Spaces:
Runtime error
Runtime error
Update TextGen/router.py
Browse files- TextGen/router.py +11 -0
TextGen/router.py
CHANGED
|
@@ -48,6 +48,17 @@ app.add_middleware(
|
|
| 48 |
def api_home():
|
| 49 |
return {'detail': 'Welcome to FastAPI TextGen Tutorial!'}
|
| 50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
@app.post("/api/generate", summary="Generate text from prompt", tags=["Generate"], response_model=Generate)
|
| 52 |
def inference(input_prompt: str):
|
| 53 |
return generate_text(prompt=input_prompt)
|
|
|
|
| 48 |
def api_home():
|
| 49 |
return {'detail': 'Welcome to FastAPI TextGen Tutorial!'}
|
| 50 |
|
| 51 |
+
|
| 52 |
+
API_URL = "https://api-inference.huggingface.co/models/google/gemma-1.1-7b-it"
|
| 53 |
+
|
| 54 |
+
def query(payload):
|
| 55 |
+
response = requests.post(API_URL, json=payload)
|
| 56 |
+
return response.json()
|
| 57 |
+
|
| 58 |
+
@app.get("/query", tags=["Query"])
|
| 59 |
+
def api_query(input: str):
|
| 60 |
+
return query({"inputs": input},"parameters":{"return_full_text":False,"max_length":2048})
|
| 61 |
+
|
| 62 |
@app.post("/api/generate", summary="Generate text from prompt", tags=["Generate"], response_model=Generate)
|
| 63 |
def inference(input_prompt: str):
|
| 64 |
return generate_text(prompt=input_prompt)
|