Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -2,11 +2,8 @@ from fastapi import FastAPI, Query
|
|
2 |
from pydantic import BaseModel
|
3 |
import cloudscraper
|
4 |
from bs4 import BeautifulSoup
|
5 |
-
from transformers import
|
6 |
-
import torch
|
7 |
import re
|
8 |
-
import os
|
9 |
-
|
10 |
|
11 |
app = FastAPI()
|
12 |
|
@@ -34,39 +31,34 @@ def scrape(url: str = Query(...)):
|
|
34 |
return ThreadResponse(question=question, replies=replies)
|
35 |
return ThreadResponse(question="", replies=[])
|
36 |
|
37 |
-
|
38 |
MODEL_NAME = "google/flan-t5-small"
|
39 |
|
40 |
-
# Load
|
41 |
-
|
42 |
-
|
43 |
-
model
|
|
|
|
|
|
|
|
|
44 |
|
45 |
class PromptRequest(BaseModel):
|
46 |
prompt: str
|
47 |
|
48 |
@app.post("/generate")
|
49 |
async def generate_text(request: PromptRequest):
|
50 |
-
#
|
51 |
-
|
52 |
-
text = tokenizer.apply_chat_template(messages, tokenize=False, enable_thinking=True)
|
53 |
-
|
54 |
-
inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
55 |
-
with torch.no_grad():
|
56 |
-
generated_ids = model.generate(**inputs, max_new_tokens=512, temperature=0.5)
|
57 |
-
output_ids = generated_ids[:, inputs.input_ids.shape[-1]:].tolist()[0]
|
58 |
-
output_text = tokenizer.decode(output_ids)
|
59 |
|
60 |
# Extract reasoning and content parts if thinking tags are present
|
61 |
-
if "</think>" in
|
62 |
-
reasoning_content =
|
63 |
-
content =
|
64 |
else:
|
65 |
reasoning_content = ""
|
66 |
-
content =
|
67 |
|
68 |
return {
|
69 |
"reasoning_content": reasoning_content,
|
70 |
"generated_text": content
|
71 |
}
|
72 |
-
|
|
|
2 |
from pydantic import BaseModel
|
3 |
import cloudscraper
|
4 |
from bs4 import BeautifulSoup
|
5 |
+
from transformers import pipeline
|
|
|
6 |
import re
|
|
|
|
|
7 |
|
8 |
app = FastAPI()
|
9 |
|
|
|
31 |
return ThreadResponse(question=question, replies=replies)
|
32 |
return ThreadResponse(question="", replies=[])
|
33 |
|
|
|
34 |
MODEL_NAME = "google/flan-t5-small"
|
35 |
|
36 |
+
# Load the pipeline once at startup with device auto-mapping
|
37 |
+
text_generator = pipeline(
|
38 |
+
"text2text-generation",
|
39 |
+
model=MODEL_NAME,
|
40 |
+
device=0 if torch.cuda.is_available() else -1,
|
41 |
+
max_new_tokens=512,
|
42 |
+
temperature=0.5
|
43 |
+
)
|
44 |
|
45 |
class PromptRequest(BaseModel):
|
46 |
prompt: str
|
47 |
|
48 |
@app.post("/generate")
|
49 |
async def generate_text(request: PromptRequest):
|
50 |
+
# Use the pipeline to generate text directly
|
51 |
+
output = text_generator(request.prompt)[0]['generated_text']
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
|
53 |
# Extract reasoning and content parts if thinking tags are present
|
54 |
+
if "</think>" in output:
|
55 |
+
reasoning_content = output.split("</think>")[0].strip()
|
56 |
+
content = output.split("</think>")[1].strip().rstrip("</s>")
|
57 |
else:
|
58 |
reasoning_content = ""
|
59 |
+
content = output.strip().rstrip("</s>")
|
60 |
|
61 |
return {
|
62 |
"reasoning_content": reasoning_content,
|
63 |
"generated_text": content
|
64 |
}
|
|