File size: 11,348 Bytes
6ad384a
 
c1a3363
5d658bb
b7fb9f7
c1a3363
 
 
 
 
 
 
 
 
 
 
34bed84
c1a3363
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3943635
 
b7fb9f7
 
 
3943635
b7fb9f7
3943635
 
c1a3363
 
 
 
 
 
 
 
 
 
 
3fa5f3b
c1a3363
 
 
 
 
 
 
 
 
8326015
c1a3363
 
 
 
 
 
3fa5f3b
 
c1a3363
 
 
 
 
 
 
f834e93
c1a3363
 
3943635
 
c1a3363
 
 
 
 
f834e93
b27ddbb
b4efea2
f834e93
 
 
c42f5d4
17d7c3a
f834e93
 
 
 
 
 
c1a3363
 
 
 
f834e93
c1a3363
 
 
 
 
 
 
 
 
 
 
 
 
9838768
 
c1a3363
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9838768
 
 
 
 
 
 
c1a3363
9838768
c1a3363
 
 
 
 
9838768
 
 
c1a3363
 
 
 
 
 
 
 
9838768
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c1a3363
 
9838768
 
 
c9a9aee
9838768
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
"Single Thread"

import os
import multiprocessing
import concurrent.futures
from langchain.document_loaders import TextLoader, DirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import FAISS
from sentence_transformers import SentenceTransformer
import faiss
import torch
import numpy as np
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
from datetime import datetime
import json
import gradio as gr
import re 

class DocumentRetrievalAndGeneration:
    def __init__(self, embedding_model_name, lm_model_id, data_folder):
        self.all_splits = self.load_documents(data_folder)
        self.embeddings = SentenceTransformer(embedding_model_name)
        self.gpu_index = self.create_faiss_index()
        self.llm = self.initialize_llm(lm_model_id)

    def load_documents(self, folder_path):
        loader = DirectoryLoader(folder_path, loader_cls=TextLoader)
        documents = loader.load()
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=5000, chunk_overlap=250)
        all_splits = text_splitter.split_documents(documents)
        print('Length of documents:', len(documents))
        print("LEN of all_splits", len(all_splits))
        for i in range(5):
            print(all_splits[i].page_content)
        return all_splits

    def create_faiss_index(self):
        all_texts = [split.page_content for split in self.all_splits]
        embeddings = self.embeddings.encode(all_texts, convert_to_tensor=True).cpu().numpy()
        index = faiss.IndexFlatL2(embeddings.shape[1])
        index.add(embeddings)
        gpu_resource = faiss.StandardGpuResources()
        gpu_index = faiss.index_cpu_to_gpu(gpu_resource, 0, index)
        return gpu_index

    def initialize_llm(self, model_id):
        bnb_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type="nf4",
            bnb_4bit_compute_dtype=torch.bfloat16
        )
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config)
        tokenizer = AutoTokenizer.from_pretrained(model_id)
        generate_text = pipeline(
            model=model,
            tokenizer=tokenizer,
            return_full_text=True,
            task='text-generation',
            temperature=0.6,
            max_new_tokens=256,
        )
        return generate_text

    def generate_response_with_timeout(self, model_inputs):
        try:
            with concurrent.futures.ThreadPoolExecutor() as executor:
                future = executor.submit(self.llm.model.generate, model_inputs, max_new_tokens=1000, do_sample=True)
                generated_ids = future.result(timeout=60)  # Timeout set to 60 seconds
            return generated_ids
        except concurrent.futures.TimeoutError:
            raise TimeoutError("Text generation process timed out")
    
    def query_and_generate_response(self, query):
        query_embedding = self.embeddings.encode(query, convert_to_tensor=True).cpu().numpy()
        distances, indices = self.gpu_index.search(np.array([query_embedding]), k=5)

        content = ""
        for idx in indices[0]:
            content += "-" * 50 + "\n"
            content += self.all_splits[idx].page_content + "\n"
            print("CHUNK", idx)
            print(self.all_splits[idx].page_content)
            print("############################")
        prompt = f"""<s>
        You are a knowledgeable assistant with access to a comprehensive database. 
        I need you to answer my question and provide related information in a specific format.
        I have provided five relatable json files {content}, choose the most suitable chunks for answering the query
        Here's what I need:
        Include a final answer without additional comments, sign-offs, or extra phrases. Be direct and to the point.
        content
        Here's my question:
        Query:{query}
        Solution==>
        RETURN ONLY SOLUTION . IF THEIR IS NO ANSWER RELATABLE IN RETRIEVED CHUNKS , RETURN " NO SOLUTION AVAILABLE"
        Example1
        Query: "How to use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM",
        Solution: "To use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM, you need to modify the configuration file of the NDK application. Specifically, change the processor reference from 'A15_0' to 'IPU1_0'.",
        
        Example2
        Query: "Can BQ25896 support I2C interface?",
        Solution: "Yes, the BQ25896 charger supports the I2C interface for communication."
        </s>
        """
        # prompt = f"Query: {query}\nSolution: {content}\n"

        # Encode and prepare inputs
        messages = [{"role": "user", "content": prompt}]
        encodeds = self.llm.tokenizer.apply_chat_template(messages, return_tensors="pt")
        model_inputs = encodeds.to(self.llm.device)
        
        # Perform inference and measure time
        start_time = datetime.now()
        generated_ids = self.generate_response_with_timeout(model_inputs)
        # generated_ids = self.llm.model.generate(model_inputs, max_new_tokens=1000, do_sample=True)
        elapsed_time = datetime.now() - start_time

        # Decode and return output
        decoded = self.llm.tokenizer.batch_decode(generated_ids)
        generated_response = decoded[0]
        match1 = re.search(r'\[/INST\](.*?)</s>', generated_response, re.DOTALL)
        
        match2 = re.search(r'Solution:(.*?)</s>', generated_response, re.DOTALL | re.IGNORECASE)
        if match1:
            solution_text = match1.group(1).strip()
            print(solution_text)
            if "Solution:" in solution_text:
                solution_text = solution_text.split("Solution:", 1)[1].strip()
        elif match2:
            solution_text = match2.group(1).strip()
            print(solution_text)
            
        else:
            solution_text=generated_response
        print("Generated response:", generated_response)
        print("Time elapsed:", elapsed_time)
        print("Device in use:", self.llm.device)

        return solution_text, content

    def qa_infer_gradio(self, query):
        response = self.query_and_generate_response(query)
        return response

if __name__ == "__main__":
    # Example usage
    embedding_model_name = 'flax-sentence-embeddings/all_datasets_v3_MiniLM-L12'
    lm_model_id = "mistralai/Mistral-7B-Instruct-v0.2"
    data_folder = 'sample_embedding_folder2'

    doc_retrieval_gen = DocumentRetrievalAndGeneration(embedding_model_name, lm_model_id, data_folder)

    """Dual Interface"""
    
    def launch_interface():
        css_code = """
            .gradio-container {
                background-color: #daccdb;
            }
            /* Button styling for all buttons */
            button {
                background-color: #927fc7; /* Default color for all other buttons */
                color: black;
                border: 1px solid black;
                padding: 10px;
                margin-right: 10px;
                font-size: 16px; /* Increase font size */
                font-weight: bold; /* Make text bold */
            }
        """
        EXAMPLES = [
            "On which devices can the VIP and CSI2 modules operate simultaneously?", 
            "I'm using Code Composer Studio 5.4.0.00091 and enabled FPv4SPD16 floating point support for CortexM4 in TDA2. However, after building the project, the .asm file shows --float_support=vfplib instead of FPv4SPD16. Why is this happening?", 
            "Could you clarify the maximum number of cameras that can be connected simultaneously to the video input ports on the TDA2x SoC, considering it supports up to 10 multiplexed input ports and includes 3 dedicated video input modules?"
        ]
    
        file_path = "ticketNames.txt"
    
        # Read the file content
        with open(file_path, "r") as file:
            content = file.read()
        ticket_names = json.loads(content)
        dropdown = gr.Dropdown(label="Sample queries", choices=ticket_names)
    
        # Define Gradio interfaces
        tab1 = gr.Interface(
            fn=doc_retrieval_gen.qa_infer_gradio,
            inputs=[gr.Textbox(label="QUERY", placeholder="Enter your query here")],
            allow_flagging='never',
            examples=EXAMPLES,
            cache_examples=False,
            outputs=[gr.Textbox(label="SOLUTION"), gr.Textbox(label="RELATED QUERIES")],
            css=css_code
        )
        tab2 = gr.Interface(
            fn=doc_retrieval_gen.qa_infer_gradio,
            inputs=[dropdown],
            allow_flagging='never',
            outputs=[gr.Textbox(label="SOLUTION"), gr.Textbox(label="RELATED QUERIES")],
            css=css_code
        )
    
        # Combine interfaces into a tabbed interface
        gr.TabbedInterface(
            [tab1, tab2],
            ["Textbox Input", "FAQs"],
            title="TI E2E FORUM",
            css=css_code
        ).launch(debug=True)
    
    # Launch the interface
    launch_interface()



    """Single Interface"""
    # def launch_interface():
    #     css_code = """
    #         .gradio-container {
    #             background-color: #daccdb;
    #         }
    #         /* Button styling for all buttons */
    #         button {
    #             background-color: #927fc7; /* Default color for all other buttons */
    #             color: black;
    #             border: 1px solid black;
    #             padding: 10px;
    #             margin-right: 10px;
    #             font-size: 16px; /* Increase font size */
    #             font-weight: bold; /* Make text bold */
    #         }
    #         """
    #     EXAMPLES = ["On which devices can the VIP and CSI2 modules operate simultaneously? ", 
    #                 "I'm using Code Composer Studio 5.4.0.00091 and enabled FPv4SPD16 floating point support for CortexM4 in TDA2. However, after building the project, the .asm file shows --float_support=vfplib instead of FPv4SPD16. Why is this happening?", 
    #                 "Could you clarify the maximum number of cameras that can be connected simultaneously to the video input ports on the TDA2x SoC, considering it supports up to 10 multiplexed input ports and includes 3 dedicated video input modules?"]
        
    #     file_path = "ticketNames.txt"

    #     # Read the file content
    #     with open(file_path, "r") as file:
    #         content = file.read()
    #     ticket_names = json.loads(content)
    #     dropdown = gr.Dropdown(label="Sample queries", choices=ticket_names)
        
    #     # Define Gradio interface
    #     interface = gr.Interface(
    #         fn=doc_retrieval_gen.qa_infer_gradio,
    #         inputs=[gr.Textbox(label="QUERY", placeholder="Enter your query here")],
    #         allow_flagging='never',
    #         examples=EXAMPLES,
    #         cache_examples=False,
    #         outputs=[gr.Textbox(label="SOLUTION"), gr.Textbox(label="RELATED QUERIES")],
    #         css=css_code
    #     )

    #     # Launch Gradio interface
    #     interface.launch(debug=True)

    # # Launch the interface
    # launch_interface()