USER
I am writing the code that run llama local through gradio:"
# Gradio, work
import os
import time
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
import gradio as gr
from threading import Thread
# Assuming you have the model downloaded locally in a directory named "local_llama_model"
MODEL_PATH = r"C:\Users\shermanhonsf\Desktop\Python\llama_develop\model\Llama3.2-1B-Instruct-HF-Model"
TITLE = """
# Llama 3.2 1B
"""
PLACEHOLDER = """
Hi, I'm Llama. Ask me anything.
"""
CSS = """
.duplicate-button {
margin: auto !important;
color: white !important;
background: black !important;
border-radius: 100vh !important;
}
h3 {
text-align: center;
}
"""
# Set the device to "cuda" if you have a compatible GPU, otherwise "cpu"
device = "cuda" if torch.cuda.is_available() else "cpu"
# Configure quantization if needed (optional)
quantization_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.bfloat16,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4"
)
# Load the tokenizer and model from the local directory
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
torch_dtype=torch.bfloat16,
device_map="auto",
#quantization_config=quantization_config
)
# Move the model to the appropriate device
model.to(device)
def stream_chat(message: str, history: list, system_prompt: str = 'You are a helpful assistant', temperature: float = 0.8, max_new_tokens: int = 1024, top_p: float = 1.0, top_k: int = 20, penalty: float = 1.2):
print(f'message: {message}')
print(f'history: {history}')
print(f'system_prompt: {system_prompt}')
print(f'temperature: {temperature}')
print(f'max_new_tokens: {max_new_tokens}')
print(f'top_p: {top_p}')
print(f'top_k: {top_k}')
print(f'penalty: {penalty}')
print(f'Message: {message}')
print(f'History: {history}')
conversation = [
{"role": "system", "content": system_prompt}
]
for prompt, answer in history:
conversation.extend([
{"role": "user", "content": prompt},
{"role": "assistant", "content": answer},
])
conversation.append({"role": "user", "content": message})
# Prepare input IDs
input_ids = tokenizer.apply_chat_template(
conversation,
#tokenize=True,
add_generation_prompt=True,
return_tensors="pt"
).to(device)
streamer = TextIteratorStreamer(
tokenizer,
timeout=60.0,
skip_prompt=True,
skip_special_tokens=True
)
generate_kwargs = dict(
input_ids=input_ids,
max_new_tokens=max_new_tokens,
do_sample=False if temperature == 0 else True,
top_p=top_p,
top_k=top_k,
temperature=temperature,
eos_token_id=[128001, 128008, 128009],
streamer=streamer,
)
# Generate the response in a background thread
with torch.no_grad():
thread = Thread(target=model.generate, kwargs=generate_kwargs)
thread.start()
# Stream the response back to the user
buffer = ""
for new_text in streamer:
buffer += new_text
yield buffer
# Create the Gradio interface
chatbot = gr.Chatbot(height=600, show_copy_button=True, layout="panel", placeholder=PLACEHOLDER)
with gr.Blocks(css=CSS) as demo:
gr.HTML(TITLE)
gr.ChatInterface(
fn=stream_chat,
chatbot=chatbot,
fill_height=True,
additional_inputs_accordion=gr.Accordion(
label="⚙️ Parameters",
open=False,
render=False
),
additional_inputs=[
gr.Textbox(
value="You are a helpful assistant",
label="System Prompt",
render=False,
),
gr.Slider(
minimum=0,
maximum=1,
step=0.1,
value=0.8,
label="Temperature",
render=False,
),
gr.Slider(
minimum=128,
maximum=8192,
step=1,
value=1024,
label="Max new tokens",
render=False,
),
gr.Slider(
minimum=0.0,
maximum=1.0,
step=0.1,
value=1.0,
label="Top_p",
render=False,
),
gr.Slider(
minimum=1,
maximum=20,
step=1,
value=20,
label="Top_k",
render=False,
),
gr.Slider(
minimum=0.0,
maximum=2.0,
step=0.1,
value=1.2,
label="Repetition penalty",
render=False,
)
],
examples=[
["Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option."],
["What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter."],
["Tell me a random fun fact about the Roman Empire."],
["Show me a code snippet of a website's sticky header in CSS and JavaScript."],
],
cache_examples=False,
)
if __name__ == "__main__":
demo.launch(debug=True)
"
Make the chat interface more like a chatbotASSISTANT
To create a more chatbot-like interface using Gradio, you can make adjustments to enhance the interactive user experience and visibility. Here are some suggestions to make the Gradio-based chatbot experience more intuitive and engaging:
1. **Add User and Assistant Avatars**: You can use avatars to distinguish between user and system messages visually.
2. **Improved Layout**: Adjust the layout for better accessibility by reorganizing UI elements.
3. **Customize with CSS**: Use CSS to provide visual cues and a more polished look.
4. **Streamlined Message Handling**: Ensure messages are handled smoothly for both input and output.
Here’s how you can start implementing these suggestions in your code:
```python
import os
import time
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
import gradio as gr
from threading import Thread
MODEL_PATH = r"C:\Users\shermanhonsf\Desktop\Python\llama_develop\model\Llama3.2-1B-Instruct-HF-Model"
TITLE = """
# Llama 3.2 1B Chatbot
"""
PLACEHOLDER = "Hi, I'm Llama. Ask me anything."
CSS = """
.duplicate-button {
margin: auto !important;
color: white !important;
background: black !important;
border-radius: 100vh !important;
}
h3 {
text-align: center;
}
.chat-interface {
background-color: #f5f5f5;
overflow-y: auto;
}
.chat-message.user {
text-align: right;
margin: 5px 0;
}
.chat-message.assistant {
text-align: left;
margin: 5px 0;
}
"""
device = "cuda" if torch.cuda.is_available() else "cpu"
quantization_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.bfloat16,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4"
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
torch_dtype=torch.bfloat16,
device_map="auto"
)
model.to(device)
def stream_chat(message: str, history: list, **kwargs):
conversation = [{"role": "system", "content": kwargs.get('system_prompt', 'You are a helpful assistant')}]
for prompt, answer in history:
conversation.append({"role": "user", "content": prompt})
conversation.append({"role": "assistant", "content": answer})
conversation.append({"role": "user", "content": message})
input_ids = tokenizer.apply_chat_template(
conversation,
add_generation_prompt=True,
return_tensors="pt"
).to(device)
streamer = TextIteratorStreamer(
tokenizer,
timeout=60.0,
skip_prompt=True,
skip_special_tokens=True
)
generate_kwargs = dict(
input_ids=input_ids,
max_new_tokens=kwargs.get('max_new_tokens', 1024),
do_sample=kwargs.get('temperature', 0) != 0,
top_p=kwargs.get('top_p', 1.0),
top_k=kwargs.get('top_k', 20),
temperature=kwargs.get('temperature', 0.8),
eos_token_id=[128001, 128008, 128009],
streamer=streamer,
)
with torch.no_grad():
thread = Thread(target=model.generate, kwargs=generate_kwargs)
thread.start()
buffer = ""
for new_text in streamer:
buffer += new_text
yield buffer
chatbot = gr.Chatbot(height=600, show_copy_button=True, layout="panel", placeholder=PLACEHOLDER).style(
container_css={"border": "1px solid #e0e0e0; padding: 10px;"},
message_css={
"user": "chat-message user",
"assistant": "chat-message assistant"
}
)
with gr.Blocks(css=CSS) as demo:
gr.HTML(TITLE)
gr.ChatInterface(
fn=stream_chat,
chatbot=chatbot,
fill_height=True,
additional_inputs_accordion=gr.Accordion(
label="⚙️ Parameters",
open=False,
render=False
),
additional_inputs=[
gr.Textbox(
value="You are a helpful assistant",
label="System Prompt",
render=False,
),
gr.Slider(
minimum=0,
maximum=1,
step=0.1,
value=0.8,
label="Temperature",
render=False,
),
gr.Slider(
minimum=128,
maximum=8192,
step=1,
value=1024,
label="Max new tokens",
render=False,
),
gr.Slider(
minimum=0.0,
maximum=1.0,
step=0.1,
value=1.0,
label="Top_p",
render=False,
),
gr.Slider(
minimum=1,
maximum=20,
step=1,
value=20,
label="Top_k",
render=False,
),
gr.Slider(
minimum=0.0,
maximum=2.0,
step=0.1,
value=1.2,
label="Repetition penalty",
render=False,
)
],
examples=[
["Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option."],
["What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter."],
["Tell me a random fun fact about the Roman Empire."],
["Show me a code snippet of a website's sticky header in CSS and JavaScript."],
],
cache_examples=False,
)
if __name__ == "__main__":
demo.launch(debug=True)
```
### Key Adjustments Made:
- **CSS Improvements**: Improved layout appearance utilizing CSS for message appearance adjustments.
- **Message Handling**: Ensured clear delineation between user and assistant chat bubbles.
- **Chat Interface Title Change**: Increased clarity of chat's role.
Implement these adjustments to provide an elevated user experience, thus making interactions with your local Llama model smoother and more engaging.