File size: 10,421 Bytes
a9c396e
 
 
 
 
 
 
 
 
17149e5
a9c396e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
17149e5
 
 
a9c396e
 
 
17149e5
 
 
 
 
 
 
 
 
 
 
 
 
a9c396e
 
057f5e7
a9c396e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
057f5e7
 
 
a9c396e
 
 
 
 
 
 
 
 
 
 
 
057f5e7
a9c396e
 
 
 
 
 
17149e5
a9c396e
 
 
17149e5
057f5e7
64987dd
057f5e7
 
a9c396e
 
 
 
 
 
057f5e7
a9c396e
057f5e7
 
 
 
a9c396e
 
057f5e7
a9c396e
 
057f5e7
a9c396e
057f5e7
 
a9c396e
 
057f5e7
 
 
 
 
 
17149e5
85c5bca
057f5e7
 
 
a9c396e
 
 
 
 
 
17149e5
 
85c5bca
 
 
17149e5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a9c396e
 
 
 
17149e5
 
 
 
 
 
a9c396e
 
 
17149e5
a9c396e
17149e5
a9c396e
 
 
 
 
 
 
057f5e7
a9c396e
057f5e7
a9c396e
85c5bca
17149e5
a9c396e
 
 
17149e5
 
 
 
 
a9c396e
 
 
17149e5
a9c396e
 
 
 
 
 
 
 
057f5e7
a9c396e
057f5e7
a9c396e
85c5bca
17149e5
a9c396e
17149e5
 
 
 
 
 
 
 
 
 
 
057f5e7
 
 
 
 
 
 
17149e5
a9c396e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
import gradio as gr
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter, RecursiveCharacterTextSplitter
from langchain.vectorstores import DocArrayInMemorySearch
from langchain.chains import RetrievalQA,  ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.chat_models import ChatOpenAI
from langchain.embeddings import HuggingFaceEmbeddings
from langchain import HuggingFaceHub
from conversadocs.llamacppmodels import LlamaCpp #from langchain.llms import LlamaCpp
from huggingface_hub import hf_hub_download
import param
import os
import torch
from langchain.document_loaders import (
    EverNoteLoader,
    TextLoader,
    UnstructuredEPubLoader,
    UnstructuredHTMLLoader,
    UnstructuredMarkdownLoader,
    UnstructuredODTLoader,
    UnstructuredPowerPointLoader,
    UnstructuredWordDocumentLoader,
    PyPDFLoader,
)
import gc
gc.collect()
torch.cuda.empty_cache()

#YOUR_HF_TOKEN = os.getenv("My_hf_token")

EXTENSIONS = {
    ".txt": (TextLoader, {"encoding": "utf8"}),
    ".pdf": (PyPDFLoader, {}),
    ".doc": (UnstructuredWordDocumentLoader, {}),
    ".docx": (UnstructuredWordDocumentLoader, {}),
    ".enex": (EverNoteLoader, {}),
    ".epub": (UnstructuredEPubLoader, {}),
    ".html": (UnstructuredHTMLLoader, {}),
    ".md": (UnstructuredMarkdownLoader, {}),
    ".odt": (UnstructuredODTLoader, {}),
    ".ppt": (UnstructuredPowerPointLoader, {}),
    ".pptx": (UnstructuredPowerPointLoader, {}),
}

#alter
def load_db(files):

    # select extensions loader
    documents = []
    for file in files:
      ext = "." + file.rsplit(".", 1)[-1]
      if ext in EXTENSIONS:
          loader_class, loader_args = EXTENSIONS[ext]
          loader = loader_class(file, **loader_args)
          documents.extend(loader.load())
      else:
        pass

    # load documents
    if documents == []:
        loader_class, loader_args = EXTENSIONS['.txt']
        loader = loader_class('demo_docs/demo.txt', **loader_args)
        documents = loader.load()

    # split documents
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=150)
    docs = text_splitter.split_documents(documents)

    # define embedding
    embeddings = HuggingFaceEmbeddings(model_name='all-MiniLM-L6-v2') # all-mpnet-base-v2 #embeddings = OpenAIEmbeddings()

    # create vector database from data
    db = DocArrayInMemorySearch.from_documents(docs, embeddings)
    return db

def q_a(db, chain_type="stuff", k=3, llm=None):
    retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": k})
    # create a chatbot chain. Memory is managed externally.
    qa = ConversationalRetrievalChain.from_llm(
        llm=llm,
        chain_type=chain_type,
        retriever=retriever,
        return_source_documents=True,
        return_generated_question=True,
    )
    return qa



class DocChat(param.Parameterized):
    chat_history = param.List([])
    answer = param.String("")
    db_query  = param.String("")
    db_response = param.List([])
    k_value = param.Integer(3)
    llm = None

    def __init__(self,  **params):
        super(DocChat, self).__init__( **params)
        self.loaded_file = ["demo_docs/demo.txt"]
        self.db = load_db(self.loaded_file)
        self.default_falcon_model(os.getenv("My_hf_token")) #self.change_llm("TheBloke/Llama-2-7B-Chat-GGML", "llama-2-7b-chat.ggmlv3.q2_K.bin", max_tokens=256, temperature=0.2, top_p=0.95, top_k=50, repeat_penalty=1.2, k=3)
        self.qa = q_a(self.db, "stuff", self.k_value, self.llm)
        

    def call_load_db(self, path_file, k):
        if not os.path.exists(path_file[0]):  # init or no file specified
            return "No file loaded"
        else:
          try:
            self.db = load_db(path_file)
            self.loaded_file = path_file
            self.qa = q_a(self.db, "stuff", k, self.llm)
            self.k_value = k
            #self.clr_history()
            return f"New DB created and history cleared | Loaded File: {self.loaded_file}"
          except:
            return f'No valid file'


    # chat
    def convchain(self, query, k_max, recall_previous_messages):
        if k_max != self.k_value:
          print("Maximum querys changed")
          self.qa = q_a(self.db, "stuff", k_max, self.llm)
          self.k_value = k_max

        if not recall_previous_messages:
          self.clr_history()

        try:
          result = self.qa({"question": query, "chat_history": self.chat_history})
        except:
          print("Error not get response from model, reloaded default llama-2 7B config")
          self.change_llm("TheBloke/Llama-2-7B-Chat-GGML", "llama-2-7b-chat.ggmlv3.q2_K.bin", max_tokens=256, temperature=0.2, top_p=0.95, top_k=50, repeat_penalty=1.2, k=3)
          self.qa = q_a(self.db, "stuff", k_max, self.llm)
          result = self.qa({"question": query, "chat_history": self.chat_history})
        
        self.chat_history.extend([(query, result["answer"])])
        self.db_query = result["generated_question"]
        self.db_response = result["source_documents"]
        self.answer = result['answer']
        return self.answer

    def summarize(self, chunk_size=2000, chunk_overlap=100):
        # load docs
        if "SET_LIMIT" == os.getenv("DEMO"):
            return "Since the space only uses the CPU, the summarization function cannot be used."
        
        documents = []
        for file in self.loaded_file:
          ext = "." + file.rsplit(".", 1)[-1]
          if ext in EXTENSIONS:
              loader_class, loader_args = EXTENSIONS[ext]
              loader = loader_class(file, **loader_args)
              documents.extend(loader.load_and_split())

        if documents == []:
            return "Error in summarization"

        # split documents
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size=chunk_size, 
            chunk_overlap=chunk_overlap,
            separators=["\n\n", "\n", "(?<=\. )", " ", ""]
            )
        docs = text_splitter.split_documents(documents)
        # summarize
        from langchain.chains.summarize import load_summarize_chain
        chain = load_summarize_chain(self.llm, chain_type='map_reduce', verbose=True)
        return chain.run(docs)

    def change_llm(self, repo_, file_, max_tokens=256, temperature=0.2, top_p=0.95, top_k=50, repeat_penalty=1.2, k=3):

        if torch.cuda.is_available():
            try:
              model_path = hf_hub_download(repo_id=repo_, filename=file_)
              
              self.qa = None
              self.llm = None
              gc.collect()
              torch.cuda.empty_cache()
              gpu_llm_layers = 35 if not '70B' in repo_.upper() else 25 # fix for 70B

              self.llm = LlamaCpp(
                  model_path=model_path,
                  n_ctx=4096,
                  n_batch=512,
                  n_gpu_layers=gpu_llm_layers, 
                  max_tokens=max_tokens,
                  verbose=False,
                  temperature=temperature,
                  top_p=top_p,
                  top_k=top_k,
                  repeat_penalty=repeat_penalty,
                  )
              self.qa = q_a(self.db, "stuff", k, self.llm)
              self.k_value = k
              return f"Loaded {file_} [GPU INFERENCE]"
            except:
              self.change_llm("TheBloke/Llama-2-7B-Chat-GGML", "llama-2-7b-chat.ggmlv3.q2_K.bin", max_tokens=256, temperature=0.2, top_p=0.95, top_k=50, repeat_penalty=1.2, k=3)
              return "No valid model | Reloaded Reloaded default llama-2 7B config"
        else:
            try:
              model_path = hf_hub_download(repo_id=repo_, filename=file_)
              
              self.qa = None
              self.llm = None
              gc.collect()
              torch.cuda.empty_cache()

              self.llm = LlamaCpp(
                  model_path=model_path,
                  n_ctx=2048,
                  n_batch=8,
                  max_tokens=max_tokens,
                  verbose=False,
                  temperature=temperature,
                  top_p=top_p,
                  top_k=top_k,
                  repeat_penalty=repeat_penalty,
                  )
              self.qa = q_a(self.db, "stuff", k, self.llm)
              self.k_value = k
              return f"Loaded {file_} [CPU INFERENCE SLOW]"
            except:
              self.change_llm("TheBloke/Llama-2-7B-Chat-GGML", "llama-2-7b-chat.ggmlv3.q2_K.bin", max_tokens=256, temperature=0.2, top_p=0.95, top_k=50, repeat_penalty=1.2, k=3)
              return "No valid model | Reloaded default llama-2 7B config"

    def default_falcon_model(self, HF_TOKEN):
      self.llm = llm_api=HuggingFaceHub(
          huggingfacehub_api_token=HF_TOKEN,
          repo_id="tiiuae/falcon-7b-instruct",
          model_kwargs={
              "temperature":0.2,
              "max_new_tokens":500,
              "top_k":50,
              "top_p":0.95,
              "repetition_penalty":1.2,
              },)
      self.qa = q_a(self.db, "stuff", self.k_value, self.llm)
      return "Loaded model Falcon 7B-instruct [API FAST INFERENCE]"

    def openai_model(self, API_KEY):
        self.llm = ChatOpenAI(temperature=0, openai_api_key=API_KEY, model_name='gpt-3.5-turbo')
        self.qa = q_a(self.db, "stuff", self.k_value, self.llm)
        API_KEY = ""
        return "Loaded model OpenAI gpt-3.5-turbo [API FAST INFERENCE]"

    @param.depends('db_query ', )
    def get_lquest(self):
        if not self.db_query :
            return print("Last question to DB: no DB accesses so far")
        return self.db_query

    @param.depends('db_response', )
    def get_sources(self):
        if not self.db_response:
            return
        #rlist=[f"Result of DB lookup:"]
        rlist=[]
        for doc in self.db_response:
          for element in doc:
            rlist.append(element)
        return rlist

    @param.depends('convchain', 'clr_history')
    def get_chats(self):
        if not self.chat_history:
            return "No History Yet"
        #rlist=[f"Current Chat History variable"]
        rlist=[]
        for exchange in self.chat_history:
            rlist.append(exchange)
        return rlist

    def clr_history(self,count=0):
        self.chat_history = []
        return "HISTORY CLEARED"