Create app.py

app.py

@@ -0,0 +1,261 @@
+import gradio as gr
+# from transformers import pipeline
+# from transformers.utils import logging
+from huggingface_hub import InferenceClient
+from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
+from llama_index.vector_stores.chroma import ChromaVectorStore
+from llama_index.core.vector_stores import (MetadataFilters, ExactMatchFilter, )
+from llama_index.embeddings.huggingface import HuggingFaceEmbedding
+import torch
+from llama_index.core import (
+    VectorStoreIndex,
+    Document,
+    Settings,
+)
+import requests
+import json
+from llama_index.llms.huggingface import (HuggingFaceLLM, )
+from llama_index.llms.huggingface_api import (HuggingFaceInferenceAPI, )
+from llama_index.core.base.llms.types import ChatMessage
+from huggingface_hub import login
+import chromadb as chromadb
+from chromadb.utils import embedding_functions
+import shutil
+import os
+from io import StringIO
+from llama_index.core.memory import ChatMemoryBuffer
+
+memory = ChatMemoryBuffer.from_defaults(token_limit=3072)
+#
+
+last = 0
+CHROMA_DATA_PATH = "chroma_data/"
+EMBED_MODEL =  "sentence-transformers/all-MiniLM-L6-v2" # "BAAI/bge-m3"
+#LLM_NAME = "mistralai/Mistral-Nemo-Instruct-2407"
+#LLM_NAME = "sswiss-ai/apertus-70b-instruct" # provider: publicai
+#LLM_NAME = "openai/gpt-oss-20b"
+LLM_NAME = "swiss-ai/apertus-8b-instruct"
+#LLM_NAME = "aisingapore/Gemma-SEA-LION-v4-27B-IT"
+#LLM_NAME = "W4D/YugoGPT-7B-Instruct-GGUF"
+CHUNK_SIZE = 800
+CHUNK_OVERLAP = 50
+max_results = 3
+min_len = 40
+min_distance = 0.35
+max_distance = 0.6
+temperature = 0.7
+max_tokens=5100
+top_p=0.85
+top_k=1000
+frequency_penalty=0.0
+repetition_penalty=1.12
+presence_penalty=0.15
+cs = "s0"
+sp_flag = True
+
+system_sr = "Zoveš se U-Chat AI asistent i pomažeš odgovorima korisniku usluga UFO Cowboy. Korisnik postavlja pitanje ili problem na koji očekuje rešenje. "
+# " Ako ne znaš odgovor, reci da ne znaš, ne izmišljaj ga."
+system_sr += " UFO Cowboy je potpuno otvoren svet fiktivnih ljudi i drugih oblika života "
+
+chroma_client = chromadb.PersistentClient(CHROMA_DATA_PATH)
+embedding_func = embedding_functions.SentenceTransformerEmbeddingFunction(
+        model_name=EMBED_MODEL
+    )
+
+collection = chroma_client.get_or_create_collection(
+        name="chroma_data",
+        embedding_function=embedding_func,
+        metadata={"hnsw:space": "cosine"},
+    )
+last = collection.count()
+
+#
+HF_TOKEN = os.getenv("HF_TOKEN")
+API_SECRET = os.getenv("api_secret")
+#
+login(token=(HF_TOKEN))
+
+#system_prompt = system_sr
+
+client = InferenceClient(LLM_NAME)
+
+# "facebook/blenderbot-400M-distill",  facebook/blenderbot-400M-distill, stabilityai/stablelm-zephyr-3b, BAAI/bge-small-en-v1.5
+Settings.llm = HuggingFaceInferenceAPI(model_name=LLM_NAME,
+#                              device_map="auto",
+#                              system_prompt = system_prompt,
+                              context_window=6144,
+                              max_new_tokens=3072,
+#                              stopping_ids=[50278, 50279, 50277, 1, 0],
+                              generate_kwargs={"temperature": temperature, "top_p":top_p, "repetition_penalty": repetition_penalty,
+                                               "presence_penalty": presence_penalty, "frequency_penalty": frequency_penalty,
+                                               "top_k": top_k, "do_sample": False },
+#                              tokenizer_kwargs={"max_length": 4096},
+                              tokenizer_name=LLM_NAME,
+                              hf_token = HF_TOKEN,
+                              src = "models",
+                              provider="publicai",
+                             )
+# "BAAI/bge-m3"
+Settings.embed_model = HuggingFaceEmbedding(model_name=EMBED_MODEL)
+#documents = [Document(text="Content ..."),
+#            ]
+#index = VectorStoreIndex.from_documents(
+#    documents,
+#)
+vector_store = ChromaVectorStore(chroma_collection=collection)
+index = VectorStoreIndex.from_vector_store(vector_store, embed_model=Settings.embed_model)
+chat_engine = index.as_chat_engine(chat_mode="condense_plus_context", memory=memory, verbose=True)
+# best condense_question context condense_plus_context
+#query_engine = index.as_query_engine(verbose=True)
+
+def upload_file(filepath):
+    documents = SimpleDirectoryReader(filepath).load_data()
+    index = VectorStoreIndex.from_documents(documents)
+    #query_engine = index.as_query_engine()
+    #condense_question condense_plus_context
+    chat_engine = index.as_chat_engine(verbose=True)
+    return filepath
+
+def resetChat():
+    chat_engine.reset()
+    print("Restarted!!!")
+    return True
+
+def rag(input_text, history, jezik, request: gr.Request): # , file):
+    global sp_flag
+#    if (btn):
+#        resetChat()
+#    print(history, input_text)
+##    if (file):
+    documents = []
+#!!!    for f in file:
+#!!!        documents += SimpleDirectoryReader(f).load_data()
+#        f = file + "*.pdf"
+##        pathname = os.path.dirname
+#        shutil.copyfile(file.name, path)
+##        print("pathname=", pathname)
+##        print("basename=", os.path.basename(file))
+##        print("filename=", file.name)        
+##        documents = SimpleDirectoryReader(file).load_data()
+#!!!        index2 = VectorStoreIndex.from_documents(documents)
+##        query_engine = index2.as_query_engine()
+#        return query_engine.query(input_text)
+#        return history.append({"role": "assistant", "content": query_engine.query(input_text)})
+##        return history + [[input_text, query_engine.query(input_text)]]
+#        collection.add(
+#            documents=documents,
+#        ids=[f"id{last+i}" for i in range(len(documents))],
+#        metadatas=[{"state": "s0", "next": "s0", "used": False, "source": 'None', "page": -1, "lang": jezik } for i in range(len(documents)) ]
+#    )
+##    else:
+###    query_results = collection.query(
+    #query_engine = index.as_query_engine(
+    #    similarity_top_k=3,
+    #    vector_store_query_mode="default",
+    #    filters=MetadataFilters(
+    #         filters=[
+    #             ExactMatchFilter(key="lang", value=jezik),
+    #        ]
+    #    ),
+    #    alpha=None,
+    #    doc_ids=None,
+    #)
+    #query_results = index.query(
+    #    query_texts = [ input_text ],
+    #    n_results = max_results,
+    #    where = { "lang": jezik },
+    #    #where = { "$and": [ {"lang": jezik}, {"page": { "$nin": [ -1 ]}}]},
+    #    #where = { "$and": [ {"$and": [ { "$or": [ {"state": self.cs },  { "page": { "$nin": [ -1 ] } } ] } , { "used": False } ] } ,
+    #    #        {"lang": jezik } ] },
+    #)
+    #jezik = "N/A"
+    system_prompt = ""
+    match jezik:
+        case 'hrvatski':
+            o_jezik = 'na hrvatskom jeziku, gramatički točno.'
+            system_prompt = system_sr + "Call centar telefon je 095 1000 444 za privatne i 095 1000 500 za poslovne korisnike. Stranica podrške je <https://tele mach.hr/podrska>." + "Odgovaraj isključivo " + o_jezik
+        case 'slovenski':
+            o_jezik = 'v slovenščini, slovnično pravilen.'
+            system_prompt = system_sr + "Call centar i pomoč za fizične uporabnike: 070 700 700.stran za podporo je <https://telemach.si/pomoc>. " + "Odgovor isključivo " + o_jezik
+        case 'srpski':
+            o_jezik = 'na srpskom jeziku, gramatički ispravno.'
+            system_prompt = system_sr + "Call centar telefon je 19900 za sve korisnike. Stranica podrške je <https://sbb.rs/podrska/>. " + "Odgovaraj isključivo " + o_jezik
+        case 'makedonski':
+            o_jezik = 'на македонски јазикот граматички точно.'
+            system_prompt = system_sr + "Stranica podrške je https://mn.nettvplus.com/me/podrska/ za NetTV. " + "Oдговори исклучиво " + o_jezik
+        case 'Eksperimentalna opcija':
+            o_jezik = 'N/A'
+            system_prompt = system_sr + "Call centar telefon je 12755 za Crnu Goru, 0800 31111 za BIH, 070 700 700 u Sloveniji, 19900 u Srbiji, 095 1000 444 za hrvatske korisnike. Odgovori na jeziku istom kao i u postavljenom pitanju ili problemu korisnika."
+    print("jezik: "+o_jezik)
+    system_prompt = system_prompt + " Sledi pitanje ili problem korisnika, sa kojim dalje komuniciraš: "
+    if sp_flag:
+        sp_flag = False
+    else:
+        system_prompt = ""
+#        if (o_jezik!='N/A'):
+#            input_text += " - odgovori " + o_jezik + "."
+#    Settings.llm.system_prompt = system_prompt
+    response = chat_engine.chat(str({"role": "user", "content": system_prompt + input_text})).response
+#    response = query_engine.query(input_text)
+    session_hash = request.session_hash
+    url = "https://houseofplaying.com/FBMod/public/api.php" # An example public API
+    payload = {
+        "session_hash": session_hash,
+        "input_query": input_text,
+        "response": response
+    }
+    headers = {
+        "Authorization": "Bearer " + API_SECRET,
+        "Content-Type": "application/json"
+    }
+    try:
+        api_response = requests.post(url, json=payload, headers=headers)
+    except:
+        print("POST error for "+session_hash)
+    return response
+# gr.Textbox(label="Pitanje:", lines=6),
+# outputs=[gr.Textbox(label="Odgovor:", lines=6)], 
+# ChatMessage(role="assistant", content="Kako Vam mogu pomoći?")
+
+with gr.Blocks() as iface:
+    ichat = gr.ChatInterface(fn=rag,        
+        title="UChat",
+        description="Postavite pitanje ili opišite problem koji imate - nakon promene jezika ili pre početka nove sesije sa agentom pritisnite dugme 'Briši sve - razgovor ispočetka'",
+        chatbot=gr.Chatbot(placeholder="Kako Vam mogu pomoći?", type="tuples", label="Agent podrške", height=350),
+        textbox=gr.Textbox(placeholder="Pitanje ili opis problema", container=False, scale=7),
+        autofocus = True,
+        theme="soft",
+        examples = [
+                    ["Ne radi mi internet", "srpski", ],
+                    ["Možete li mi popraviti kompjuter koji koristi internet?", "srpski", ],
+                    ["Ne radi mi daljinski upravljač, šta da radim?", "srpski", ],
+                    ["EON daljinski upravljalnik mi ne deluje, kaj naj naredim?", "slovenski", ],
+                    ["Мојот кабелски прием не работи, што треба да направам?", "makedonski", ],
+        ],
+        cache_examples=False,
+        retry_btn=None,
+        undo_btn=None,
+        clear_btn="Briši sve - razgovor ispočetka",
+        additional_inputs = [gr.Dropdown(["slovenski", "hrvatski", "srpski", "makedonski", "Eksperimentalna opcija"], value="srpski", label="Jezik", info="komunikacije"),
+#                             gr.File()
+                         ],
+        additional_inputs_accordion="Jezik i ostale opcije",
+    )
+#    login_button = gr.LoginButton("Hugging Face login", size="lg")
+    ichat.clear_btn.click(resetChat)
+
+#with gr.Blocks() as iface:
+#    gr.Markdown("Uchat")
+#    file_out = gr.File()
+#    with gr.Row():
+#        with gr.Column(scale=1):
+#            inp = gr.Textbox(label="Pitanje:", lines=6)
+#            u = gr.UploadButton("Upload a file", file_count="single")
+#        with gr.Column(scale=1):
+#            out = gr.Textbox(label="Odgovor:", lines=6)
+#            sub = gr.Button("Pokreni")
+#
+#    u.upload(upload_file, u, file_out)
+#    sub.click(rag, inp, out)
+    
+iface.launch(share=True)