File size: 4,471 Bytes
926e371 d7fcb3e 00c54fb d7fcb3e 00c54fb d7fcb3e 00c54fb d7fcb3e 00c54fb d7fcb3e 00c54fb d7fcb3e 00c54fb d7fcb3e 00c54fb d7fcb3e 00c54fb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 |
import gradio as gr
from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
import torch # Import torch
# --- Model Loading (Do this only once, outside the function) ---
# Option 1: Pipeline (High-Level, Easier)
use_pipeline = True # Set to False to use the manual method
if use_pipeline:
pipe = pipeline("text-generation", model="kakaocorp/kanana-nano-2.1b-base", device="cpu") #Explicitly on CPU
else:
# Option 2: Manual Tokenizer and Model (More Control)
tokenizer = AutoTokenizer.from_pretrained("kakaocorp/kanana-nano-2.1b-base")
model = AutoModelForCausalLM.from_pretrained("kakaocorp/kanana-nano-2.1b-base")
# No need to move to GPU. It will default to CPU.
print("Model loaded on CPU")
# --- Generation Function ---
def generate_text(prompt, max_length=50, temperature=1.0, top_k=50, top_p=1.0, no_repeat_ngram_size=0, num_return_sequences=1):
"""Generates text based on the given prompt and parameters."""
if use_pipeline:
messages = [{"role": "user", "content": prompt}] # Format for pipeline
try:
result = pipe(
messages,
max_length=max_length,
temperature=temperature,
top_k=top_k,
top_p=top_p,
no_repeat_ngram_size=no_repeat_ngram_size,
num_return_sequences=num_return_sequences,
return_full_text=False, # Important: We only want generated text
pad_token_id=pipe.tokenizer.eos_token_id # Prevent warning, pipeline knows the EOS token
)
# Pipeline returns a list of dictionaries, each with 'generated_text'
return "\n\n".join([res['generated_text'] for res in result])
except Exception as e:
return f"Error during generation: {e}"
else: # Manual method
try:
inputs = tokenizer(prompt, return_tensors="pt")
# No need to move to GPU. Inputs will default to CPU.
outputs = model.generate(
**inputs,
max_length=max_length,
temperature=temperature,
top_k=top_k,
top_p=top_p,
no_repeat_ngram_size=no_repeat_ngram_size,
num_return_sequences=num_return_sequences,
pad_token_id=tokenizer.eos_token_id, # Ensure padding is correct
do_sample=True # Ensure sampling happens.
)
generated_texts = []
for i in range(outputs.shape[0]):
generated_text = tokenizer.decode(outputs[i], skip_special_tokens=True)
generated_texts.append(generated_text)
return "\n\n".join(generated_texts)
except Exception as e:
return f"Error during generation: {e}"
# --- Gradio Interface ---
with gr.Blocks() as demo:
gr.Markdown("# Text Generation with kakaocorp/kanana-nano-2.1b-base")
with gr.Row():
with gr.Column():
prompt_input = gr.Textbox(label="Prompt", placeholder="Enter your prompt here...")
with gr.Accordion("Generation Parameters", open=False):
max_length_slider = gr.Slider(label="Max Length", minimum=10, maximum=512, value=50, step=1)
temperature_slider = gr.Slider(label="Temperature", minimum=0.1, maximum=2.0, value=1.0, step=0.1)
top_k_slider = gr.Slider(label="Top K", minimum=0, maximum=100, value=50, step=1)
top_p_slider = gr.Slider(label="Top P", minimum=0.0, maximum=1.0, value=1.0, step=0.05)
no_repeat_ngram_size_slider = gr.Slider(label="No Repeat N-gram Size", minimum=0, maximum=10, value=0, step=1) # Add the slider
num_return_sequences_slider = gr.Slider(label="Number of Return Sequences", minimum=1, maximum=5, value=1, step=1)
generate_button = gr.Button("Generate")
with gr.Column():
output_text = gr.Textbox(label="Generated Text", interactive=False) # Use interactive=False
generate_button.click(
generate_text,
inputs=[
prompt_input,
max_length_slider,
temperature_slider,
top_k_slider,
top_p_slider,
no_repeat_ngram_size_slider,
num_return_sequences_slider
],
outputs=output_text,
)
demo.launch() # Remove share=True for local testing, add it back for deployment |