Spaces:
Runtime error
Runtime error
Use dropdown to select source (#71)
Browse files* add dropdown menu for switching data sources
* Add ability to update Buster's config on the fly
* Add lightning, godot documentation sources
* add download script for the weights (from huggingface dataset)
* update tests
* Add logging to pytest
* Fix source titles when returning results
* return percentages instead of cosine score
* change source directly when you call chat
- .gitattributes +0 -1
- .gitignore +1 -0
- buster/apps/bot_configs.py +175 -0
- buster/apps/gradio_app.py +33 -50
- buster/buster.py +47 -29
- buster/data/documents.db +0 -3
- buster/documents/utils.py +13 -0
- buster/formatter/base.py +3 -2
- buster/formatter/gradio.py +1 -1
- buster/formatter/html.py +1 -1
- buster/formatter/markdown.py +1 -1
- buster/formatter/slack.py +1 -1
- pyproject.toml +4 -0
- tests/test_chatbot.py +33 -14
.gitattributes
DELETED
@@ -1 +0,0 @@
|
|
1 |
-
*.db filter=lfs diff=lfs merge=lfs -text
|
|
|
|
.gitignore
CHANGED
@@ -1,3 +1,4 @@
|
|
|
|
1 |
# Byte-compiled / optimized / DLL files
|
2 |
__pycache__/
|
3 |
*.py[cod]
|
|
|
1 |
+
buster/apps/data/
|
2 |
# Byte-compiled / optimized / DLL files
|
3 |
__pycache__/
|
4 |
*.py[cod]
|
buster/apps/bot_configs.py
ADDED
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from buster.buster import BusterConfig
|
2 |
+
|
3 |
+
huggingface_cfg = BusterConfig(
|
4 |
+
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
5 |
+
embedding_model="text-embedding-ada-002",
|
6 |
+
top_k=3,
|
7 |
+
thresh=0.7,
|
8 |
+
max_words=3000,
|
9 |
+
completer_cfg={
|
10 |
+
"name": "ChatGPT",
|
11 |
+
"text_before_documents": (
|
12 |
+
"You are a chatbot assistant answering technical questions about huggingface transformers, a library to train transformers in python. "
|
13 |
+
"You can only respond to a question if the content necessary to answer the question is contained in the following provided documentation. "
|
14 |
+
"If the answer is in the documentation, summarize it in a helpful way to the user. "
|
15 |
+
"If it isn't, simply reply that you cannot answer the question. "
|
16 |
+
"Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
17 |
+
"Here is the documentation: "
|
18 |
+
"<DOCUMENTS> "
|
19 |
+
),
|
20 |
+
"text_before_prompt": (
|
21 |
+
"<\DOCUMENTS>\n"
|
22 |
+
"REMEMBER:\n"
|
23 |
+
"You are a chatbot assistant answering technical questions about huggingface transformers, a library to train transformers in python. "
|
24 |
+
"Here are the rules you must follow:\n"
|
25 |
+
"1) You must only respond with information contained in the documentation above. Say you do not know if the information is not provided.\n"
|
26 |
+
"2) Make sure to format your answers in Markdown format, including code block and snippets.\n"
|
27 |
+
"3) Do not reference any links, urls or hyperlinks in your answers.\n"
|
28 |
+
"4) If you do not know the answer to a question, or if it is completely irrelevant to the library usage, simply reply with:\n"
|
29 |
+
"5) Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
30 |
+
"'I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?'"
|
31 |
+
"For example:\n"
|
32 |
+
"What is the meaning of life for huggingface?\n"
|
33 |
+
"I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"
|
34 |
+
"Now answer the following question:\n"
|
35 |
+
),
|
36 |
+
"completion_kwargs": {
|
37 |
+
"model": "gpt-3.5-turbo",
|
38 |
+
},
|
39 |
+
},
|
40 |
+
response_format="gradio",
|
41 |
+
source="huggingface",
|
42 |
+
)
|
43 |
+
|
44 |
+
|
45 |
+
pytorch_cfg = BusterConfig(
|
46 |
+
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the pytorch library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
47 |
+
embedding_model="text-embedding-ada-002",
|
48 |
+
top_k=3,
|
49 |
+
thresh=0.7,
|
50 |
+
max_words=3000,
|
51 |
+
completer_cfg={
|
52 |
+
"name": "ChatGPT",
|
53 |
+
"text_before_documents": (
|
54 |
+
"You are a chatbot assistant answering technical questions about pytorch, a library to train neural networks in python. "
|
55 |
+
"You can only respond to a question if the content necessary to answer the question is contained in the following provided documentation. "
|
56 |
+
"If the answer is in the documentation, summarize it in a helpful way to the user. "
|
57 |
+
"If it isn't, simply reply that you cannot answer the question. "
|
58 |
+
"Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
59 |
+
"Here is the documentation: "
|
60 |
+
"<DOCUMENTS> "
|
61 |
+
),
|
62 |
+
"text_before_prompt": (
|
63 |
+
"<\DOCUMENTS>\n"
|
64 |
+
"REMEMBER:\n"
|
65 |
+
"You are a chatbot assistant answering technical questions about pytorch transformers, a library to train neural networks in python. "
|
66 |
+
"Here are the rules you must follow:\n"
|
67 |
+
"1) You must only respond with information contained in the documentation above. Say you do not know if the information is not provided.\n"
|
68 |
+
"2) Make sure to format your answers in Markdown format, including code block and snippets.\n"
|
69 |
+
"3) Do not include any links, urls or hyperlinks in your answers.\n"
|
70 |
+
"4) If you do not know the answer to a question, or if it is completely irrelevant to the library usage, simply reply with:\n"
|
71 |
+
"5) Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
72 |
+
"'I'm sorry, but I am an AI language model trained to assist with questions related to the pytorch transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?'"
|
73 |
+
"For example:\n"
|
74 |
+
"What is the meaning of life for pytorch?\n"
|
75 |
+
"I'm sorry, but I am an AI language model trained to assist with questions related to the pytorch library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"
|
76 |
+
"Now answer the following question:\n"
|
77 |
+
),
|
78 |
+
"completion_kwargs": {
|
79 |
+
"model": "gpt-3.5-turbo",
|
80 |
+
},
|
81 |
+
},
|
82 |
+
response_format="gradio",
|
83 |
+
source="pytorch",
|
84 |
+
)
|
85 |
+
|
86 |
+
lightning_cfg = BusterConfig(
|
87 |
+
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the pytorch lightning library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
88 |
+
embedding_model="text-embedding-ada-002",
|
89 |
+
top_k=3,
|
90 |
+
thresh=0.7,
|
91 |
+
max_words=3000,
|
92 |
+
completer_cfg={
|
93 |
+
"name": "ChatGPT",
|
94 |
+
"text_before_documents": (
|
95 |
+
"You are a chatbot assistant answering technical questions about pytorch lightning, a library to train neural networks in python. "
|
96 |
+
"You can only respond to a question if the content necessary to answer the question is contained in the following provided documentation. "
|
97 |
+
"If the answer is in the documentation, summarize it in a helpful way to the user. "
|
98 |
+
"If it isn't, simply reply that you cannot answer the question. "
|
99 |
+
"Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
100 |
+
"Here is the documentation: "
|
101 |
+
"<DOCUMENTS> "
|
102 |
+
),
|
103 |
+
"text_before_prompt": (
|
104 |
+
"<\DOCUMENTS>\n"
|
105 |
+
"REMEMBER:\n"
|
106 |
+
"You are a chatbot assistant answering technical questions about pytorch lightning transformers, a library to train neural networks in python. "
|
107 |
+
"Here are the rules you must follow:\n"
|
108 |
+
"1) You must only respond with information contained in the documentation above. Say you do not know if the information is not provided.\n"
|
109 |
+
"2) Make sure to format your answers in Markdown format, including code block and snippets.\n"
|
110 |
+
"3) Do not include any links, urls or hyperlinks in your answers.\n"
|
111 |
+
"4) If you do not know the answer to a question, or if it is completely irrelevant to the library usage, simply reply with:\n"
|
112 |
+
"5) Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
113 |
+
"'I'm sorry, but I am an AI language model trained to assist with questions related to the pytorch lightning library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?'"
|
114 |
+
"For example:\n"
|
115 |
+
"What is the meaning of life for pytorch lightning?\n"
|
116 |
+
"I'm sorry, but I am an AI language model trained to assist with questions related to the pytorch lightning library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"
|
117 |
+
"Now answer the following question:\n"
|
118 |
+
),
|
119 |
+
"completion_kwargs": {
|
120 |
+
"model": "gpt-3.5-turbo",
|
121 |
+
},
|
122 |
+
},
|
123 |
+
response_format="gradio",
|
124 |
+
source="lightning",
|
125 |
+
)
|
126 |
+
|
127 |
+
|
128 |
+
godot_cfg = BusterConfig(
|
129 |
+
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the godot library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
130 |
+
embedding_model="text-embedding-ada-002",
|
131 |
+
top_k=3,
|
132 |
+
thresh=0.7,
|
133 |
+
max_words=3000,
|
134 |
+
completer_cfg={
|
135 |
+
"name": "ChatGPT",
|
136 |
+
"text_before_documents": (
|
137 |
+
"You are a chatbot assistant answering technical questions about godot, a game-engine library. "
|
138 |
+
"You can only respond to a question if the content necessary to answer the question is contained in the following provided documentation. "
|
139 |
+
"If the answer is in the documentation, summarize it in a helpful way to the user. "
|
140 |
+
"If it isn't, simply reply that you cannot answer the question. "
|
141 |
+
"Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
142 |
+
"Here is the documentation: "
|
143 |
+
"<DOCUMENTS> "
|
144 |
+
),
|
145 |
+
"text_before_prompt": (
|
146 |
+
"<\DOCUMENTS>\n"
|
147 |
+
"REMEMBER:\n"
|
148 |
+
"You are a chatbot assistant answering technical questions about godot, a game-engine library."
|
149 |
+
"Here are the rules you must follow:\n"
|
150 |
+
"1) You must only respond with information contained in the documentation above. Say you do not know if the information is not provided.\n"
|
151 |
+
"2) Make sure to format your answers in Markdown format, including code block and snippets.\n"
|
152 |
+
"3) Do not include any links, urls or hyperlinks in your answers.\n"
|
153 |
+
"4) If you do not know the answer to a question, or if it is completely irrelevant to the library usage, simply reply with:\n"
|
154 |
+
"5) Do not refer to the documentation directly, but use the instructions provided within it to answer questions. "
|
155 |
+
"'I'm sorry, but I am an AI language model trained to assist with questions related to the godot library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?'"
|
156 |
+
"For example:\n"
|
157 |
+
"What is the meaning of life for godot?\n"
|
158 |
+
"I'm sorry, but I am an AI language model trained to assist with questions related to the pytorch lightning library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"
|
159 |
+
"Now answer the following question:\n"
|
160 |
+
),
|
161 |
+
"completion_kwargs": {
|
162 |
+
"model": "gpt-3.5-turbo",
|
163 |
+
},
|
164 |
+
},
|
165 |
+
response_format="gradio",
|
166 |
+
source="godot",
|
167 |
+
)
|
168 |
+
|
169 |
+
|
170 |
+
available_configs = {
|
171 |
+
"huggingface": huggingface_cfg,
|
172 |
+
"pytorch": pytorch_cfg,
|
173 |
+
"pytorch-lightning": lightning_cfg,
|
174 |
+
"godot": godot_cfg,
|
175 |
+
}
|
buster/apps/gradio_app.py
CHANGED
@@ -3,53 +3,27 @@ import pathlib
|
|
3 |
|
4 |
import gradio as gr
|
5 |
|
|
|
6 |
from buster.buster import Buster, BusterConfig
|
|
|
|
|
7 |
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
"name": "ChatGPT",
|
19 |
-
"text_before_documents": (
|
20 |
-
"You are a chatbot assistant answering technical questions about huggingface transformers, a library to train transformers in python. "
|
21 |
-
"You can only respond to a question if the content necessary to answer the question is contained in the following provided documentation. "
|
22 |
-
"If it isn't, simply reply that you cannot answer the question. "
|
23 |
-
"Here is the documentation: "
|
24 |
-
"<BEGIN_DOCUMENTATION> "
|
25 |
-
),
|
26 |
-
"text_before_prompt": (
|
27 |
-
"<\END_DOCUMENTATION>\n"
|
28 |
-
"REMINDER:\n"
|
29 |
-
"You are a chatbot assistant answering technical questions about huggingface transformers, a library to train transformers in python. "
|
30 |
-
"Here are the rules you must follow:\n"
|
31 |
-
"1) You must only respond with information contained in the documentation above. Say you do not know if the information is not provided.\n"
|
32 |
-
"2) Make sure to format your answers in Markdown format, including code block and snippets.\n"
|
33 |
-
"3) Do not include any links to urls or hyperlinks in your answers.\n"
|
34 |
-
"4) If you do not know the answer to a question, or if it is completely irrelevant to the library usage, simply reply with:\n"
|
35 |
-
"'I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?'"
|
36 |
-
"For example:\n"
|
37 |
-
"What is the meaning of life for huggingface?\n"
|
38 |
-
"I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"
|
39 |
-
"Now answer the following question:\n"
|
40 |
-
),
|
41 |
-
"completion_kwargs": {
|
42 |
-
"model": "gpt-3.5-turbo",
|
43 |
-
},
|
44 |
-
},
|
45 |
-
response_format="gradio",
|
46 |
-
)
|
47 |
-
buster = Buster(buster_cfg)
|
48 |
-
|
49 |
-
|
50 |
-
def chat(question, history):
|
51 |
-
history = history or []
|
52 |
|
|
|
|
|
|
|
|
|
|
|
53 |
answer = buster.process_input(question)
|
54 |
|
55 |
# formatting hack for code blocks to render properly every time
|
@@ -59,11 +33,20 @@ def chat(question, history):
|
|
59 |
return history, history
|
60 |
|
61 |
|
62 |
-
block = gr.Blocks(css=".
|
63 |
|
64 |
with block:
|
65 |
with gr.Row():
|
66 |
-
gr.Markdown("<h3><center>Buster 🤖: A Question-Answering Bot for
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
|
68 |
chatbot = gr.Chatbot()
|
69 |
|
@@ -75,11 +58,11 @@ with block:
|
|
75 |
)
|
76 |
submit = gr.Button(value="Send", variant="secondary").style(full_width=False)
|
77 |
|
78 |
-
gr.Examples(
|
|
|
79 |
examples=[
|
80 |
"What kind of models should I use for images and text?",
|
81 |
"When should I finetune a model vs. training it form scratch?",
|
82 |
-
"How can I deploy my trained huggingface model?",
|
83 |
"Can you give me some python code to quickly finetune a model on my sentiment analysis dataset?",
|
84 |
],
|
85 |
inputs=message,
|
@@ -95,8 +78,8 @@ with block:
|
|
95 |
state = gr.State()
|
96 |
agent_state = gr.State()
|
97 |
|
98 |
-
submit.click(chat, inputs=[message, state], outputs=[chatbot, state])
|
99 |
-
message.submit(chat, inputs=[message, state], outputs=[chatbot, state])
|
100 |
|
101 |
|
102 |
block.launch(debug=True)
|
|
|
3 |
|
4 |
import gradio as gr
|
5 |
|
6 |
+
from buster.apps.bot_configs import available_configs
|
7 |
from buster.buster import Buster, BusterConfig
|
8 |
+
from buster.documents.base import DocumentsManager
|
9 |
+
from buster.documents.utils import download_db, get_documents_manager_from_extension
|
10 |
|
11 |
+
DEFAULT_CONFIG = "huggingface"
|
12 |
+
DB_URL = "https://huggingface.co/datasets/jerpint/buster-data/resolve/main/documents.db"
|
13 |
+
|
14 |
+
# Download the db...
|
15 |
+
documents_filepath = download_db(db_url=DB_URL, output_dir="./data")
|
16 |
+
documents: DocumentsManager = get_documents_manager_from_extension(documents_filepath)(documents_filepath)
|
17 |
+
|
18 |
+
# initialize buster with the default config...
|
19 |
+
default_cfg: BusterConfig = available_configs.get(DEFAULT_CONFIG)
|
20 |
+
buster = Buster(cfg=default_cfg, documents=documents)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
+
|
23 |
+
def chat(question, history, bot_source):
|
24 |
+
history = history or []
|
25 |
+
cfg = available_configs.get(bot_source)
|
26 |
+
buster.update_cfg(cfg)
|
27 |
answer = buster.process_input(question)
|
28 |
|
29 |
# formatting hack for code blocks to render properly every time
|
|
|
33 |
return history, history
|
34 |
|
35 |
|
36 |
+
block = gr.Blocks(css="#chatbot .overflow-y-auto{height:500px}")
|
37 |
|
38 |
with block:
|
39 |
with gr.Row():
|
40 |
+
gr.Markdown("<h3><center>Buster 🤖: A Question-Answering Bot for open-source libraries </center></h3>")
|
41 |
+
|
42 |
+
doc_source = gr.Dropdown(
|
43 |
+
choices=sorted(list(available_configs.keys())),
|
44 |
+
value=DEFAULT_CONFIG,
|
45 |
+
interactive=True,
|
46 |
+
multiselect=False,
|
47 |
+
label="Source of Documentation",
|
48 |
+
info="The source of documentation to select from",
|
49 |
+
)
|
50 |
|
51 |
chatbot = gr.Chatbot()
|
52 |
|
|
|
58 |
)
|
59 |
submit = gr.Button(value="Send", variant="secondary").style(full_width=False)
|
60 |
|
61 |
+
examples = gr.Examples(
|
62 |
+
# TODO: seems not possible (for now) to update examples on change...
|
63 |
examples=[
|
64 |
"What kind of models should I use for images and text?",
|
65 |
"When should I finetune a model vs. training it form scratch?",
|
|
|
66 |
"Can you give me some python code to quickly finetune a model on my sentiment analysis dataset?",
|
67 |
],
|
68 |
inputs=message,
|
|
|
78 |
state = gr.State()
|
79 |
agent_state = gr.State()
|
80 |
|
81 |
+
submit.click(chat, inputs=[message, state, doc_source], outputs=[chatbot, state])
|
82 |
+
message.submit(chat, inputs=[message, state, doc_source], outputs=[chatbot, state])
|
83 |
|
84 |
|
85 |
block.launch(debug=True)
|
buster/buster.py
CHANGED
@@ -1,12 +1,12 @@
|
|
1 |
import logging
|
2 |
from dataclasses import dataclass, field
|
|
|
3 |
|
4 |
import numpy as np
|
5 |
import pandas as pd
|
6 |
from openai.embeddings_utils import cosine_similarity, get_embedding
|
7 |
|
8 |
from buster.completers import get_completer
|
9 |
-
from buster.documents import get_documents_manager_from_extension
|
10 |
from buster.formatter import (
|
11 |
Response,
|
12 |
ResponseFormatter,
|
@@ -33,6 +33,7 @@ class BusterConfig:
|
|
33 |
unknown_prompt: Prompt to use to generate the "I don't know" embedding to compare to.
|
34 |
text_before_prompt: Text to prompt GPT with before the user prompt, but after the documentation.
|
35 |
reponse_footnote: Generic response to add the the chatbot's reply.
|
|
|
36 |
"""
|
37 |
|
38 |
documents_file: str = "buster/data/document_embeddings.tar.gz"
|
@@ -60,34 +61,45 @@ class BusterConfig:
|
|
60 |
response_format: str = "slack"
|
61 |
unknown_prompt: str = "I Don't know how to answer your question."
|
62 |
response_footnote: str = "I'm a bot 🤖 and not always perfect."
|
|
|
|
|
|
|
|
|
63 |
|
64 |
|
65 |
class Buster:
|
66 |
-
def __init__(self, cfg: BusterConfig):
|
67 |
-
|
68 |
self.cfg = cfg
|
69 |
-
self.
|
70 |
-
|
71 |
-
self.
|
72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
-
def
|
|
|
|
|
|
|
|
|
|
|
75 |
self.response_formatter = response_formatter_factory(
|
76 |
format=self.cfg.response_format, response_footnote=self.cfg.response_footnote
|
77 |
)
|
|
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
logger.info(
|
82 |
-
|
83 |
-
logger.info(f"embeddings loaded.")
|
84 |
-
|
85 |
-
def _init_unk_embedding(self):
|
86 |
-
logger.info("Generating UNK embedding...")
|
87 |
-
self.unk_embedding = get_embedding(
|
88 |
-
self.cfg.unknown_prompt,
|
89 |
-
engine=self.cfg.embedding_model,
|
90 |
-
)
|
91 |
|
92 |
def rank_documents(
|
93 |
self,
|
@@ -95,16 +107,17 @@ class Buster:
|
|
95 |
top_k: float,
|
96 |
thresh: float,
|
97 |
engine: str,
|
|
|
98 |
) -> pd.DataFrame:
|
99 |
"""
|
100 |
Compare the question to the series of documents and return the best matching documents.
|
101 |
"""
|
102 |
|
103 |
-
query_embedding = get_embedding(
|
104 |
query,
|
105 |
engine=engine,
|
106 |
)
|
107 |
-
matched_documents = self.documents.retrieve(query_embedding, top_k)
|
108 |
|
109 |
# log matched_documents to the console
|
110 |
logger.info(f"matched documents before thresh: {matched_documents}")
|
@@ -119,7 +132,9 @@ class Buster:
|
|
119 |
def prepare_documents(self, matched_documents: pd.DataFrame, max_words: int) -> str:
|
120 |
# gather the documents in one large plaintext variable
|
121 |
documents_list = matched_documents.content.to_list()
|
122 |
-
documents_str = "
|
|
|
|
|
123 |
|
124 |
# truncate the documents to fit
|
125 |
# TODO: increase to actual token count
|
@@ -135,11 +150,13 @@ class Buster:
|
|
135 |
self,
|
136 |
response,
|
137 |
matched_documents: pd.DataFrame,
|
138 |
-
unknown_prompt: str,
|
139 |
):
|
140 |
logger.info(f"GPT Response:\n{response.text}")
|
141 |
sources = (
|
142 |
-
Source(
|
|
|
|
|
|
|
143 |
)
|
144 |
|
145 |
return sources
|
@@ -154,7 +171,7 @@ class Buster:
|
|
154 |
|
155 |
set the unk_threshold to 0 to essentially turn off this feature.
|
156 |
"""
|
157 |
-
response_embedding = get_embedding(
|
158 |
completion,
|
159 |
engine=engine,
|
160 |
)
|
@@ -180,17 +197,18 @@ class Buster:
|
|
180 |
top_k=self.cfg.top_k,
|
181 |
thresh=self.cfg.thresh,
|
182 |
engine=self.cfg.embedding_model,
|
|
|
183 |
)
|
184 |
|
185 |
if len(matched_documents) == 0:
|
186 |
-
response = Response(
|
187 |
sources = tuple()
|
188 |
return self.response_formatter(response, sources)
|
189 |
|
190 |
# generate a completion
|
191 |
documents: str = self.prepare_documents(matched_documents, max_words=self.cfg.max_words)
|
192 |
-
response = self.completer.generate_response(user_input, documents)
|
193 |
-
sources = self.add_sources(response, matched_documents
|
194 |
|
195 |
# check for relevance
|
196 |
relevant = self.check_response_relevance(
|
|
|
1 |
import logging
|
2 |
from dataclasses import dataclass, field
|
3 |
+
from functools import lru_cache
|
4 |
|
5 |
import numpy as np
|
6 |
import pandas as pd
|
7 |
from openai.embeddings_utils import cosine_similarity, get_embedding
|
8 |
|
9 |
from buster.completers import get_completer
|
|
|
10 |
from buster.formatter import (
|
11 |
Response,
|
12 |
ResponseFormatter,
|
|
|
33 |
unknown_prompt: Prompt to use to generate the "I don't know" embedding to compare to.
|
34 |
text_before_prompt: Text to prompt GPT with before the user prompt, but after the documentation.
|
35 |
reponse_footnote: Generic response to add the the chatbot's reply.
|
36 |
+
source: the source of the document to consider
|
37 |
"""
|
38 |
|
39 |
documents_file: str = "buster/data/document_embeddings.tar.gz"
|
|
|
61 |
response_format: str = "slack"
|
62 |
unknown_prompt: str = "I Don't know how to answer your question."
|
63 |
response_footnote: str = "I'm a bot 🤖 and not always perfect."
|
64 |
+
source: str = ""
|
65 |
+
|
66 |
+
|
67 |
+
from buster.documents.base import DocumentsManager
|
68 |
|
69 |
|
70 |
class Buster:
|
71 |
+
def __init__(self, cfg: BusterConfig, documents: DocumentsManager):
|
72 |
+
self._unk_embedding = None
|
73 |
self.cfg = cfg
|
74 |
+
self.update_cfg(cfg)
|
75 |
+
|
76 |
+
self.documents = documents
|
77 |
+
|
78 |
+
@property
|
79 |
+
def unk_embedding(self):
|
80 |
+
return self._unk_embedding
|
81 |
+
|
82 |
+
@unk_embedding.setter
|
83 |
+
def unk_embedding(self, embedding):
|
84 |
+
logger.info("Setting new UNK embedding...")
|
85 |
+
self._unk_embedding = embedding
|
86 |
+
return self._unk_embedding
|
87 |
|
88 |
+
def update_cfg(self, cfg: BusterConfig):
|
89 |
+
"""Every time we set a new config, we update the things that need to be updated."""
|
90 |
+
logger.info(f"Updating config to {cfg.source}:\n{cfg}")
|
91 |
+
self.cfg = cfg
|
92 |
+
self.completer = get_completer(cfg.completer_cfg)
|
93 |
+
self.unk_embedding = self.get_embedding(self.cfg.unknown_prompt, engine=self.cfg.embedding_model)
|
94 |
self.response_formatter = response_formatter_factory(
|
95 |
format=self.cfg.response_format, response_footnote=self.cfg.response_footnote
|
96 |
)
|
97 |
+
logger.info(f"Config Updated.")
|
98 |
|
99 |
+
@lru_cache
|
100 |
+
def get_embedding(self, query: str, engine: str):
|
101 |
+
logger.info("generating embedding")
|
102 |
+
return get_embedding(query, engine=engine)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
103 |
|
104 |
def rank_documents(
|
105 |
self,
|
|
|
107 |
top_k: float,
|
108 |
thresh: float,
|
109 |
engine: str,
|
110 |
+
source: str,
|
111 |
) -> pd.DataFrame:
|
112 |
"""
|
113 |
Compare the question to the series of documents and return the best matching documents.
|
114 |
"""
|
115 |
|
116 |
+
query_embedding = self.get_embedding(
|
117 |
query,
|
118 |
engine=engine,
|
119 |
)
|
120 |
+
matched_documents = self.documents.retrieve(query_embedding, top_k=top_k, source=source)
|
121 |
|
122 |
# log matched_documents to the console
|
123 |
logger.info(f"matched documents before thresh: {matched_documents}")
|
|
|
132 |
def prepare_documents(self, matched_documents: pd.DataFrame, max_words: int) -> str:
|
133 |
# gather the documents in one large plaintext variable
|
134 |
documents_list = matched_documents.content.to_list()
|
135 |
+
documents_str = ""
|
136 |
+
for idx, doc in enumerate(documents_list):
|
137 |
+
documents_str += f"<DOCUMENT> {doc} <\DOCUMENT>"
|
138 |
|
139 |
# truncate the documents to fit
|
140 |
# TODO: increase to actual token count
|
|
|
150 |
self,
|
151 |
response,
|
152 |
matched_documents: pd.DataFrame,
|
|
|
153 |
):
|
154 |
logger.info(f"GPT Response:\n{response.text}")
|
155 |
sources = (
|
156 |
+
Source(
|
157 |
+
source=dct["source"], title=dct["title"], url=dct["url"], question_similarity=dct["similarity"] * 100
|
158 |
+
)
|
159 |
+
for dct in matched_documents.to_dict(orient="records")
|
160 |
)
|
161 |
|
162 |
return sources
|
|
|
171 |
|
172 |
set the unk_threshold to 0 to essentially turn off this feature.
|
173 |
"""
|
174 |
+
response_embedding = self.get_embedding(
|
175 |
completion,
|
176 |
engine=engine,
|
177 |
)
|
|
|
197 |
top_k=self.cfg.top_k,
|
198 |
thresh=self.cfg.thresh,
|
199 |
engine=self.cfg.embedding_model,
|
200 |
+
source=self.cfg.source,
|
201 |
)
|
202 |
|
203 |
if len(matched_documents) == 0:
|
204 |
+
response = Response(self.cfg.unknown_prompt)
|
205 |
sources = tuple()
|
206 |
return self.response_formatter(response, sources)
|
207 |
|
208 |
# generate a completion
|
209 |
documents: str = self.prepare_documents(matched_documents, max_words=self.cfg.max_words)
|
210 |
+
response: Response = self.completer.generate_response(user_input, documents)
|
211 |
+
sources = self.add_sources(response, matched_documents)
|
212 |
|
213 |
# check for relevance
|
214 |
relevant = self.check_response_relevance(
|
buster/data/documents.db
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:b86c2b4f5a2ec410c2b9132ed62213528ba10c0dc260162f689e30ba677815f1
|
3 |
-
size 244338688
|
|
|
|
|
|
|
|
buster/documents/utils.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
import os
|
|
|
2 |
from typing import Type
|
3 |
|
4 |
from buster.documents.base import DocumentsManager
|
@@ -12,6 +13,18 @@ def get_file_extension(filepath: str) -> str:
|
|
12 |
return os.path.splitext(filepath)[1]
|
13 |
|
14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
def get_documents_manager_from_extension(filepath: str) -> Type[DocumentsManager]:
|
16 |
ext = get_file_extension(filepath)
|
17 |
|
|
|
1 |
import os
|
2 |
+
import urllib.request
|
3 |
from typing import Type
|
4 |
|
5 |
from buster.documents.base import DocumentsManager
|
|
|
13 |
return os.path.splitext(filepath)[1]
|
14 |
|
15 |
|
16 |
+
def download_db(db_url: str, output_dir: str):
|
17 |
+
os.makedirs(output_dir, exist_ok=True)
|
18 |
+
fname = os.path.join(output_dir, "documents.db")
|
19 |
+
if not os.path.exists(fname):
|
20 |
+
print(f"Downloading db file from {db_url} to {fname}...")
|
21 |
+
urllib.request.urlretrieve(db_url, fname)
|
22 |
+
print("Downloaded.")
|
23 |
+
else:
|
24 |
+
print("File already exists. Skipping.")
|
25 |
+
return fname
|
26 |
+
|
27 |
+
|
28 |
def get_documents_manager_from_extension(filepath: str) -> Type[DocumentsManager]:
|
29 |
ext = get_file_extension(filepath)
|
30 |
|
buster/formatter/base.py
CHANGED
@@ -4,9 +4,10 @@ from typing import Iterable, NamedTuple
|
|
4 |
|
5 |
# Should be from the `documents` module.
|
6 |
class Source(NamedTuple):
|
7 |
-
|
8 |
url: str
|
9 |
question_similarity: float
|
|
|
10 |
# TODO Add answer similarity.
|
11 |
# answer_similarity: float
|
12 |
|
@@ -22,7 +23,7 @@ class Response:
|
|
22 |
@dataclass
|
23 |
class ResponseFormatter:
|
24 |
response_footnote: str
|
25 |
-
source_template: str = "{source.name} (relevance: {source.question_similarity:2.
|
26 |
error_msg_template: str = """Something went wrong:\n{response.error_msg}"""
|
27 |
error_fallback_template: str = "Something went very wrong."
|
28 |
sourced_answer_template: str = (
|
|
|
4 |
|
5 |
# Should be from the `documents` module.
|
6 |
class Source(NamedTuple):
|
7 |
+
title: str
|
8 |
url: str
|
9 |
question_similarity: float
|
10 |
+
source: str = ""
|
11 |
# TODO Add answer similarity.
|
12 |
# answer_similarity: float
|
13 |
|
|
|
23 |
@dataclass
|
24 |
class ResponseFormatter:
|
25 |
response_footnote: str
|
26 |
+
source_template: str = "{source.name} (relevance: {source.question_similarity:2.1f})"
|
27 |
error_msg_template: str = """Something went wrong:\n{response.error_msg}"""
|
28 |
error_fallback_template: str = "Something went very wrong."
|
29 |
sourced_answer_template: str = (
|
buster/formatter/gradio.py
CHANGED
@@ -17,7 +17,7 @@ class GradioResponseFormatter(ResponseFormatter):
|
|
17 |
"""{footnote}"""
|
18 |
)
|
19 |
unsourced_answer_template: str = "{response.text}<br><br>{footnote}"
|
20 |
-
source_template: str = """[🔗 {source.
|
21 |
|
22 |
def sources_list(self, sources: Iterable[Source]) -> str | None:
|
23 |
"""Format sources into a list."""
|
|
|
17 |
"""{footnote}"""
|
18 |
)
|
19 |
unsourced_answer_template: str = "{response.text}<br><br>{footnote}"
|
20 |
+
source_template: str = """[🔗 {source.title}]({source.url}), relevance: {source.question_similarity:2.1f} %"""
|
21 |
|
22 |
def sources_list(self, sources: Iterable[Source]) -> str | None:
|
23 |
"""Format sources into a list."""
|
buster/formatter/html.py
CHANGED
@@ -37,5 +37,5 @@ class HTMLResponseFormatter(ResponseFormatter):
|
|
37 |
response.error,
|
38 |
html.escape(response.error_msg) if response.error_msg else response.error_msg,
|
39 |
)
|
40 |
-
sources = (Source(html.escape(source.
|
41 |
return super().__call__(response, sources)
|
|
|
37 |
response.error,
|
38 |
html.escape(response.error_msg) if response.error_msg else response.error_msg,
|
39 |
)
|
40 |
+
sources = (Source(html.escape(source.title), source.url, source.question_similarity) for source in sources)
|
41 |
return super().__call__(response, sources)
|
buster/formatter/markdown.py
CHANGED
@@ -8,7 +8,7 @@ from buster.formatter.base import ResponseFormatter, Source
|
|
8 |
class MarkdownResponseFormatter(ResponseFormatter):
|
9 |
"""Format the answer in markdown."""
|
10 |
|
11 |
-
source_template: str = """[🔗 {source.
|
12 |
|
13 |
def sources_list(self, sources: Iterable[Source]) -> str | None:
|
14 |
"""Format sources into a list."""
|
|
|
8 |
class MarkdownResponseFormatter(ResponseFormatter):
|
9 |
"""Format the answer in markdown."""
|
10 |
|
11 |
+
source_template: str = """[🔗 {source.title}]({source.url}), relevance: {source.question_similarity:2.3f}"""
|
12 |
|
13 |
def sources_list(self, sources: Iterable[Source]) -> str | None:
|
14 |
"""Format sources into a list."""
|
buster/formatter/slack.py
CHANGED
@@ -8,7 +8,7 @@ from buster.formatter import ResponseFormatter, Source
|
|
8 |
class SlackResponseFormatter(ResponseFormatter):
|
9 |
"""Format the answer for Slack."""
|
10 |
|
11 |
-
source_template: str = """<{source.url}|🔗 {source.
|
12 |
|
13 |
def sources_list(self, sources: Iterable[Source]) -> str | None:
|
14 |
"""Format sources into a list."""
|
|
|
8 |
class SlackResponseFormatter(ResponseFormatter):
|
9 |
"""Format the answer for Slack."""
|
10 |
|
11 |
+
source_template: str = """<{source.url}|🔗 {source.title}>, relevance: {source.question_similarity:2.3f}"""
|
12 |
|
13 |
def sources_list(self, sources: Iterable[Source]) -> str | None:
|
14 |
"""Format sources into a list."""
|
pyproject.toml
CHANGED
@@ -18,3 +18,7 @@ profile = "black"
|
|
18 |
|
19 |
[tool.black]
|
20 |
line-length = 120
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
[tool.black]
|
20 |
line-length = 120
|
21 |
+
|
22 |
+
[tool.pytest.ini_options]
|
23 |
+
log_cli = true
|
24 |
+
log_cli_level = "INFO"
|
tests/test_chatbot.py
CHANGED
@@ -5,7 +5,9 @@ import numpy as np
|
|
5 |
import pandas as pd
|
6 |
|
7 |
from buster.buster import Buster, BusterConfig
|
8 |
-
from buster.
|
|
|
|
|
9 |
|
10 |
TEST_DATA_DIR = Path(__file__).resolve().parent / "data"
|
11 |
DOCUMENTS_FILE = os.path.join(str(TEST_DATA_DIR), "document_embeddings_huggingface_subset.tar.gz")
|
@@ -16,6 +18,17 @@ def get_fake_embedding(length=1536):
|
|
16 |
return list(rng.random(length, dtype=np.float32))
|
17 |
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
class DocumentsMock(DocumentsManager):
|
20 |
def __init__(self, filepath):
|
21 |
self.filepath = filepath
|
@@ -39,20 +52,24 @@ class DocumentsMock(DocumentsManager):
|
|
39 |
return self.documents
|
40 |
|
41 |
|
|
|
|
|
|
|
|
|
|
|
42 |
def test_chatbot_mock_data(tmp_path, monkeypatch):
|
43 |
gpt_expected_answer = "this is GPT answer"
|
44 |
-
monkeypatch.setattr("
|
45 |
-
monkeypatch.setattr("buster.buster.
|
46 |
-
monkeypatch.setattr("openai.Completion.create", lambda **kwargs: {"choices": [{"text": gpt_expected_answer}]})
|
47 |
|
48 |
hf_transformers_cfg = BusterConfig(
|
49 |
-
documents_file=tmp_path / "not_a_real_file.tar.gz",
|
50 |
unknown_prompt="This doesn't seem to be related to the huggingface library. I am not sure how to answer.",
|
51 |
embedding_model="text-embedding-ada-002",
|
52 |
top_k=3,
|
53 |
-
thresh=0
|
54 |
max_words=3000,
|
55 |
response_format="slack",
|
|
|
56 |
completer_cfg={
|
57 |
"name": "GPT3",
|
58 |
"text_before_prompt": (
|
@@ -72,7 +89,9 @@ def test_chatbot_mock_data(tmp_path, monkeypatch):
|
|
72 |
},
|
73 |
},
|
74 |
)
|
75 |
-
|
|
|
|
|
76 |
answer = buster.process_input("What is a transformer?")
|
77 |
assert isinstance(answer, str)
|
78 |
assert answer.startswith(gpt_expected_answer)
|
@@ -80,7 +99,6 @@ def test_chatbot_mock_data(tmp_path, monkeypatch):
|
|
80 |
|
81 |
def test_chatbot_real_data__chatGPT():
|
82 |
hf_transformers_cfg = BusterConfig(
|
83 |
-
documents_file=DOCUMENTS_FILE,
|
84 |
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
85 |
embedding_model="text-embedding-ada-002",
|
86 |
top_k=3,
|
@@ -101,14 +119,14 @@ def test_chatbot_real_data__chatGPT():
|
|
101 |
},
|
102 |
},
|
103 |
)
|
104 |
-
|
|
|
105 |
answer = buster.process_input("What is a transformer?")
|
106 |
assert isinstance(answer, str)
|
107 |
|
108 |
|
109 |
def test_chatbot_real_data__chatGPT_OOD():
|
110 |
buster_cfg = BusterConfig(
|
111 |
-
documents_file=DOCUMENTS_FILE,
|
112 |
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
113 |
embedding_model="text-embedding-ada-002",
|
114 |
top_k=3,
|
@@ -122,7 +140,7 @@ def test_chatbot_real_data__chatGPT_OOD():
|
|
122 |
"""Do not include any links to urls or hyperlinks in your answers. """
|
123 |
"""If you do not know the answer to a question, or if it is completely irrelevant to the library usage, let the user know you cannot answer. """
|
124 |
"""Use this response: """
|
125 |
-
"""I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"""
|
126 |
"""For example:\n"""
|
127 |
"""What is the meaning of life for huggingface?\n"""
|
128 |
"""I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"""
|
@@ -135,7 +153,8 @@ def test_chatbot_real_data__chatGPT_OOD():
|
|
135 |
},
|
136 |
response_format="gradio",
|
137 |
)
|
138 |
-
|
|
|
139 |
answer = buster.process_input("What is a good recipe for brocolli soup?")
|
140 |
assert isinstance(answer, str)
|
141 |
assert buster_cfg.unknown_prompt in answer
|
@@ -143,7 +162,6 @@ def test_chatbot_real_data__chatGPT_OOD():
|
|
143 |
|
144 |
def test_chatbot_real_data__GPT():
|
145 |
hf_transformers_cfg = BusterConfig(
|
146 |
-
documents_file=DOCUMENTS_FILE,
|
147 |
unknown_prompt="This doesn't seem to be related to the huggingface library. I am not sure how to answer.",
|
148 |
embedding_model="text-embedding-ada-002",
|
149 |
top_k=3,
|
@@ -169,6 +187,7 @@ def test_chatbot_real_data__GPT():
|
|
169 |
},
|
170 |
},
|
171 |
)
|
172 |
-
|
|
|
173 |
answer = buster.process_input("What is a transformer?")
|
174 |
assert isinstance(answer, str)
|
|
|
5 |
import pandas as pd
|
6 |
|
7 |
from buster.buster import Buster, BusterConfig
|
8 |
+
from buster.completers.base import Completer
|
9 |
+
from buster.documents import DocumentsManager, get_documents_manager_from_extension
|
10 |
+
from buster.formatter.base import Response
|
11 |
|
12 |
TEST_DATA_DIR = Path(__file__).resolve().parent / "data"
|
13 |
DOCUMENTS_FILE = os.path.join(str(TEST_DATA_DIR), "document_embeddings_huggingface_subset.tar.gz")
|
|
|
18 |
return list(rng.random(length, dtype=np.float32))
|
19 |
|
20 |
|
21 |
+
class MockCompleter(Completer):
|
22 |
+
def __init__(self, expected_answer):
|
23 |
+
self.expected_answer = expected_answer
|
24 |
+
|
25 |
+
def complete(self):
|
26 |
+
return
|
27 |
+
|
28 |
+
def generate_response(self, user_input, documents) -> Response:
|
29 |
+
return Response(self.expected_answer)
|
30 |
+
|
31 |
+
|
32 |
class DocumentsMock(DocumentsManager):
|
33 |
def __init__(self, filepath):
|
34 |
self.filepath = filepath
|
|
|
52 |
return self.documents
|
53 |
|
54 |
|
55 |
+
import logging
|
56 |
+
|
57 |
+
logging.basicConfig(level=logging.INFO)
|
58 |
+
|
59 |
+
|
60 |
def test_chatbot_mock_data(tmp_path, monkeypatch):
|
61 |
gpt_expected_answer = "this is GPT answer"
|
62 |
+
monkeypatch.setattr(Buster, "get_embedding", lambda self, prompt, engine: get_fake_embedding())
|
63 |
+
monkeypatch.setattr("buster.buster.get_completer", lambda x: MockCompleter(expected_answer=gpt_expected_answer))
|
|
|
64 |
|
65 |
hf_transformers_cfg = BusterConfig(
|
|
|
66 |
unknown_prompt="This doesn't seem to be related to the huggingface library. I am not sure how to answer.",
|
67 |
embedding_model="text-embedding-ada-002",
|
68 |
top_k=3,
|
69 |
+
thresh=0,
|
70 |
max_words=3000,
|
71 |
response_format="slack",
|
72 |
+
source="fake source",
|
73 |
completer_cfg={
|
74 |
"name": "GPT3",
|
75 |
"text_before_prompt": (
|
|
|
89 |
},
|
90 |
},
|
91 |
)
|
92 |
+
filepath = tmp_path / "not_a_real_file.tar.gz"
|
93 |
+
documents = DocumentsMock(filepath)
|
94 |
+
buster = Buster(cfg=hf_transformers_cfg, documents=documents)
|
95 |
answer = buster.process_input("What is a transformer?")
|
96 |
assert isinstance(answer, str)
|
97 |
assert answer.startswith(gpt_expected_answer)
|
|
|
99 |
|
100 |
def test_chatbot_real_data__chatGPT():
|
101 |
hf_transformers_cfg = BusterConfig(
|
|
|
102 |
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
103 |
embedding_model="text-embedding-ada-002",
|
104 |
top_k=3,
|
|
|
119 |
},
|
120 |
},
|
121 |
)
|
122 |
+
documents = get_documents_manager_from_extension(DOCUMENTS_FILE)(DOCUMENTS_FILE)
|
123 |
+
buster = Buster(cfg=hf_transformers_cfg, documents=documents)
|
124 |
answer = buster.process_input("What is a transformer?")
|
125 |
assert isinstance(answer, str)
|
126 |
|
127 |
|
128 |
def test_chatbot_real_data__chatGPT_OOD():
|
129 |
buster_cfg = BusterConfig(
|
|
|
130 |
unknown_prompt="I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?",
|
131 |
embedding_model="text-embedding-ada-002",
|
132 |
top_k=3,
|
|
|
140 |
"""Do not include any links to urls or hyperlinks in your answers. """
|
141 |
"""If you do not know the answer to a question, or if it is completely irrelevant to the library usage, let the user know you cannot answer. """
|
142 |
"""Use this response: """
|
143 |
+
"""'I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?'\n"""
|
144 |
"""For example:\n"""
|
145 |
"""What is the meaning of life for huggingface?\n"""
|
146 |
"""I'm sorry, but I am an AI language model trained to assist with questions related to the huggingface transformers library. I cannot answer that question as it is not relevant to the library or its usage. Is there anything else I can assist you with?"""
|
|
|
153 |
},
|
154 |
response_format="gradio",
|
155 |
)
|
156 |
+
documents = get_documents_manager_from_extension(DOCUMENTS_FILE)(DOCUMENTS_FILE)
|
157 |
+
buster = Buster(cfg=buster_cfg, documents=documents)
|
158 |
answer = buster.process_input("What is a good recipe for brocolli soup?")
|
159 |
assert isinstance(answer, str)
|
160 |
assert buster_cfg.unknown_prompt in answer
|
|
|
162 |
|
163 |
def test_chatbot_real_data__GPT():
|
164 |
hf_transformers_cfg = BusterConfig(
|
|
|
165 |
unknown_prompt="This doesn't seem to be related to the huggingface library. I am not sure how to answer.",
|
166 |
embedding_model="text-embedding-ada-002",
|
167 |
top_k=3,
|
|
|
187 |
},
|
188 |
},
|
189 |
)
|
190 |
+
documents = get_documents_manager_from_extension(DOCUMENTS_FILE)(DOCUMENTS_FILE)
|
191 |
+
buster = Buster(cfg=hf_transformers_cfg, documents=documents)
|
192 |
answer = buster.process_input("What is a transformer?")
|
193 |
assert isinstance(answer, str)
|