Upload inference.py
Browse files- helpers/inference.py +66 -0
helpers/inference.py
ADDED
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from helpers import get_credentials
|
2 |
+
import requests
|
3 |
+
|
4 |
+
def hf_inference(prompt, model_id, temperature, max_new_tokens):
|
5 |
+
|
6 |
+
hf_token, _ = get_credentials.get_hf_credentials()
|
7 |
+
|
8 |
+
API_URL = "https://router.huggingface.co/together/v1/chat/completions"
|
9 |
+
headers = {
|
10 |
+
"Authorization": f"Bearer {hf_token}",
|
11 |
+
}
|
12 |
+
|
13 |
+
response = requests.post(
|
14 |
+
API_URL,
|
15 |
+
headers=headers,
|
16 |
+
json={
|
17 |
+
"messages": [
|
18 |
+
{
|
19 |
+
"role": "user",
|
20 |
+
"content": [
|
21 |
+
{
|
22 |
+
"type": "text",
|
23 |
+
"text": prompt
|
24 |
+
},
|
25 |
+
]
|
26 |
+
}
|
27 |
+
],
|
28 |
+
"model": model_id,
|
29 |
+
'temperature': temperature,
|
30 |
+
'max_new_tokens': max_new_tokens,
|
31 |
+
}
|
32 |
+
)
|
33 |
+
|
34 |
+
return response.json()["choices"][0]["message"]
|
35 |
+
|
36 |
+
def replicate_inference(prompt, model_id, temperature, max_new_tokens):
|
37 |
+
|
38 |
+
repl_token = get_credentials.get_replicate_credentials()
|
39 |
+
|
40 |
+
API_URL = f"https://api.replicate.com/v1/models/{model_id}/predictions"
|
41 |
+
headers = {
|
42 |
+
"Authorization": f"Bearer {repl_token}",
|
43 |
+
"Content-Type": "application/json",
|
44 |
+
"Prefer": "wait"
|
45 |
+
}
|
46 |
+
|
47 |
+
response = requests.post(
|
48 |
+
API_URL,
|
49 |
+
headers=headers,
|
50 |
+
json={
|
51 |
+
"input": {
|
52 |
+
"prompt": prompt,
|
53 |
+
"temperature": temperature,
|
54 |
+
"max_tokens": max_new_tokens,
|
55 |
+
}
|
56 |
+
}
|
57 |
+
)
|
58 |
+
|
59 |
+
return {
|
60 |
+
"content": "".join(response.json()['output'])
|
61 |
+
}
|
62 |
+
|
63 |
+
INFERENCE_HANDLER = {
|
64 |
+
'huggingface': hf_inference,
|
65 |
+
'replicate': replicate_inference
|
66 |
+
}
|