File size: 3,413 Bytes
646b139 ab1c53c 646b139 03dfa3e 8db5361 646b139 03dfa3e 646b139 8db5361 646b139 03dfa3e 646b139 |
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 |
```python
import gradio as gr
import os
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch
# Load Hugging Face API token securely
api_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
if not api_token:
raise ValueError("❌ ERROR: Hugging Face API token is not set. Please set it as an environment variable.")
# Define model names
base_model_name = "unsloth/qwen2.5-math-7b-bnb-4bit"
peft_model_name = "Hrushi02/Root_Math"
# Load base model with authentication
base_model = AutoModelForCausalLM.from_pretrained(
base_model_name,
torch_dtype=torch.float16,
device_map="auto",
use_auth_token=api_token # ✅ Correct
)
# Load fine-tuned model
model = PeftModel.from_pretrained(base_model, peft_model_name, token=api_token)
# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(base_model_name, token=api_token)
# Ensure pad_token is set
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_token
def respond(
message,
history: list[tuple[str, str]],
system_message,
max_tokens,
temperature,
top_p,
):
# Build messages list
messages = [{"role": "system", "content": system_message}]
for val in history:
if val[0]:
messages.append({"role": "user", "content": val[0]})
if val[1]:
messages.append({"role": "assistant", "content": val[1]})
messages.append({"role": "user", "content": message})
# Apply chat template
prompt = tokenizer.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
# Tokenize input
inputs = tokenizer([prompt], return_tensors="pt").to(model.device)
# Generate response with streaming
with torch.no_grad():
for new_token in model.generate(
**inputs,
max_new_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
do_sample=True,
pad_token_id=tokenizer.eos_token_id,
repetition_penalty=1.1,
streamer=None, # We'll handle streaming manually
):
# Decode the new token
new_token_decoded = tokenizer.decode(new_token[-1:], skip_special_tokens=True)
yield new_token_decoded
# Note: For true token-by-token streaming in Gradio, the above yields per-token.
# If you want full sentence streaming, accumulate and yield periodically, but this matches the original's per-token yield.
"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
respond,
additional_inputs=[
gr.Textbox(value="You are a helpful math assistant specialized in solving equations and finding roots.", label="System message"),
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.95,
step=0.05,
label="Top-p (nucleus sampling)",
),
],
chatbot=gr.Chatbot(type="messages"), # Modern format to avoid deprecation
title="Root Math Chatbot",
description="A fine-tuned Qwen2.5-Math model for solving roots and math problems."
)
if __name__ == "__main__":
demo.launch()
``` |