Update app.py
Browse files
app.py
CHANGED
|
@@ -1,8 +1,10 @@
|
|
| 1 |
import os
|
| 2 |
import time
|
|
|
|
| 3 |
import gradio as gr
|
| 4 |
from huggingface_hub import get_inference_endpoint
|
| 5 |
|
|
|
|
| 6 |
endpoint_url = os.getenv('ENDPOINT_URL')
|
| 7 |
personal_secret_token = os.getenv('PERSONAL_HF_TOKEN')
|
| 8 |
|
|
@@ -11,7 +13,17 @@ system_symbol = os.getenv('SYSTEM_SYMBOL')
|
|
| 11 |
user_symbol = os.getenv('USER_SYMBOL')
|
| 12 |
assistant_symbol = os.getenv('ASSISTANT_SYMBOL')
|
| 13 |
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
def respond(
|
| 17 |
message,
|
|
@@ -49,11 +61,10 @@ def respond(
|
|
| 49 |
max_new_tokens=max_new_tokens,
|
| 50 |
do_sample=temperature > 0,
|
| 51 |
top_p=top_p,
|
| 52 |
-
temperature=temperature
|
| 53 |
-
# stream=True,
|
| 54 |
)
|
| 55 |
|
| 56 |
-
response =
|
| 57 |
"inputs": turn_breaker.join(all_messages),
|
| 58 |
"parameters": generation_kwargs
|
| 59 |
})
|
|
|
|
| 1 |
import os
|
| 2 |
import time
|
| 3 |
+
import requests
|
| 4 |
import gradio as gr
|
| 5 |
from huggingface_hub import get_inference_endpoint
|
| 6 |
|
| 7 |
+
endpoint_name = os.getenv('ENDPOINT_NAME')
|
| 8 |
endpoint_url = os.getenv('ENDPOINT_URL')
|
| 9 |
personal_secret_token = os.getenv('PERSONAL_HF_TOKEN')
|
| 10 |
|
|
|
|
| 13 |
user_symbol = os.getenv('USER_SYMBOL')
|
| 14 |
assistant_symbol = os.getenv('ASSISTANT_SYMBOL')
|
| 15 |
|
| 16 |
+
headers = {
|
| 17 |
+
"Accept" : "application/json",
|
| 18 |
+
"Authorization": f"Bearer {personal_secret_token}",
|
| 19 |
+
"Content-Type": "application/json"
|
| 20 |
+
}
|
| 21 |
+
|
| 22 |
+
def query(payload):
|
| 23 |
+
response = requests.post(endpoint_url, headers=headers, json=payload)
|
| 24 |
+
return response.json()
|
| 25 |
+
|
| 26 |
+
endpoint = get_inference_endpoint(endpoint_name, token=personal_secret_token)
|
| 27 |
|
| 28 |
def respond(
|
| 29 |
message,
|
|
|
|
| 61 |
max_new_tokens=max_new_tokens,
|
| 62 |
do_sample=temperature > 0,
|
| 63 |
top_p=top_p,
|
| 64 |
+
temperature=temperature
|
|
|
|
| 65 |
)
|
| 66 |
|
| 67 |
+
response = query({
|
| 68 |
"inputs": turn_breaker.join(all_messages),
|
| 69 |
"parameters": generation_kwargs
|
| 70 |
})
|