Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,49 +1,62 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
|
| 3 |
|
| 4 |
-
# 1. Load
|
| 5 |
MODEL_NAME = "himu1780/ai-python-model"
|
| 6 |
|
| 7 |
print(f"Loading {MODEL_NAME}...")
|
| 8 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
| 9 |
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
|
| 10 |
|
| 11 |
-
# 2. Create the
|
| 12 |
generator = pipeline(
|
| 13 |
"text-generation",
|
| 14 |
model=model,
|
| 15 |
tokenizer=tokenizer,
|
| 16 |
-
device=-1
|
| 17 |
)
|
| 18 |
|
| 19 |
def chat_function(user_message, history):
|
| 20 |
-
# 3. THE
|
| 21 |
-
# We add "
|
| 22 |
-
|
|
|
|
| 23 |
|
| 24 |
-
# 4. Generate with
|
| 25 |
response = generator(
|
| 26 |
prompt,
|
| 27 |
-
max_new_tokens=250,
|
| 28 |
-
temperature=0.
|
| 29 |
-
repetition_penalty=1.2,
|
| 30 |
do_sample=True,
|
| 31 |
pad_token_id=tokenizer.eos_token_id
|
| 32 |
)
|
| 33 |
|
| 34 |
-
# 5. Clean output
|
| 35 |
full_text = response[0]['generated_text']
|
| 36 |
-
# Only keep the part after "Response:"
|
| 37 |
-
answer = full_text.split("### Response:")[-1].strip()
|
| 38 |
|
| 39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
-
# 6. Launch
|
| 42 |
demo = gr.ChatInterface(
|
| 43 |
fn=chat_function,
|
| 44 |
-
title="🐍 AI Python
|
| 45 |
-
description="I am
|
| 46 |
-
examples=[
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
)
|
| 48 |
|
| 49 |
demo.launch()
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
|
| 3 |
|
| 4 |
+
# 1. Load Your Smart Model (Step 1000)
|
| 5 |
MODEL_NAME = "himu1780/ai-python-model"
|
| 6 |
|
| 7 |
print(f"Loading {MODEL_NAME}...")
|
| 8 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
| 9 |
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
|
| 10 |
|
| 11 |
+
# 2. Create the Generator Pipeline
|
| 12 |
generator = pipeline(
|
| 13 |
"text-generation",
|
| 14 |
model=model,
|
| 15 |
tokenizer=tokenizer,
|
| 16 |
+
device=-1 # Use CPU
|
| 17 |
)
|
| 18 |
|
| 19 |
def chat_function(user_message, history):
|
| 20 |
+
# 3. THE FIX: Force the prompt structure
|
| 21 |
+
# We add "```python" at the end. This TRICKS the model.
|
| 22 |
+
# It thinks: "Oh, I already started writing a code block, I must finish it."
|
| 23 |
+
prompt = f"### Instruction:\n{user_message}\n\n### Response:\n```python\n"
|
| 24 |
|
| 25 |
+
# 4. Generate with Anti-Loop Settings
|
| 26 |
response = generator(
|
| 27 |
prompt,
|
| 28 |
+
max_new_tokens=250, # Give it enough space to write a full function
|
| 29 |
+
temperature=0.4, # Low temperature = Focused logic (No creative babbling)
|
| 30 |
+
repetition_penalty=1.2, # STRICT penalty. If it repeats "Hello", it gets banned.
|
| 31 |
do_sample=True,
|
| 32 |
pad_token_id=tokenizer.eos_token_id
|
| 33 |
)
|
| 34 |
|
| 35 |
+
# 5. Clean up the output
|
| 36 |
full_text = response[0]['generated_text']
|
|
|
|
|
|
|
| 37 |
|
| 38 |
+
# Isolate just the new code
|
| 39 |
+
# We split by "### Response:" to get the answer
|
| 40 |
+
answer_part = full_text.split("### Response:")[-1].strip()
|
| 41 |
+
|
| 42 |
+
# Ensure the formatting looks nice in the chat
|
| 43 |
+
# We stripped the ```python start, so we add it back for the UI
|
| 44 |
+
if not answer_part.startswith("```"):
|
| 45 |
+
answer_part = "```python\n" + answer_part
|
| 46 |
+
|
| 47 |
+
return answer_part
|
| 48 |
|
| 49 |
+
# 6. Launch the Chat Interface
|
| 50 |
demo = gr.ChatInterface(
|
| 51 |
fn=chat_function,
|
| 52 |
+
title="🐍 AI Python Assistant",
|
| 53 |
+
description="I am trained! Ask me to 'Write a function to...' or 'Create a loop...'",
|
| 54 |
+
examples=[
|
| 55 |
+
"Write a python function to add two numbers",
|
| 56 |
+
"Create a loop from 1 to 10",
|
| 57 |
+
"Write a script to calculate the area of a circle"
|
| 58 |
+
],
|
| 59 |
+
theme="soft"
|
| 60 |
)
|
| 61 |
|
| 62 |
demo.launch()
|