Update app.py
Browse files
app.py
CHANGED
@@ -1,51 +1,33 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
-
import torch
|
4 |
-
import tempfile
|
5 |
import os
|
|
|
|
|
6 |
|
7 |
-
#
|
8 |
-
|
9 |
-
os.makedirs(offload_dir, exist_ok=True)
|
10 |
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
torch_dtype=torch.bfloat16,
|
15 |
-
device_map="auto",
|
16 |
-
low_cpu_mem_usage=True,
|
17 |
-
offload_folder=offload_dir
|
18 |
)
|
19 |
-
tokenizer = AutoTokenizer.from_pretrained("tiiuae/meta-llama/Meta-Llama-3-8B")
|
20 |
-
|
21 |
-
|
22 |
-
def generate_text(input_text):
|
23 |
-
input_ids = tokenizer.encode(input_text, return_tensors="pt")
|
24 |
-
attention_mask = torch.ones(input_ids.shape)
|
25 |
-
|
26 |
-
output = model.generate(
|
27 |
-
input_ids,
|
28 |
-
attention_mask=attention_mask,
|
29 |
-
max_length=200,
|
30 |
-
do_sample=True,
|
31 |
-
top_k=10,
|
32 |
-
num_return_sequences=1,
|
33 |
-
eos_token_id=tokenizer.eos_token_id,
|
34 |
-
)
|
35 |
-
|
36 |
-
output_text = tokenizer.decode(output[0], skip_special_tokens=True)
|
37 |
-
print(output_text)
|
38 |
-
|
39 |
-
# Remove Prompt Echo from Generated Text
|
40 |
-
cleaned_output_text = output_text.replace(input_text, "")
|
41 |
-
return cleaned_output_text
|
42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
|
44 |
-
|
45 |
-
|
46 |
-
inputs=[
|
47 |
-
gr.inputs.Textbox(label="Input Text"),
|
48 |
-
],
|
49 |
-
outputs=gr.inputs.Textbox(label="Generated Text"),
|
50 |
-
title="---LLM---",
|
51 |
-
).launch()
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
+
import gradio as gr
|
3 |
+
from huggingface_hub import InferenceClient
|
4 |
|
5 |
+
# Retrieve the token from environment variable
|
6 |
+
token = os.getenv("HF_TOKEN")
|
|
|
7 |
|
8 |
+
client = InferenceClient(
|
9 |
+
"meta-llama/Llama-3.2-3B-Instruct",
|
10 |
+
token=token,
|
|
|
|
|
|
|
|
|
11 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
|
13 |
+
def chat_with_llama(user_input):
|
14 |
+
response = ""
|
15 |
+
for message in client.chat_completion(
|
16 |
+
messages=[{"role": "user", "content": user_input}],
|
17 |
+
max_tokens=500,
|
18 |
+
stream=True,
|
19 |
+
):
|
20 |
+
response += message.choices[0].delta.content
|
21 |
+
return response
|
22 |
+
|
23 |
+
# Create a Gradio interface
|
24 |
+
interface = gr.Interface(
|
25 |
+
fn=chat_with_llama,
|
26 |
+
inputs=gr.Textbox(label="Input Text", placeholder="Ask something..."),
|
27 |
+
outputs="text",
|
28 |
+
title="Chat with Llama 3",
|
29 |
+
description="Enter your message to chat with Llama 3. Type your question or prompt below.",
|
30 |
+
)
|
31 |
|
32 |
+
if __name__ == "__main__":
|
33 |
+
interface.launch()
|
|
|
|
|
|
|
|
|
|
|
|