File size: 6,536 Bytes
8f17b3d
 
 
 
 
 
 
 
 
 
 
 
 
0102b8a
8f17b3d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0d13133
9f3ddbf
8f17b3d
 
 
 
 
 
 
 
 
d35069c
 
0102b8a
fc8de1e
0102b8a
 
 
d35069c
8f17b3d
 
 
 
 
 
 
 
0b76a3e
b3c801a
0b76a3e
b3c801a
0b76a3e
b3c801a
0b76a3e
b3c801a
0b76a3e
b3c801a
0b76a3e
 
 
 
 
 
98a9e92
8f17b3d
 
b3c801a
8f17b3d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0d13133
 
aa1f83e
 
 
 
 
 
 
 
6c3e183
 
aa1f83e
 
 
 
 
 
 
 
 
 
0b76a3e
aa1f83e
 
 
 
 
 
 
 
8663794
aa1f83e
0102b8a
aa1f83e
 
 
d35069c
aa1f83e
 
 
 
 
 
 
 
 
 
 
0102b8a
 
 
 
 
 
 
 
 
 
 
6c3e183
8f17b3d
 
 
 
0d13133
8f17b3d
 
 
 
 
 
0d13133
8f17b3d
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
"""
Credit to Derek Thomas, derek@huggingface.co
"""
import os
import logging
from pathlib import Path
from time import perf_counter

import gradio as gr
from jinja2 import Environment, FileSystemLoader

from backend.query_llm import generate_hf, generate_openai
from backend.semantic_search import retrieve
from backend.reranker import rerank_documents


TOP_K = int(os.getenv("TOP_K", 4))

proj_dir = Path(__file__).parent
# Setting up the logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Set up the template environment with the templates directory
env = Environment(loader=FileSystemLoader(proj_dir / 'templates'))

# Load the templates directly from the environment
template = env.get_template('template.j2')
template_html = env.get_template('template_html.j2')


def add_text(history, text):
    history = [] if history is None else history
    history = history + [(text, None)]
    return history, gr.Textbox(value="", interactive=False)


def bot(history, chunk_table, embedding_model, llm_model, cross_encoder, top_k_param, rerank_topk ):
    top_k_param = int(top_k_param)
    query = history[-1][0]

    if not query:
        raise gr.Warning("Please submit a non-empty string as a prompt")

    logger.info('Retrieving documents...')
    # Retrieve documents relevant to query
    document_start = perf_counter()

    #documents = retrieve(query, TOP_K)
    documents = retrieve(query, top_k_param, chunk_table, embedding_model)
    if cross_encoder != "None" and len(documents) > 1:
        documents = rerank_documents(cross_encoder, documents, query, top_k_rerank=rerank_topk)
        #"cross-encoder/ms-marco-MiniLM-L-6-v2"




    document_time = perf_counter() - document_start
    logger.info(f'Finished Retrieving documents in {round(document_time, 2)} seconds...')

    # Create Prompt
    prompt = template.render(documents=documents, query=query)
    prompt_html = template_html.render(documents=documents, query=query)

    if llm_model == "mistralai/Mistral-7B-Instruct-v0.2":
        generate_fn = generate_hf
    if llm_model == "mistralai/Mistral-7B-v0.1":
        generate_fn = generate_hf
    if llm_model == "mistralai/Mixtral-8x7B-Instruct-v0.1":
        generate_fn = generate_hf 
    if llm_model == "gpt-3.5-turbo":
        generate_fn = generate_openai
    if llm_model == "gpt-4-turbo-preview":
        generate_fn = generate_openai

    #if api_kind == "HuggingFace":
    #     generate_fn = generate_hf
    #elif api_kind == "OpenAI":
    #     generate_fn = generate_openai
    #else:
    #     raise gr.Error(f"API {api_kind} is not supported")

    history[-1][1] = ""
    for character in generate_fn(prompt, history[:-1], llm_model):
        history[-1][1] = character
        yield history, prompt_html


with gr.Blocks() as demo:
    chatbot = gr.Chatbot(
            [],
            elem_id="chatbot",
            avatar_images=('https://aui.atlassian.com/aui/8.8/docs/images/avatar-person.svg',
                           'https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.svg'),
            bubble_full_width=False,
            show_copy_button=True,
            show_share_button=True,
            )

    with gr.Row():
        txt = gr.Textbox(
                scale=3,
                show_label=False,
                placeholder="Enter text and press enter",
                container=False,
                )
        txt_btn = gr.Button(value="Submit text", scale=1)

    #api_kind = gr.Radio(choices=["HuggingFace",
    #                             "OpenAI"], value="HuggingFace")
    
    chunk_table = gr.Radio(choices=["BGE_CharacterTextSplitter", 
                                    "BGE_FixedSizeSplitter",
                                    "BGE_RecursiveCharacterTextSplitter",
                                    "MiniLM_CharacterTextSplitter", 
                                    "MiniLM_FixedSizeSplitter",
                                    "MiniLM_RecursiveCharacterSplitter"
                                    ], 
                                    value="MiniLM_CharacterTextSplitter",
                                    label="Chunk table")
    embedding_model = gr.Radio(
                choices=[
                    "BAAI/bge-large-en-v1.5",
                    "sentence-transformers/all-MiniLM-L6-v2",
                ],
                value="sentence-transformers/all-MiniLM-L6-v2",
                label='Embedding model'
            )
    llm_model = gr.Radio(
                choices=[
                    "mistralai/Mistral-7B-Instruct-v0.2",
                    "gpt-3.5-turbo",
                    "gpt-4-turbo-preview",
                    "mistralai/Mistral-7B-v0.1",
                    "mistralai/Mixtral-8x7B-Instruct-v0.1"
                ],
                value="gpt-3.5-turbo",
                label='LLM'
            )    
    cross_encoder = gr.Radio(
                choices=[
                    "None",
                    "BAAI/bge-reranker-large",
                    "cross-encoder/ms-marco-MiniLM-L-6-v2",
                ],
                value="None",
                label='Cross-encoder model'
            )
    top_k_param = gr.Radio(
                choices=[
                    "5",
                    "10",
                    "20",
                    "50",
                ],
                value="5",
                label='top-K'
            )
    rerank_topk = gr.Radio(
                choices=[
                    "5",
                    "10",
                    "20",
                    "50",
                ],
                value="5",
                label='rerank-top-K'
            )  


    prompt_html = gr.HTML()
    # Turn off interactivity while generating if you click
    txt_msg = txt_btn.click(add_text, [chatbot, txt], [chatbot, txt], queue=False).then(
            bot, [chatbot, chunk_table, embedding_model, llm_model, cross_encoder, top_k_param, rerank_topk], [chatbot, prompt_html])

    # Turn it back on
    txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False)

    # Turn off interactivity while generating if you hit enter
    txt_msg = txt.submit(add_text, [chatbot, txt], [chatbot, txt], queue=False).then(
            bot, [chatbot, chunk_table, embedding_model, llm_model, cross_encoder, top_k_param, rerank_topk], [chatbot, prompt_html])

    # Turn it back on
    txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False)

demo.queue()
demo.launch(debug=True)