Spaces:
Paused
Paused
Daniel Marques
commited on
Commit
•
1ae7653
1
Parent(s):
81fca18
feat: add backend
Browse files
main.py
CHANGED
@@ -32,9 +32,7 @@ class Predict(BaseModel):
|
|
32 |
class Delete(BaseModel):
|
33 |
filename: str
|
34 |
|
35 |
-
|
36 |
-
def on_llm_new_token(self, token: str, **kwargs) -> None:
|
37 |
-
print(f"token: {token}")
|
38 |
|
39 |
# if torch.backends.mps.is_available():
|
40 |
# DEVICE_TYPE = "mps"
|
@@ -57,7 +55,31 @@ DB = Chroma(
|
|
57 |
|
58 |
RETRIEVER = DB.as_retriever()
|
59 |
|
60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
|
62 |
template = """you are a helpful, respectful and honest assistant. You should only use the source documents provided to answer the questions.
|
63 |
You should only respond only topics that contains in documents use to training. Use the following pieces of context to answer the question at the end.
|
@@ -233,12 +255,14 @@ async def websocket_endpoint(websocket: WebSocket):
|
|
233 |
|
234 |
await websocket.accept()
|
235 |
while True:
|
236 |
-
|
237 |
|
238 |
-
data
|
239 |
|
240 |
-
|
|
|
|
|
|
|
|
|
241 |
|
242 |
-
print(f"${res}")
|
243 |
|
244 |
-
await websocket_state.send_text(f"Message text was:")
|
|
|
32 |
class Delete(BaseModel):
|
33 |
filename: str
|
34 |
|
35 |
+
|
|
|
|
|
36 |
|
37 |
# if torch.backends.mps.is_available():
|
38 |
# DEVICE_TYPE = "mps"
|
|
|
55 |
|
56 |
RETRIEVER = DB.as_retriever()
|
57 |
|
58 |
+
|
59 |
+
machine = Machine(lump)
|
60 |
+
solid = State('solid')
|
61 |
+
liquid = State('liquid')
|
62 |
+
|
63 |
+
class MyCustomSyncHandler(BaseCallbackHandler):
|
64 |
+
def __init__(self):
|
65 |
+
self.end = False
|
66 |
+
|
67 |
+
async def on_llm_start(
|
68 |
+
self, serialized: Dict[str, Any], prompts: List[str], **kwargs: Any
|
69 |
+
) -> None:
|
70 |
+
self.end = False
|
71 |
+
|
72 |
+
def on_llm_end(self, response: LLMResult, **kwargs: Any) -> None:
|
73 |
+
self.end = True
|
74 |
+
|
75 |
+
def on_llm_new_token(self, token: str, **kwargs) -> Any:
|
76 |
+
return token
|
77 |
+
|
78 |
+
|
79 |
+
handlerToken = MyCustomSyncHandler()
|
80 |
+
|
81 |
+
|
82 |
+
LLM = load_model(device_type=DEVICE_TYPE, model_id=MODEL_ID, model_basename=MODEL_BASENAME, stream=True, callbacks=[handlerToken])
|
83 |
|
84 |
template = """you are a helpful, respectful and honest assistant. You should only use the source documents provided to answer the questions.
|
85 |
You should only respond only topics that contains in documents use to training. Use the following pieces of context to answer the question at the end.
|
|
|
255 |
|
256 |
await websocket.accept()
|
257 |
while True:
|
258 |
+
data = await websocket.receive_text()
|
259 |
|
260 |
+
QA(data)
|
261 |
|
262 |
+
finish = False
|
263 |
+
while finish == False:
|
264 |
+
finish = handlerToken.end
|
265 |
+
token = handlerToken.on_llm_new_token()
|
266 |
+
await websocket.send_text(f"result: {token}")
|
267 |
|
|
|
268 |
|
|