Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,43 +1,17 @@
|
|
1 |
import gradio as gr
|
2 |
import pandas as pd
|
3 |
import requests
|
4 |
-
from transformers import GPT2LMHeadModel, GPT2Tokenizer, AutoTokenizer, pipeline, AutoModelForCausalLM
|
5 |
-
from huggingface_hub import HfFolder
|
6 |
from io import StringIO
|
7 |
import os
|
8 |
-
import torch
|
9 |
|
10 |
# Access the Hugging Face API token from environment variables
|
11 |
hf_token = os.getenv('HF_API_TOKEN')
|
12 |
|
13 |
if not hf_token:
|
14 |
raise ValueError("Hugging Face API token is not set. Please set the HF_API_TOKEN environment variable.")
|
15 |
-
HfFolder.save_token(hf_token)
|
16 |
|
17 |
-
# Set
|
18 |
-
|
19 |
-
|
20 |
-
# Load the tokenizer and model
|
21 |
-
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
|
22 |
-
model_gpt2 = GPT2LMHeadModel.from_pretrained('gpt2')
|
23 |
-
|
24 |
-
# Create a pipeline for text generation using GPT-2
|
25 |
-
text_generator = pipeline("text-generation", model=model_gpt2, tokenizer=tokenizer)
|
26 |
-
|
27 |
-
# Lazy loading function for Llama-3 model
|
28 |
-
model_llama = None
|
29 |
-
tokenizer_llama = None
|
30 |
-
|
31 |
-
def load_llama_model():
|
32 |
-
global model_llama, tokenizer_llama
|
33 |
-
if model_llama is None:
|
34 |
-
model_name = "meta-llama/Meta-Llama-3.1-8B"
|
35 |
-
model_llama = AutoModelForCausalLM.from_pretrained(
|
36 |
-
model_name,
|
37 |
-
torch_dtype=torch.float16, # Use FP16 for reduced memory
|
38 |
-
use_auth_token=hf_token
|
39 |
-
)
|
40 |
-
tokenizer_llama = AutoTokenizer.from_pretrained(model_name, token=hf_token)
|
41 |
|
42 |
# Define your prompt template
|
43 |
prompt_template = """\
|
@@ -62,53 +36,45 @@ Columns:
|
|
62 |
Output: """
|
63 |
|
64 |
def preprocess_user_prompt(user_prompt):
|
65 |
-
|
66 |
-
return generated_text
|
67 |
|
68 |
def format_prompt(description, columns):
|
69 |
processed_description = preprocess_user_prompt(description)
|
70 |
prompt = prompt_template.format(description=processed_description, columns=",".join(columns))
|
71 |
return prompt
|
72 |
|
73 |
-
generation_params = {
|
74 |
-
"top_p": 0.90,
|
75 |
-
"temperature": 0.8,
|
76 |
-
"max_new_tokens": 512,
|
77 |
-
"return_full_text": False,
|
78 |
-
"use_cache": False
|
79 |
-
}
|
80 |
-
|
81 |
def generate_synthetic_data(description, columns):
|
82 |
try:
|
83 |
-
#
|
84 |
-
load_llama_model()
|
85 |
-
|
86 |
-
# Prepare the input for the Llama model
|
87 |
formatted_prompt = format_prompt(description, columns)
|
88 |
|
89 |
-
#
|
90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
max_length=512,
|
97 |
-
top_p=generation_params["top_p"],
|
98 |
-
temperature=generation_params["temperature"],
|
99 |
-
num_return_sequences=1
|
100 |
-
)
|
101 |
-
|
102 |
-
# Decode the generated output
|
103 |
-
generated_text = tokenizer_llama.decode(outputs[0], skip_special_tokens=True)
|
104 |
|
105 |
-
#
|
|
|
106 |
return generated_text
|
|
|
107 |
except Exception as e:
|
108 |
print(f"Error in generate_synthetic_data: {e}")
|
109 |
return f"Error: {e}"
|
110 |
|
111 |
-
|
112 |
def generate_large_synthetic_data(description, columns, num_rows=1000, rows_per_generation=100):
|
113 |
data_frames = []
|
114 |
num_iterations = num_rows // rows_per_generation
|
@@ -147,7 +113,6 @@ iface = gr.Interface(
|
|
147 |
api_name="generate" # Set the API name directly here
|
148 |
)
|
149 |
|
150 |
-
|
151 |
iface.api_name = "generate"
|
152 |
|
153 |
# Run the Gradio app
|
|
|
1 |
import gradio as gr
|
2 |
import pandas as pd
|
3 |
import requests
|
|
|
|
|
4 |
from io import StringIO
|
5 |
import os
|
|
|
6 |
|
7 |
# Access the Hugging Face API token from environment variables
|
8 |
hf_token = os.getenv('HF_API_TOKEN')
|
9 |
|
10 |
if not hf_token:
|
11 |
raise ValueError("Hugging Face API token is not set. Please set the HF_API_TOKEN environment variable.")
|
|
|
12 |
|
13 |
+
# Set the inference endpoint URL
|
14 |
+
inference_endpoint = "https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3.1-8B"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
# Define your prompt template
|
17 |
prompt_template = """\
|
|
|
36 |
Output: """
|
37 |
|
38 |
def preprocess_user_prompt(user_prompt):
|
39 |
+
return user_prompt
|
|
|
40 |
|
41 |
def format_prompt(description, columns):
|
42 |
processed_description = preprocess_user_prompt(description)
|
43 |
prompt = prompt_template.format(description=processed_description, columns=",".join(columns))
|
44 |
return prompt
|
45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
def generate_synthetic_data(description, columns):
|
47 |
try:
|
48 |
+
# Format the prompt
|
|
|
|
|
|
|
49 |
formatted_prompt = format_prompt(description, columns)
|
50 |
|
51 |
+
# Send a POST request to the Hugging Face Inference API
|
52 |
+
headers = {
|
53 |
+
"Authorization": f"Bearer {hf_token}",
|
54 |
+
"Content-Type": "application/json"
|
55 |
+
}
|
56 |
+
data = {
|
57 |
+
"inputs": formatted_prompt,
|
58 |
+
"parameters": {
|
59 |
+
"max_new_tokens": 512,
|
60 |
+
"top_p": 0.90,
|
61 |
+
"temperature": 0.8
|
62 |
+
}
|
63 |
+
}
|
64 |
|
65 |
+
response = requests.post(inference_endpoint, json=data, headers=headers)
|
66 |
+
|
67 |
+
if response.status_code != 200:
|
68 |
+
return f"Error: {response.status_code}, {response.text}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
+
# Extract the generated text from the response
|
71 |
+
generated_text = response.json()[0]['generated_text']
|
72 |
return generated_text
|
73 |
+
|
74 |
except Exception as e:
|
75 |
print(f"Error in generate_synthetic_data: {e}")
|
76 |
return f"Error: {e}"
|
77 |
|
|
|
78 |
def generate_large_synthetic_data(description, columns, num_rows=1000, rows_per_generation=100):
|
79 |
data_frames = []
|
80 |
num_iterations = num_rows // rows_per_generation
|
|
|
113 |
api_name="generate" # Set the API name directly here
|
114 |
)
|
115 |
|
|
|
116 |
iface.api_name = "generate"
|
117 |
|
118 |
# Run the Gradio app
|