26 |
+
elif role == "assistant":
|
| 27 |
+
prompt += f"<|assistant|>\n{content}\n"
|
| 28 |
+
|
| 29 |
+
# Handle old tuple format
|
| 30 |
+
elif isinstance(item, (list, tuple)) and len(item) == 2:
|
| 31 |
+
user_msg, bot_msg = item
|
| 32 |
+
|
| 33 |
+
prompt += f"<|user|>\n{user_msg}\n"
|
| 34 |
+
prompt += f"<|assistant|>\n{bot_msg}\n"
|
| 35 |
+
|
| 36 |
+
except Exception:
|
| 37 |
+
pass
|
| 38 |
|
| 39 |
prompt += f"<|user|>\n{user_input}\n<|assistant|>\n"
|
| 40 |
|
| 41 |
return prompt
|
| 42 |
|
| 43 |
|
| 44 |
+
# Main chatbot function
|
| 45 |
def chat(message, history):
|
| 46 |
+
|
| 47 |
try:
|
|
|
|
| 48 |
prompt = build_prompt(history, message)
|
| 49 |
|
|
|
|
| 50 |
response = model(
|
| 51 |
prompt,
|
| 52 |
max_new_tokens=150,
|
|
|
|
| 57 |
return_full_text=False
|
| 58 |
)
|
| 59 |
|
|
|
|
| 60 |
answer = response[0]["generated_text"].strip()
|
| 61 |
|
| 62 |
+
# Remove unwanted continuation
|
| 63 |
stop_tokens = [
|
| 64 |
"<|user|>",
|
| 65 |
"<|assistant|>",
|
|
|
|
| 71 |
if token in answer:
|
| 72 |
answer = answer.split(token)[0].strip()
|
| 73 |
|
|
|
|
| 74 |
if not answer:
|
| 75 |
answer = "I could not generate a response."
|
| 76 |
|
|
|
|
| 80 |
return f"Error: {str(e)}"
|
| 81 |
|
| 82 |
|
| 83 |
+
# Create UI
|
| 84 |
demo = gr.ChatInterface(
|
| 85 |
fn=chat,
|
| 86 |
title="Symbolic Alpha AI Agent",
|
| 87 |
+
description="Stable Phi-3 Mini chatbot running on Hugging Face Spaces"
|
| 88 |
)
|
| 89 |
|
| 90 |
# Launch app
|