Sharyar commited on
Commit ·
c0d3986
1
Parent(s): 3666e81
Refactor: Switch to google/flan-t5-xxl for reliability
Browse files
app.py
CHANGED
|
@@ -2,30 +2,28 @@ import gradio as gr
|
|
| 2 |
from huggingface_hub import InferenceClient
|
| 3 |
import os
|
| 4 |
|
| 5 |
-
# Use a reliable
|
| 6 |
-
MODEL_ID = "
|
| 7 |
|
| 8 |
-
# It's recommended to set your Hugging Face token as an environment variable
|
| 9 |
-
# For local testing, you can also pass it directly: client = InferenceClient(token="hf_...")
|
| 10 |
client = InferenceClient(model=MODEL_ID)
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
|
|
|
| 23 |
""".strip()
|
| 24 |
|
| 25 |
def improve_description(user_input: str) -> str:
|
| 26 |
"""
|
| 27 |
-
Rewrites a rough project description
|
| 28 |
-
using a zero-shot large language model.
|
| 29 |
"""
|
| 30 |
if not user_input or not user_input.strip():
|
| 31 |
return "Please provide a description to improve."
|
|
@@ -34,31 +32,24 @@ def improve_description(user_input: str) -> str:
|
|
| 34 |
if len(user_text) < 10:
|
| 35 |
return "The description is too short. Please add more details about your project."
|
| 36 |
|
| 37 |
-
# Construct the prompt for the model
|
| 38 |
-
|
| 39 |
-
{"role": "system", "content": SYSTEM_PROMPT},
|
| 40 |
-
{
|
| 41 |
-
"role": "user",
|
| 42 |
-
"content": (
|
| 43 |
-
"Please rewrite the following project description into a professional "
|
| 44 |
-
"and technical summary, following all system guidelines.\n\n"
|
| 45 |
-
f"Original Description:\n'{user_text}'"
|
| 46 |
-
),
|
| 47 |
-
},
|
| 48 |
-
]
|
| 49 |
|
| 50 |
try:
|
| 51 |
-
#
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
temperature=0.
|
| 56 |
-
top_p=0.
|
|
|
|
| 57 |
)
|
| 58 |
-
|
|
|
|
|
|
|
| 59 |
|
| 60 |
-
# Final check to prevent returning the original text
|
| 61 |
-
if improved_text.lower() == user_text.lower():
|
| 62 |
return "Could not generate a meaningful improvement. Please try rephrasing or adding more detail."
|
| 63 |
|
| 64 |
return improved_text
|
|
@@ -69,7 +60,7 @@ def improve_description(user_input: str) -> str:
|
|
| 69 |
return "Sorry, the AI model service is currently unavailable. Please try again later."
|
| 70 |
|
| 71 |
|
| 72 |
-
# --- Gradio Interface ---
|
| 73 |
with gr.Blocks(title="Technical Description Assistant", theme=gr.themes.Soft()) as demo:
|
| 74 |
gr.Markdown(
|
| 75 |
"""
|
|
|
|
| 2 |
from huggingface_hub import InferenceClient
|
| 3 |
import os
|
| 4 |
|
| 5 |
+
# Use a reliable and powerful text-to-text model from the Hugging Face Inference API
|
| 6 |
+
MODEL_ID = "google/flan-t5-xxl"
|
| 7 |
|
|
|
|
|
|
|
| 8 |
client = InferenceClient(model=MODEL_ID)
|
| 9 |
|
| 10 |
+
# System prompt is less critical for non-instruct models, but we can use it in the user prompt.
|
| 11 |
+
REWRITE_TASK_PROMPT = """
|
| 12 |
+
Please rewrite the following rough project description into a clear, concise, and professional technical summary.
|
| 13 |
+
- Preserve all core technical details.
|
| 14 |
+
- Use a neutral and informative tone.
|
| 15 |
+
- Correct grammar and improve the structure.
|
| 16 |
+
- Do not repeat the original input.
|
| 17 |
+
|
| 18 |
+
Original Description:
|
| 19 |
+
'{user_text}'
|
| 20 |
+
|
| 21 |
+
Professional Summary:
|
| 22 |
""".strip()
|
| 23 |
|
| 24 |
def improve_description(user_input: str) -> str:
|
| 25 |
"""
|
| 26 |
+
Rewrites a rough project description using a zero-shot text-to-text model.
|
|
|
|
| 27 |
"""
|
| 28 |
if not user_input or not user_input.strip():
|
| 29 |
return "Please provide a description to improve."
|
|
|
|
| 32 |
if len(user_text) < 10:
|
| 33 |
return "The description is too short. Please add more details about your project."
|
| 34 |
|
| 35 |
+
# Construct the prompt for the text-to-text model
|
| 36 |
+
prompt = REWRITE_TASK_PROMPT.format(user_text=user_text)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
try:
|
| 39 |
+
# Use the `text_generation` method for non-chat models
|
| 40 |
+
improved_text = client.text_generation(
|
| 41 |
+
prompt,
|
| 42 |
+
max_new_tokens=300,
|
| 43 |
+
temperature=0.3,
|
| 44 |
+
top_p=0.95,
|
| 45 |
+
repetition_penalty=1.2,
|
| 46 |
)
|
| 47 |
+
|
| 48 |
+
# The output is a single string, clean it up.
|
| 49 |
+
improved_text = improved_text.strip()
|
| 50 |
|
| 51 |
+
# Final check to prevent returning the original text or empty strings
|
| 52 |
+
if not improved_text or improved_text.lower() == user_text.lower():
|
| 53 |
return "Could not generate a meaningful improvement. Please try rephrasing or adding more detail."
|
| 54 |
|
| 55 |
return improved_text
|
|
|
|
| 60 |
return "Sorry, the AI model service is currently unavailable. Please try again later."
|
| 61 |
|
| 62 |
|
| 63 |
+
# --- Gradio Interface (no changes needed here) ---
|
| 64 |
with gr.Blocks(title="Technical Description Assistant", theme=gr.themes.Soft()) as demo:
|
| 65 |
gr.Markdown(
|
| 66 |
"""
|