initial_commit

app.py

@@ -1,63 +1,236 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
+## Import required packages
 
+from langchain.llms import CTransformers
+from langchain.prompts import FewShotChatMessagePromptTemplate, ChatPromptTemplate, FewShotPromptTemplate
+import gradio as gr
+from langchain.chains import LLMChain
+from langchain.prompts import PromptTemplate
+# from langchain.document_loaders import WikipediaLoader
+from langchain.document_loaders import JSONLoader
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.embeddings import HuggingFaceEmbeddings
+from langchain.vectorstores import Chroma
+from langchain.chains import RetrievalQA
+from operator import itemgetter
+from langchain_core.output_parsers import JsonOutputParser
+from langchain_core.outputs import Generation
+from typing import Any, List, Optional, Type, TypeVar, Union
+
+## Defining few variables
+MODEL_PATH = "TheBloke/Mistral-7B-Claude-Chat-GGUF"
+MODEL_FILE = "mistral-7b-claude-chat.Q4_K_M.gguf"
+MODEL_TYPE = "mistral"
+MAX_NEW_TOKENS = 100
+temperature = 1
+top_p = 0.95
+top_k = 50
+repetition_penalty = 1.5
+
+## Defining Model
+llm = CTransformers(
+    model = MODEL_PATH,
+    model_file=MODEL_FILE,
+    model_type = MODEL_TYPE,
+    config = {
+        "max_new_tokens":MAX_NEW_TOKENS,
+        "temperature": temperature,
+        "top_p": top_p,
+        "top_k": top_k,
+        "repetition_penalty": repetition_penalty,
+        "last_n_tokens": 4,
+        "stream": True,
+        "gpu_layers": 1000
+    }
+)
+
+# One shot inferencing
+examples = [
+    {
+        "query": "Please classify this name: Ketan Jogadankar",
+        "answer":"""{
+            "name": "Ketan Jogadankar",
+            "label": "person",
+            "score": 0.99,
+            "reason": "Ketan is a most famous first name and Jogadankar looks like a surname."
+        }"""
+    }
+]
+
+example_template = """
+User: {query}
+{answer}
 """
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+
+example_prompt = ChatPromptTemplate.from_messages(
+    [("human", "{query}"),
+        ("ai", "{answer}")]
+)
+
+prefix = """Act as an AI assistant that classifies names into 3 categories (person, business and other) based on the provided rules and example data.
+{format_instructions}
+Do not append any text to human input.
+Rules:
+* If the names contains the word "POD", classify it as a other.
+* If the names contains the word "trust", classify it as a other.
+* If the names contains the word "llc", classify it as a business.
+* If the name is non-profit organization then classify it as a other.
+Here are some examples:
 """
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
+suffix = """Please classify this name: {name}
+"""
+
+few_shot_prompt_template = FewShotChatMessagePromptTemplate(
+    examples = examples,
+    example_prompt = example_prompt
+)
+
+prompt = ChatPromptTemplate.from_messages(
+    [
+        ("system",prefix),
+        few_shot_prompt_template,
+        ("human", suffix)
+    ]
+)
+
+format_instructions = """The output should be formatted as a JSON instance that conforms to the JSON schema below.
+Here is the output schema:
+```
+{"properties": {"name": {"title": "Name", "description": "this is the input name passed by human", "type": "string"}, "label": {"title": "Label", "description": "this is the label predicted for input name", "type": "string"}, "score": {"title": "Score", "description": "This is confidence score for predicted label", "type": "number"}, "reason": {"title": "Reason", "description": "This is to explain why AI has predicted that label", "type": "string"}}, "required": ["name", "label", "score", "reason"]}
+```
+"""
+# RAG
+data_loader = JSONLoader(file_path="/content/sample_data/anscombe.json",
+                         jq_schema='.',text_content=False)
+data = data_loader.load()
+data = [doc.page_content for doc in data]
+
+splitter = CharacterTextSplitter(chunk_size=2, chunk_overlap=1)
+documents = splitter.create_documents(texts=data)
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+docs_str = [doc.page_content for doc in documents]
+sentence_emb = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+db = Chroma.from_texts(docs_str, sentence_emb, persist_directory="./temp_db")
+db.persist()
 
-    messages.append({"role": "user", "content": message})
+retriever = db.as_retriever(
+    search_type="similarity",
+    search_kwargs={'k':1})
 
-    response = ""
+# Pydantic output validator
+from pydantic import BaseModel, Field
+class NameClassification(BaseModel):
+    name:str = Field(description="this is the input name passed by human")
+    label:str = Field(description="this is the label predicted for input name")
+    score:float = Field(description="This is confidence score for predicted label")
+    reason:str = Field(description="This is to explain why AI has predicted that label")
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+    def remove_junks(self, text):
+        start_index = text.index("{")
+        stop_index = text.index("}") + 1
+        return text[start_index:stop_index+1]
 
-        response += token
-        yield response
+    def parse(self, text):
+        text = self.remove_junks(text)
+        super().invoke(text)
+
+class CustomParser(JsonOutputParser):
+
+    def parse_result(self, result: List[Generation], *, partial: bool = False) -> Any:
+        text = result[0].text
+        text = text.strip()
+        text = self.remove_junks(text)
+        result = [Generation(text= text)]
+        return super().parse_result(result=result,partial=partial)
+
+
+    def remove_junks(self, text):
+        start_index = text.index("{")
+        stop_index = text.index("}") + 1
+        return text[start_index:stop_index+1]
+
+parser = CustomParser(pydantic_object=NameClassification)
+
+chain = (
+        {"context": itemgetter("name") | retriever,
+         "format_instructions": itemgetter("format_instructions"),
+         "name": itemgetter("name")}
+        | prompt
+        | llm
+        | parser
+)
 
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
 """
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+
+def predict(message, history, min_hist_memo = 3):
+    # streamer = chain(message)
+    streamer = chain.invoke({"name":message, "format_instructions":format_instructions})
+    # partial_message = ""
+    # for new_token in streamer:
+    #     if new_token != '<':
+    #         partial_message += new_token
+    #         yield partial_message
+    yield str(streamer)
+
+gr.ChatInterface(predict, title="Mistral 7B").queue().launch(debug=True)
+
+# def respond(
+#     message,
+#     history: list[tuple[str, str]],
+#     system_message,
+#     max_tokens,
+#     temperature,
+#     top_p,
+# ):
+#     messages = [{"role": "system", "content": system_message}]
+
+#     for val in history:
+#         if val[0]:
+#             messages.append({"role": "user", "content": val[0]})
+#         if val[1]:
+#             messages.append({"role": "assistant", "content": val[1]})
+
+#     messages.append({"role": "user", "content": message})
+
+#     response = ""
+
+#     for message in client.chat_completion(
+#         messages,
+#         max_tokens=max_tokens,
+#         stream=True,
+#         temperature=temperature,
+#         top_p=top_p,
+#     ):
+#         token = message.choices[0].delta.content
+
+#         response += token
+#         yield response
+
+# """
+# For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+# """
+# demo = gr.ChatInterface(
+#     respond,
+#     additional_inputs=[
+#         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+#         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+#         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+#         gr.Slider(
+#             minimum=0.1,
+#             maximum=1.0,
+#             value=0.95,
+#             step=0.05,
+#             label="Top-p (nucleus sampling)",
+#         ),
+#     ],
+# )
 
 
-if __name__ == "__main__":
-    demo.launch()
+# if __name__ == "__main__":
+#     demo.launch()