Upload app.py

app.py

@@ -1,152 +1,220 @@
-import subprocess
-import sys
-
-subprocess.check_call([sys.executable,"-m","pip","install",'causal-conv1d'])
-subprocess.check_call([sys.executable, "-m", "pip", "install", 'miditok','mamba-ssm','gradio'])
-subprocess.check_call(["apt-get", "install", "timidity", "-y"])
-
-# !pip install pretty_midi midi2audio
-# !pip install miditok
-# !apt-get install fluidsynth
-# !apt install timidity -y
-# !pip install causal-conv1d>=1.1.0
-# !pip install mamba-ssm
-# !pip install gradio
-
-
-
-# !export LC_ALL="en_US.UTF-8"
-# !export LD_LIBRARY_PATH="/usr/lib64-nvidia"
-# !export LIBRARY_PATH="/usr/local/cuda/lib64/stubs"
-
-# subprocess.check_call(['export', 'LC_ALL="en_US.UTF-8"'])
-# subprocess.check_call(['export', 'LD_LIBRARY_PATH="/usr/lib64-nvidia"'])
-# subprocess.check_call(['export', 'LIBRARY_PATH="/usr/local/cuda/lib64/stubs"'])
-import os
-
-os.environ['LC_ALL'] = "en_US.UTF-8"
-os.environ['LD_LIBRARY_PATH'] = "/usr/lib64-nvidia"
-os.environ['LIBRARY_PATH'] = "/usr/local/cuda/lib64/stubs"
-
-
-
 import gradio as gr
-import torch
-from mamba_ssm import Mamba
-from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
-from mamba_ssm.models.config_mamba import MambaConfig
-import numpy as np
-
-device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
-if torch.cuda.is_available():
-    subprocess.check_call(['ldconfig', '/usr/lib64-nvidia'])
-    # !ldconfig /usr/lib64-nvidia
-
-# !wget "https://huggingface.co/krystv/MIDI_Mamba-159M/resolve/main/MIDI_Mamba-159M_1536VS.pt"
-# !wget "https://huggingface.co/krystv/MIDI_Mamba-159M/resolve/main/tokenizer_1536mix_BPE.json"
-if os.path.isfile("MIDI_Mamba-159M_1536VS.pt") == False:
-    subprocess.check_call(['wget', 'https://huggingface.co/krystv/MIDI_Mamba-159M/resolve/main/MIDI_Mamba-159M_1536VS.pt'])
-
-if os.path.isfile("tokenizer_1536mix_BPE.json") == False:
-    subprocess.check_call(['wget', 'https://huggingface.co/krystv/MIDI_Mamba-159M/resolve/main/tokenizer_1536mix_BPE.json'])
-
-
-
-mc = MambaConfig()
-mc.d_model = 768
-mc.n_layer = 42
-mc.vocab_size = 1536
-
-from miditok import MIDILike,REMI,TokenizerConfig
-from pathlib import Path
-import torch
-
-tokenizer = REMI(params='tokenizer_1536mix_BPE.json')
-
-
-
-mf = MambaLMHeadModel(config=mc,device=device)
-mf.load_state_dict(torch.load("/content/MIDI_Mamba-159M_1536VS.pt",map_location=device))
-
-
-
-twitter_follow_link = "https://twitter.com/iamhemantindia"
-instagram_follow_link = "https://instagram.com/iamhemantindia"
-
-custom_html = f"""
-<div style='text-align: center;'>
-    <a href="{twitter_follow_link}" target="_blank" style="margin-right: 5px;">
-        <img src="https://img.icons8.com/fluent/24/000000/twitter.png" alt="Follow on Twitter"/>
-    </a>
-    <a href="{instagram_follow_link}" target="_blank">
-        <img src="https://img.icons8.com/fluent/24/000000/instagram-new.png" alt="Follow on Instagram"/>
-    </a>
-</div>
+import time
+import os
+import shutil
+import streamlit as st
+openai_api = st.secrets["OPENAI_API_KEY"]
+
+doc_store_path = os.path.join(os.path.dirname(__file__), "doc_dir")
+if not os.path.isdir(doc_store_path):
+    os.makedirs(doc_store_path)
+
+from llama_index.core import SimpleDirectoryReader, VectorStoreIndex,Settings
+from llama_index.core.node_parser import SentenceSplitter,SemanticSplitterNodeParser
+from llama_index.llms.openai import OpenAI
+from llama_index.llms.openai import OpenAI as OpenAIsum
+from llama_index.embeddings.openai import OpenAIEmbedding
+from llama_index.core.storage import StorageContext
+from llama_index.vector_stores.chroma import ChromaVectorStore
+from llama_index.core.storage.chat_store import SimpleChatStore
+from llama_index.core.memory import ChatMemoryBuffer,ChatSummaryMemoryBuffer
+
+import json
+import chromadb
+import tiktoken
+
+
+chat_store = SimpleChatStore()
+# chat_memory = ChatMemoryBuffer.from_defaults(
+#     token_limit=3000,
+#     chat_store=chat_store,
+#     chat_store_key="user1",
+# )
+
+
+sum_llm = OpenAIsum(api_key=openai_api, model="gpt-3.5-turbo", max_tokens=256)
+chat_summary_memory = ChatSummaryMemoryBuffer.from_defaults(
+    token_limit=256,
+    chat_store=chat_store,
+    chat_store_key="user1",
+    llm = sum_llm,
+    tokenizer_fn = tiktoken.encoding_for_model("gpt-3.5-turbo").encode
+)
+
+
+chat_store = SimpleChatStore.from_persist_path(
+    persist_path="chat_store.json"
+)
+
+
+
+# documents = SimpleDirectoryReader("./data").load_data()
+db = chromadb.PersistentClient(path="./chroma_db")
+
+chroma_collection = db.get_or_create_collection("quickstart")
+
+vector_store = ChromaVectorStore(chroma_collection=chroma_collection)
+storage_context = StorageContext.from_defaults(vector_store=vector_store)
+
+Settings.llm = OpenAI(model="gpt-3.5-turbo",api_key=openai_api,)
+Settings.embed_model = OpenAIEmbedding(model="text-embedding-ada-002")
+
+vector_index = VectorStoreIndex.from_vector_store(vector_store, storage_context=storage_context,)
+query_engine = vector_index.as_chat_engine(chat_memory=chat_summary_memory,storage_context=storage_context,use_async=True,similarity_top_k=2)
+
+current_refs = ""
+
+def metadata_from_doc(vec_index: VectorStoreIndex) -> dict:
+    qe = vec_index.as_chat_engine()
+    # f_prompt = """
+    # Given the text excerpts, analyze and provide the document's title and creation date in a structured JSON format. Here are a few examples:
+    
+    # In this format:
+
+    # {
+    # "creation_date": "YYYY-MM-DD",
+    # "title": "Title of the Document"
+    # }
+
+    # Text: 'An analysis of historical events. Written by Alex Johnson on 5 March 2019.'
+    # Output: { "title": "An analysis of historical events", "creation_date": "2019-03-05" }
+
+    # Text: 'Exploring the depths of the ocean. This comprehensive guide was authored by Dr. Emily White, published on 10-July 2021.'
+    # Output: { "title": "Exploring the depths of the ocean", "creation_date": "2021-07-10" }
+
+    # Text: 'The history of the Roman Empire.'
+    # Output: { "title": "The history of the Roman Empire", "creation_date": "Unknown" }
+
+
+    # Now, analyze the context from the provided document and generate json object.
+    # """
+    f_prompt ="""give me a only the data when this document was written and title of this document? in json format parameter (created_date,title),
+        example context: 'An analysis of historical events. Written by Alex Johnson on 5 March 2019.'
+        example output: { "title": "An analysis of historical events", "creation_date": "2019-03-05" }
+    now analyse the context make sure to return output only in json format object only.
+    """
+    res = qe.query(f_prompt)
+    parsed = json.loads(res.response)
+    return parsed
+
+def filter_unsaved(file_paths:list):
+    for i in file_paths:
+        if os.path.isfile(os.path.join(doc_store_path,os.path.basename(i))):
+            file_paths.remove(i)
+            print("File already exist : {}".format(i))
+        else:
+            shutil.copy2(i,doc_store_path)
+    return file_paths
+
+def add_doc(file_paths:list):
+    print(file_paths)
+    file_paths = filter_unsaved(file_paths)
+    print(file_paths)
+    if len(file_paths) == 0:
+        return
+    docs = SimpleDirectoryReader(input_files=file_paths).load_data()
+    splitter = SemanticSplitterNodeParser(buffer_size=1, breakpoint_percentile_threshold=95, embed_model=Settings.embed_model,chunk_size=256)
+    nodes = splitter.get_nodes_from_documents(docs)
+    vector_index2 = VectorStoreIndex(nodes)
+    for i in range (5):
+        try:
+            meta = metadata_from_doc(vector_index2)
+            break
+        except:
+            meta = {
+                "title": "Unknown",
+                "creation_date": "Unknown"
+            }
+            continue
+    
+    print(meta)
+    for i in range(len(nodes)):
+        nodes[i].metadata.update(meta)
+    vector_index.insert_nodes(nodes)
+    
+
+
+
+
+CSS ="""
+.contain { display: flex; flex-direction: column; }
+.gradio-container { height: 100vh !important; }
+#component-0 { height: 100%; }
+#chatbot { flex-grow: 1; overflow: auto;}
 """
 
 
-@spaces.GPU(duration=120)
-def generate(number,top_k_selector,top_p_selector, temperature_selector):
-    input_ids = torch.tensor([[1,]]).to(device)
-    out = mf.generate(
-    input_ids=input_ids,
-    max_length=int(number),
-    temperature=temperature_selector,
-    top_p=top_p_selector,
-    top_k=top_k_selector,
-
-    eos_token_id=2,)
-    m = tokenizer.decode(np.array(out[0].to('cpu')))
-    np.array(out.to('cpu')).shape
-    m.dump_midi('output.mid')
-    # !timidity output.mid -Ow -o - | ffmpeg -y -f wav -i - output.mp3
-    timidity_cmd = ['timidity', 'output.mid', '-Ow', '-o', 'output.wav']
-    subprocess.check_call(timidity_cmd)
-
-    # Then convert the WAV to MP3 using ffmpeg
-    ffmpeg_cmd = ['ffmpeg', '-y', '-f', 'wav', '-i', 'output.wav', 'output.mp3']
-    subprocess.check_call(ffmpeg_cmd)
-
-    return "output.mp3"
-
-
-# text_box = gr.Textbox(label="Enter Text")
-
-
-def generate_and_save(number,top_k_selector,top_p_selector, temperature_selector,generate_button,custom_html_wid):
-    output_audio = generate(number,top_k_selector,top_p_selector, temperature_selector)
-    return gr.Audio(output_audio,autoplay=True),gr.File(label="Download MIDI",value="output.mid"),generate_button
-
-
-
-
-
-
-# iface = gr.Interface(fn=generate_and_save,
-#  inputs=[number_selector,top_k_selector,top_p_selector, temperature_selector,generate_button,custom_html_wid],
-#   outputs=[output_box,download_midi_button],
-#    title="MIDI Mamba-159M",submit_btn=False,
-#    clear_btn=False,
-#    description="MIDI Mamba is a Mamba based model trained on MIDI data collected from open internet to train music model.",
-#    allow_flagging=False,)
-
-with gr.Blocks() as b1:
-    gr.Markdown("<h1 style='text-align: center;'>MIDI Mamba-159M <h1/> ")
-    gr.Markdown("<h3 style='text-align: center;'>MIDI Mamba is a Mamba based model trained on MIDI data collected from open internet to train music model. <br> by Hemant Kumar<h3/>")
+def new_chat(chatbot:gr.Chatbot,textbox):
+    query_engine.reset()
+    return gr.update(value=""),[],"",gr.File(visible=False),gr.File(visible=False)
+
+
+def chat(history, input):
+    response = query_engine.chat(str(input))
+    global current_refs
+    files = []
+    current_refs = ""
+    for node in response.source_nodes:
+        try:
+            current_refs += f"{str(node.metadata['title'])},"
+        except:
+            current_refs += ""
+        try:
+            current_refs += f"Pg - {str(node.metadata['page_label'])},"
+        except:
+            current_refs += "Pg - ,"
+        try:
+            current_refs += f"File - {str(node.metadata['file_name'])} \n,"
+        except:
+            current_refs += "File - ,\n"
+        
+        try:
+            files.append({'path':node.metadata['file_path'],'show':True,})
+        except:
+            files.append({'path':None,'show':False,})
+
+    if len(files) < 2:
+        for _ in range(2-len(files)):
+            files.append({'path':None,'show':False,})
+        
+    return gr.update(value=""),history + [(input, response.response)],current_refs,gr.update(visible=files[0]['show'],value=files[0]['path']),gr.update(visible=files[1]['show'],value=files[1]['path'])
+
+def file_upload(file,chatbot):
+    print(file)
+    add_doc(file)
+    return gr.update(value="ChatDoc"),chatbot
+
+with gr.Blocks(fill_height=True, css=CSS) as demo:
     with gr.Row():
-        with gr.Column():
-            number_selector = gr.Number(label="Select Length of output",value=512)
-            top_p_selector = gr.Slider(label="Select Top P", minimum=0, maximum=1.0, step=0.05, value=0.9)
-            temperature_selector = gr.Slider(label="Select Temperature", minimum=0, maximum=1.0, step=0.1, value=0.9)
-            top_k_selector = gr.Slider(label="Select Top K", minimum=1, maximum=1536, step=1, value=30)
-            generate_button = gr.Button(value="Generate",variant="primary")
-            custom_html_wid = gr.HTML(custom_html)
-        with gr.Column():
-            output_box = gr.Audio("output.mp3",autoplay=True,)
-            download_midi_button = gr.File(label="Download MIDI")
-            generate_button.click(generate_and_save,inputs=[number_selector,top_k_selector,top_p_selector, temperature_selector,generate_button,custom_html_wid],outputs=[output_box,download_midi_button,generate_button])
-
-
-
-
-b1.launch(share=True)
+        with gr.Column(scale=1):
+            title = gr.Label(value="chatdoc", label="ChatDoc")
+            files = gr.UploadButton(
+                    "📁 Upload PDF or doc files", file_types=[
+                        '.pdf',
+                        '.doc'
+                    ],
+                    file_count="multiple")
+            references = gr.Textbox(label="References",interactive=False)
+            file_down1 = gr.File(visible=False)
+            file_down2 = gr.File(visible=False) 
+            
+
+        
+        with gr.Column(scale=9,):
+            chatbot = gr.Chatbot(
+                elem_id="chatbot",
+                bubble_full_width=False,
+                label="ChatDoc",
+                avatar_images=["https://www.freeiconspng.com/thumbs/person-icon-blue/person-icon-blue-25.png","https://cdn-icons-png.flaticon.com/512/8943/8943377.png"],
+            )
+            with gr.Row():
+                textbox = gr.Textbox(label="Type your message", scale=10)
+                clear = gr.Button(value="New Chat", size="sm", scale=1)
+                clear.click(new_chat,[],[textbox, chatbot,references,file_down1,file_down2])
+                textbox.submit(chat, [chatbot, textbox], [textbox, chatbot,references,file_down1,file_down2])
+                
+
+        files.upload(file_upload,[files,chatbot],[title,chatbot])
+                
+    
+demo.launch(share=True)