code complete
Browse files- app.py +1 -0
- app_modules/llm_chat_chain.py +2 -1
- app_modules/llm_loader.py +4 -2
- server.py +39 -81
app.py
CHANGED
@@ -6,6 +6,7 @@ from timeit import default_timer as timer
|
|
6 |
|
7 |
import gradio as gr
|
8 |
from anyio.from_thread import start_blocking_portal
|
|
|
9 |
from app_modules.init import app_init
|
10 |
from app_modules.utils import print_llm_response, remove_extra_spaces
|
11 |
|
|
|
6 |
|
7 |
import gradio as gr
|
8 |
from anyio.from_thread import start_blocking_portal
|
9 |
+
|
10 |
from app_modules.init import app_init
|
11 |
from app_modules.utils import print_llm_response, remove_extra_spaces
|
12 |
|
app_modules/llm_chat_chain.py
CHANGED
@@ -1,7 +1,8 @@
|
|
|
|
1 |
from langchain.chains import ConversationalRetrievalChain
|
2 |
from langchain.chains.base import Chain
|
3 |
from langchain.memory import ConversationBufferMemory
|
4 |
-
|
5 |
from app_modules.llm_inference import LLMInference
|
6 |
|
7 |
|
|
|
1 |
+
from langchain import LLMChain, PromptTemplate
|
2 |
from langchain.chains import ConversationalRetrievalChain
|
3 |
from langchain.chains.base import Chain
|
4 |
from langchain.memory import ConversationBufferMemory
|
5 |
+
|
6 |
from app_modules.llm_inference import LLMInference
|
7 |
|
8 |
|
app_modules/llm_loader.py
CHANGED
@@ -93,7 +93,7 @@ class LLMLoader:
|
|
93 |
def __init__(self, llm_model_type, max_tokens_limit: int = 2048):
|
94 |
self.llm_model_type = llm_model_type
|
95 |
self.llm = None
|
96 |
-
self.streamer =
|
97 |
self.max_tokens_limit = max_tokens_limit
|
98 |
self.search_kwargs = {"k": 4}
|
99 |
|
@@ -138,7 +138,9 @@ class LLMLoader:
|
|
138 |
bnb_8bit_use_double_quant=load_quantized_model == "8bit",
|
139 |
)
|
140 |
|
141 |
-
callbacks = [
|
|
|
|
|
142 |
if custom_handler is not None:
|
143 |
callbacks.append(custom_handler)
|
144 |
|
|
|
93 |
def __init__(self, llm_model_type, max_tokens_limit: int = 2048):
|
94 |
self.llm_model_type = llm_model_type
|
95 |
self.llm = None
|
96 |
+
self.streamer = None
|
97 |
self.max_tokens_limit = max_tokens_limit
|
98 |
self.search_kwargs = {"k": 4}
|
99 |
|
|
|
138 |
bnb_8bit_use_double_quant=load_quantized_model == "8bit",
|
139 |
)
|
140 |
|
141 |
+
callbacks = []
|
142 |
+
if self.streamer is not None:
|
143 |
+
callbacks.append(self.streamer)
|
144 |
if custom_handler is not None:
|
145 |
callbacks.append(custom_handler)
|
146 |
|
server.py
CHANGED
@@ -1,74 +1,21 @@
|
|
1 |
"""Main entrypoint for the app."""
|
2 |
import json
|
3 |
import os
|
4 |
-
import time
|
5 |
-
from queue import Queue
|
6 |
from timeit import default_timer as timer
|
7 |
from typing import List, Optional
|
8 |
|
9 |
-
from langchain.embeddings import HuggingFaceInstructEmbeddings
|
10 |
-
from langchain.vectorstores.chroma import Chroma
|
11 |
-
from langchain.vectorstores.faiss import FAISS
|
12 |
from lcserve import serving
|
13 |
from pydantic import BaseModel
|
14 |
|
15 |
-
from app_modules.
|
16 |
-
from app_modules.
|
17 |
-
from app_modules.utils import
|
18 |
|
19 |
-
|
20 |
-
init_settings()
|
21 |
|
22 |
-
# https://github.com/huggingface/transformers/issues/17611
|
23 |
-
os.environ["CURL_CA_BUNDLE"] = ""
|
24 |
-
|
25 |
-
hf_embeddings_device_type, hf_pipeline_device_type = get_device_types()
|
26 |
-
print(f"hf_embeddings_device_type: {hf_embeddings_device_type}")
|
27 |
-
print(f"hf_pipeline_device_type: {hf_pipeline_device_type}")
|
28 |
-
|
29 |
-
hf_embeddings_model_name = (
|
30 |
-
os.environ.get("HF_EMBEDDINGS_MODEL_NAME") or "hkunlp/instructor-xl"
|
31 |
-
)
|
32 |
-
n_threds = int(os.environ.get("NUMBER_OF_CPU_CORES") or "4")
|
33 |
-
index_path = os.environ.get("FAISS_INDEX_PATH") or os.environ.get("CHROMADB_INDEX_PATH")
|
34 |
-
using_faiss = os.environ.get("FAISS_INDEX_PATH") is not None
|
35 |
-
llm_model_type = os.environ.get("LLM_MODEL_TYPE")
|
36 |
chat_history_enabled = os.environ.get("CHAT_HISTORY_ENABLED") == "true"
|
37 |
-
show_param_settings = os.environ.get("SHOW_PARAM_SETTINGS") == "true"
|
38 |
-
share_gradio_app = os.environ.get("SHARE_GRADIO_APP") == "true"
|
39 |
-
|
40 |
-
|
41 |
-
streaming_enabled = True # llm_model_type in ["openai", "llamacpp"]
|
42 |
-
|
43 |
-
start = timer()
|
44 |
-
embeddings = HuggingFaceInstructEmbeddings(
|
45 |
-
model_name=hf_embeddings_model_name,
|
46 |
-
model_kwargs={"device": hf_embeddings_device_type},
|
47 |
-
)
|
48 |
-
end = timer()
|
49 |
-
|
50 |
-
print(f"Completed in {end - start:.3f}s")
|
51 |
-
|
52 |
-
start = timer()
|
53 |
-
|
54 |
-
print(f"Load index from {index_path} with {'FAISS' if using_faiss else 'Chroma'}")
|
55 |
|
56 |
-
|
57 |
-
raise ValueError(f"{index_path} does not exist!")
|
58 |
-
elif using_faiss:
|
59 |
-
vectorstore = FAISS.load_local(index_path, embeddings)
|
60 |
-
else:
|
61 |
-
vectorstore = Chroma(embedding_function=embeddings, persist_directory=index_path)
|
62 |
-
|
63 |
-
end = timer()
|
64 |
-
|
65 |
-
print(f"Completed in {end - start:.3f}s")
|
66 |
-
|
67 |
-
start = timer()
|
68 |
-
qa_chain = QAChain(vectorstore, llm_model_type)
|
69 |
-
qa_chain.init(n_threds=n_threds, hf_pipeline_device_type=hf_pipeline_device_type)
|
70 |
-
end = timer()
|
71 |
-
print(f"Completed in {end - start:.3f}s")
|
72 |
|
73 |
|
74 |
class ChatResponse(BaseModel):
|
@@ -80,30 +27,41 @@ class ChatResponse(BaseModel):
|
|
80 |
|
81 |
|
82 |
@serving(websocket=True)
|
83 |
-
def chat(
|
|
|
|
|
|
|
84 |
# Get the `streaming_handler` from `kwargs`. This is used to stream data to the client.
|
85 |
-
streaming_handler = kwargs.get("streaming_handler")
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
resp.
|
103 |
-
|
104 |
-
|
105 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
106 |
|
107 |
|
108 |
if __name__ == "__main__":
|
109 |
-
print_llm_response(json.loads(chat("What
|
|
|
1 |
"""Main entrypoint for the app."""
|
2 |
import json
|
3 |
import os
|
|
|
|
|
4 |
from timeit import default_timer as timer
|
5 |
from typing import List, Optional
|
6 |
|
|
|
|
|
|
|
7 |
from lcserve import serving
|
8 |
from pydantic import BaseModel
|
9 |
|
10 |
+
from app_modules.init import app_init
|
11 |
+
from app_modules.llm_chat_chain import ChatChain
|
12 |
+
from app_modules.utils import print_llm_response
|
13 |
|
14 |
+
llm_loader, qa_chain = app_init()
|
|
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
chat_history_enabled = os.environ.get("CHAT_HISTORY_ENABLED") == "true"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
+
uuid_to_chat_chain_mapping = dict()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
|
21 |
class ChatResponse(BaseModel):
|
|
|
27 |
|
28 |
|
29 |
@serving(websocket=True)
|
30 |
+
def chat(
|
31 |
+
question: str, history: Optional[List] = [], uuid: Optional[str] = None, **kwargs
|
32 |
+
) -> str:
|
33 |
+
print(f"uuid: {uuid}")
|
34 |
# Get the `streaming_handler` from `kwargs`. This is used to stream data to the client.
|
35 |
+
streaming_handler = kwargs.get("streaming_handler")
|
36 |
+
if uuid is None:
|
37 |
+
chat_history = []
|
38 |
+
if chat_history_enabled:
|
39 |
+
for element in history:
|
40 |
+
item = (element[0] or "", element[1] or "")
|
41 |
+
chat_history.append(item)
|
42 |
+
|
43 |
+
start = timer()
|
44 |
+
result = qa_chain.call_chain(
|
45 |
+
{"question": question, "chat_history": chat_history}, streaming_handler
|
46 |
+
)
|
47 |
+
end = timer()
|
48 |
+
print(f"Completed in {end - start:.3f}s")
|
49 |
+
|
50 |
+
resp = ChatResponse(sourceDocs=result["source_documents"])
|
51 |
+
|
52 |
+
return json.dumps(resp.dict())
|
53 |
+
else:
|
54 |
+
if uuid in uuid_to_chat_chain_mapping:
|
55 |
+
chat = uuid_to_chat_chain_mapping[uuid]
|
56 |
+
else:
|
57 |
+
chat = ChatChain(llm_loader)
|
58 |
+
uuid_to_chat_chain_mapping[uuid] = chat
|
59 |
+
result = chat.call_chain({"question": question}, streaming_handler)
|
60 |
+
print(f"result: {result}")
|
61 |
+
|
62 |
+
resp = ChatResponse(sourceDocs=[])
|
63 |
+
return json.dumps(resp.dict())
|
64 |
|
65 |
|
66 |
if __name__ == "__main__":
|
67 |
+
print_llm_response(json.loads(chat("What's deep learning?", [])))
|