Update app.py

app.py

@@ -1,33 +1,74 @@
-import json
-import subprocess
+#########################################################################################
+# Title:  Gradio Interface to LLM-chatbot with on premises 
+# Author: Andreas Fischer
+# Date:   October 15th, 2023
+# Last update: January 21st, 2024
+##########################################################################################
+
+#https://github.com/abetlen/llama-cpp-python/issues/306
+#sudo apt install libclblast-dev
+#CMAKE_ARGS="-DLLAMA_CLBLAST=on" FORCE_CMAKE=1 pip install llama-cpp-python --force-reinstall --upgrade --no-cache-dir -v
+
+# Get model 
+#-----------
+
+import os
 import requests
-from llama_cpp import Llama
-import gradio as gr
 
-#url="https://huggingface.co/TheBloke/WizardLM-13B-V1.2-GGUF/resolve/main/wizardlm-13b-v1.2.Q4_0.gguf"
-#url="https://huggingface.co/TheBloke/Mixtral-8x7B-Instruct-v0.1-GGUF/resolve/main/mixtral-8x7b-instruct-v0.1.Q4_0.gguf?download=true"
-url="https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_0.gguf?download=true"
-response = requests.get(url)
-with open("./model.gguf", mode="wb") as file:
-  file.write(response.content)
-print("Model downloaded")
+##modelPath="/home/af/gguf/models/phi-2.Q4_0.gguf"
+#modelPath="/home/af/gguf/models/openchat-3.5-0106.Q4_0.gguf"
+modelPath="/home/af/gguf/models/SauerkrautLM-7b-HerO-q8_0.gguf"
+#modelPath="/home/af/gguf/models/sauerkrautlm-una-solar-instruct.Q4_0.gguf"
+#modelPath="/home/af/gguf/models/decilm-7b-uniform-gqa-q8_0.gguf"
+#modelPath="/home/af/gguf/models/mixtral-8x7b-instruct-v0.1.Q4_0.gguf"
+if(os.path.exists(modelPath)==False):
+  #url="https://huggingface.co/TheBloke/WizardLM-13B-V1.2-GGUF/resolve/main/wizardlm-13b-v1.2.Q4_0.gguf"
+  #url="https://huggingface.co/TheBloke/Mixtral-8x7B-Instruct-v0.1-GGUF/resolve/main/mixtral-8x7b-instruct-v0.1.Q4_0.gguf?download=true"
+  url="https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_0.gguf?download=true"
+  response = requests.get(url)
+  with open("./model.gguf", mode="wb") as file:
+    file.write(response.content)
+  print("Model downloaded")  
+  modelPath="./model.gguf"
+
+print(modelPath)
+
 
-command = ["python3", "-m", "llama_cpp.server", "--model", "./model.gguf", "--host", "0.0.0.0", "--port", "2600", "--n_threads", "2"]
+# Llama-cpp-Server
+#------------------
+
+import subprocess
+command = ["python3", "-m", "llama_cpp.server", "--model", modelPath, "--host", "0.0.0.0", "--port", "2600", "--n_threads", "4", "--n_gpu_layers","20"]
 subprocess.Popen(command)
-print("Model ready!")
+print("Server ready!")
+
+
 
-#llm = Llama(model_path="./model.gguf")
-#def response(input_text, history):
-#    output = llm(f"Q: {input_text} A:", max_tokens=256, stop=["Q:", "\n"], echo=True)
-#    return output['choices'][0]['text']
+# Gradio-GUI
+#------------
 
+import gradio as gr
+import requests
+import json
 def response(message, history):
+  prompt=message
+  system="Du bist ein KI-basiertes Assistenzsystem."
+  if("mixtral-8x7b-instruct" in modelPath): 
+    prompt=f"[INST] {prompt} [/INST]"
+  if("mixtral-8x7b-instruct" in modelPath): 
+    prompt=f"[INST] {prompt} [/INST]"
+  if("openchat-3.5" in modelPath):
+    prompt=f"GPT4 Correct User: {system} {prompt}<|end_of_turn|>GPT4 Correct Assistant:"
+  if("SauerkrautLM-7b-HerO" in modelPath):    
+    prompt=f"<|im_start|>system\n{system}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
+  if("WizardLM-13B-V1.2" in modelPath):
+    prompt=f"{system} USER: {prompt} ASSISTANT: "
+  if("phi-2" in modelPath):
+    prompt=f"Instruct: {prompt}\nOutput:"
+  print(prompt)
   #url="https://afischer1985-wizardlm-13b-v1-2-q4-0-gguf.hf.space/v1/completions"
   url="http://0.0.0.0:2600/v1/completions"  
-  #body={"prompt":"Im Folgenden findest du eine Instruktion, die eine Aufgabe bescheibt. Schreibe eine Antwort, um die Aufgabe zu lösen.\n\n### Instruktion:\n"+message+"\n\n### Antwort:","max_tokens":500, "echo":"False","stream":"True"}
-  #body={"prompt":" chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.\n\nUSER:\n"+message+"\n\nASSISTANT:","max_tokens":500, "echo":"False","stream":"True"}    
-  #body={"prompt":system+"### Instruktion:\n"+message+"\n\n### Antwort:","max_tokens":500, "echo":"False","stream":"True"} #e.g. SauerkrautLM
-  body={"prompt":"[INST]"+message+"[/INST]","max_tokens":500, "echo":"False","stream":"True"} #e.g. Mixtral-Instruct
+  body={"prompt":prompt,"max_tokens":1000, "echo":"False","stream":"True"} #e.g. Mixtral-Instruct
   response=""
   buffer=""
   print("URL: "+url)
@@ -56,5 +97,5 @@ def response(message, history):
         pass
     yield response 
 
-gr.ChatInterface(response,title="Mistral-7B-Instruct-v0.2-GGUF Chatbot").queue().launch(share=True) #False, server_name="0.0.0.0", server_port=7864)
-print("Interface up and running!")
+gr.ChatInterface(response, chatbot=gr.Chatbot(render_markdown=True),title="AI-Interface").queue().launch(share=True) #False, server_name="0.0.0.0", server_port=7864)
+print("Interface up and running!")