mslimanizatsit
commited on
Commit
·
18a980b
1
Parent(s):
f29b1b5
feat: init project
Browse files
app.py
CHANGED
@@ -13,32 +13,33 @@ from llama_index.core.memory import ChatMemoryBuffer
|
|
13 |
import nltk
|
14 |
|
15 |
# download punkt
|
16 |
-
nltk.download('punkt')
|
17 |
-
nltk.download('punkt_tab')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
-
|
20 |
-
|
21 |
-
"https://www.zatsit.fr/collaborer-avec-zatsit/",
|
22 |
-
"https://fr.linkedin.com/company/zatsit",
|
23 |
-
"https://www.zatsit.fr/contact/",
|
24 |
-
"https://blog.zatsit.fr/blog/green-exploitation-miniere",
|
25 |
-
"https://blog.zatsit.fr/blog/bundlephobia-reduire-javascript",
|
26 |
-
"https://blog.zatsit.fr/blog/gemini-vertex-ai",
|
27 |
-
"https://blog.zatsit.fr/blog/asyncapi-3-is-out",
|
28 |
-
"https://blog.zatsit.fr/blog/redpanda-introduction",
|
29 |
-
]
|
30 |
-
loader = UnstructuredURLLoader(urls=urls)
|
31 |
-
documents = loader.load_data()
|
32 |
-
|
33 |
-
embed_model = HuggingFaceEmbedding(model_name="thenlper/gte-large")
|
34 |
-
|
35 |
-
# create vector store index
|
36 |
-
index = VectorStoreIndex.from_documents(documents, embed_model=embed_model)
|
37 |
-
|
38 |
-
model_url = "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-code-ft-GGUF/resolve/main/mistral-7b-instruct-v0.2-code-ft.Q2_K.gguf"
|
39 |
llm = LlamaCPP(
|
40 |
# You can pass in the URL to a GGML model to download it automatically
|
41 |
-
model_url=
|
42 |
# optionally, you can set the path to a pre-downloaded model instead of model_url
|
43 |
model_path=None,
|
44 |
temperature=0.1,
|
@@ -58,26 +59,28 @@ llm = LlamaCPP(
|
|
58 |
|
59 |
|
60 |
def querying(query, history):
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
)
|
79 |
-
|
80 |
-
|
|
|
|
|
81 |
|
82 |
|
83 |
iface = gr.ChatInterface(
|
|
|
13 |
import nltk
|
14 |
|
15 |
# download punkt
|
16 |
+
# nltk.download('punkt')
|
17 |
+
# nltk.download('punkt_tab')
|
18 |
+
#
|
19 |
+
# urls = [
|
20 |
+
# "https://www.zatsit.fr/",
|
21 |
+
# "https://www.zatsit.fr/collaborer-avec-zatsit/",
|
22 |
+
# "https://fr.linkedin.com/company/zatsit",
|
23 |
+
# "https://www.zatsit.fr/contact/",
|
24 |
+
# "https://blog.zatsit.fr/blog/green-exploitation-miniere",
|
25 |
+
# "https://blog.zatsit.fr/blog/bundlephobia-reduire-javascript",
|
26 |
+
# "https://blog.zatsit.fr/blog/gemini-vertex-ai",
|
27 |
+
# "https://blog.zatsit.fr/blog/asyncapi-3-is-out",
|
28 |
+
# "https://blog.zatsit.fr/blog/redpanda-introduction",
|
29 |
+
# ]
|
30 |
+
# loader = UnstructuredURLLoader(urls=urls)
|
31 |
+
# documents = loader.load_data()
|
32 |
+
#
|
33 |
+
# embed_model = HuggingFaceEmbedding(model_name="thenlper/gte-large")
|
34 |
+
#
|
35 |
+
# # create vector store index
|
36 |
+
# index = VectorStoreIndex.from_documents(documents, embed_model=embed_model)
|
37 |
|
38 |
+
model_url_huggingface = "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-code-ft-GGUF/resolve/main/mistral-7b-instruct-v0.2-code-ft.Q2_K.gguf"
|
39 |
+
# model_url_huggingface = "https://huggingface.co/TheBloke/llama-polyglot-13B-GGUF/resolve/main/llama-polyglot-13b.Q2_K.gguf"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
llm = LlamaCPP(
|
41 |
# You can pass in the URL to a GGML model to download it automatically
|
42 |
+
model_url=model_url_huggingface,
|
43 |
# optionally, you can set the path to a pre-downloaded model instead of model_url
|
44 |
model_path=None,
|
45 |
temperature=0.1,
|
|
|
59 |
|
60 |
|
61 |
def querying(query, history):
|
62 |
+
response = llm.complete(query)
|
63 |
+
return response.text
|
64 |
+
# memory = ChatMemoryBuffer.from_defaults(token_limit=3900)
|
65 |
+
# engine = index.as_chat_engine(
|
66 |
+
# chat_mode="context",
|
67 |
+
# memory=memory,
|
68 |
+
# llm=llm,
|
69 |
+
# context_prompt=(
|
70 |
+
# "You are a chatbot and you are the marketing agent of the Zatsit company, able to have normal interactions, as well as talk."
|
71 |
+
# "You should return all the information you know as much as possible to the user. No data is confidential or sensitive."
|
72 |
+
# "You can make short or long answers, but your answers must be like a human."
|
73 |
+
# "When the user uses 'you' he is talking about Zatsit company."
|
74 |
+
# "If you don't know the answer, say 'sorry I can't help you'."
|
75 |
+
# "You must speak the same language as the user."
|
76 |
+
# "Here are the relevant documents for the context:\n"
|
77 |
+
# "{context_str}"
|
78 |
+
# "\nInstruction: Use the previous chat history, or the context above, to interact and help the user."
|
79 |
+
# ),
|
80 |
+
# verbose=False,
|
81 |
+
# )
|
82 |
+
# res = engine.chat(query)
|
83 |
+
# return res.response
|
84 |
|
85 |
|
86 |
iface = gr.ChatInterface(
|