From 4af16bacd9ad2dd16d938a7323eda41851e13fe0 Mon Sep 17 00:00:00 2001
From: "Steve L. Nyemba" <steve.l.nyemba@vumc.org>
Date: Wed, 19 Mar 2025 20:39:05 +0000
Subject: [PATCH] Delete www/html/medix.py

---
 www/html/medix.py | 138 ----------------------------------------------
 1 file changed, 138 deletions(-)
 delete mode 100644 www/html/medix.py

diff --git a/www/html/medix.py b/www/html/medix.py
deleted file mode 100644
index 2f19882..0000000
--- a/www/html/medix.py
+++ /dev/null
@@ -1,138 +0,0 @@
-from langchain_openai import AzureOpenAIEmbeddings, AzureOpenAI, AzureChatOpenAI
-from langchain_ollama import OllamaEmbeddings, OllamaLLM
-import io
-from langchain.chains import LLMChain
-from langchain_core.prompts import PromptTemplate
-import os
-import json
-import transport
-
-import cms
-import uuid
-from multiprocessing import Process 
-import pandas as pd
-import copy
-
-from gtts import gTTS
-import io
-import requests
-
-# USE_OPENAI = os.environ.get('USE_OPENAI',0)
-
-class ILLM :
-    def __init__(self,**_args) :
-        self.USE_OPENAI = 'openai' in _args
-        # self._args = _args['ollama'] if not self.USE_OPENAI else _args['openai']
-        _path = _args['ollama'] if not self.USE_OPENAI else _args['openai']
-        f = open(_path)
-        self._args = json.loads( f.read() )
-        f.close()
-        self._prompt = _args['prompt'] if 'prompt' in _args else {}
-    def embed(self,_question):
-        _pointer =  AzureOpenAIEmbeddings if self.USE_OPENAI else OllamaEmbeddings
-        _kwargs = self._args if 'embedding' not in self._args else self._args['embedding']
-        _handler = _pointer(**_kwargs)
-        return _handler.embed_query(_question)
-    def answer(self,_question,_context) :
-        _pointer =  AzureChatOpenAI if self.USE_OPENAI else OllamaLLM
-        _kwargs = self._args if 'completion' not in self._args else self._args['completion']
-        _llm =  _pointer(**_kwargs)
-        _prompt = PromptTemplate(**self._prompt)
-        chain   = LLMChain(llm=_llm,prompt=_prompt)
-        _input = {'context':_context,'question':_question}
-        _input = json.loads(json.dumps(_input))
-        resp = chain.invoke(_input)
-        return resp
-    def schema(self):
-        return 'openai' if self.USE_OPENAI else 'public'
-    def documents(self,_vector) :
-        _schema = 'openai' if self.USE_OPENAI else 'ollama'
-        pgr = transport.get.reader(label='llm', schema=_schema)
-        sql = f"""SELECT file, name, page, content, embeddings <-> '{json.dumps(_vector)}' similarity FROM {_schema}.documents
-        ORDER BY similarity ASC
-        LIMIT 5
-        """
-        _df = pgr.read(sql=sql)
-        pgr.close()
-
-        return _df
-    def lookup (self,index:int,token) :
-        _schema = 'openai' if self.USE_OPENAI else 'ollama'
-        pgr = transport.get.reader(label='llm', schema=_schema)
-        index = int(index)  + 1
-        _sql = f"SELECT * FROM (select row_number() over(partition by token) as row_index, answer from llm_logs where token='{token}') as _x where row_index    = {index}"
-        print (_sql)
-        _df = pgr.read(sql= _sql)
-        return _df.answer[0] if _df.shape[0] > 0 else None
-@cms.Plugin(mimetype="application/json",method="POST")
-def answer (**_args):
-    
-    _request = _args['request']
-    _config = _args['config']['system']['source']['llm']
-
-    _question = _request.json['question']
-    token = str(uuid.uuid4()) if 'token' not in _request.json else _request.json['token']
-    _index = _request.json['index'] if 'index' in _request.json else 0 
-
-    _llmproc = ILLM(**_config)
-
-    #
-    # Turn the question into a vector and send it to the LLM Server
-    #
-    _vector = _llmproc.embed(_question)
-    _df = _llmproc.documents(_vector)
-    _pages = _df.apply(lambda row: row.content,axis=1).tolist()
-    
-    #
-    # submit the request to the 
-
-
-    # return _df[['name','page','similarity']].to_dict(orient='records')
-    #
-    # Let us submit the question to the llm-server
-    #
-    resp =  _llmproc.answer(_question, _pages)
-    #
-    # @TODO :
-    #   - Log questions, answers and sources to see what things are like
-    _context = _args['config']['system']['context'].strip()
-    _out = {"token":token,"openai":_llmproc.USE_OPENAI,"answer":resp["text"],"documents": _df[["name","page"]].to_dict(orient='records')}
-    
-    
-    try:
-        def _logger():
-            _log = pd.DataFrame([dict(_out,**{'question':_question})])
-            _log.documents = _df[["name","page"]].to_json(orient='records')
-            pgw = transport.get.writer (label='llm',table='llm_logs')
-            pgw.write(_log)
-            pgw.close()
-        #
-        # send the thread 
-        pthread = Process(target=_logger)
-        pthread.start()
-    except Exception as e:
-        print (e)
-    # _out['stream'] = f'{_context}/api/medix/audio?token={token}&index={_index}'
-    return _out
-
-@cms.Plugin(mimetype="text/plain",method="POST")
-def info (**_args):
-    _config = _args['config']
-    return list(_config['system']['source']['llm'].keys())[0]
-@cms.Plugin(mimetype="audio/mpeg",method="GET")
-def audio (**_args):
-    _request = _args['request']
-    _config = _args['config']['system']['source']['llm']
-    _index = _request.args['index']
-    _token = _request.args['token']
-    _llmproc = ILLM(**_config)
-
-    _llmproc = ILLM(**_config)
-    _text = _llmproc.lookup(_index,_token)
-    g = gTTS(_text,lang='en')
-    return g.stream()
-    # stream = io.BytesIO()
-    # for line in g.stream() :
-    #     stream.write(line)
-    # stream.seek(0)
-    # return stream #g.stream()
\ No newline at end of file