update extend_prompt

run.py

@@ -2,7 +2,7 @@
 # Title:  Gradio AI-Interface with Memory-RAG
 # Author: Andreas Fischer
 # Date:   October 15th, 2023
-# Last update: March 1st, 2024
+# Last update: May 27th, 2024
 ##########################################################################################
 
 #https://github.com/abetlen/llama-cpp-python/issues/306
@@ -139,12 +139,22 @@ def extend_prompt(message="", history=None, system=None, RAGAddon=None, system2=
   template0=" [INST]{system}\n  [/INST] </s>" 
   template1=" [INST] {message} [/INST]"
   template2=" {response}</s>"
+  if("command-r" in modelPath): #https://huggingface.co/CohereForAI/c4ai-command-r-v01
+    ## <BOS_TOKEN><|START_OF_TURN_TOKEN|><|USER_TOKEN|>Hello, how are you?<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>
+    template0="<BOS_TOKEN><|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|> {system}<|END_OF_TURN_TOKEN|>" 
+    template1="<|START_OF_TURN_TOKEN|><|USER_TOKEN|>{message}<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>"
+    template2="{response}<|END_OF_TURN_TOKEN|>"
   if("Gemma-" in modelPath): # https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1
     template0="<start_of_turn>user{system}</end_of_turn>" 
     template1="<start_of_turn>user{message}</end_of_turn><start_of_turn>model"
-    template2="{response}</end_of_turn>"
-  if("Mixtral-8x7b-instruct" in modelPath): # https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1
+    template2="{response}</end_of_turn>"      
+  if("Mixtral-8x22B-Instruct" in modelPath): # AutoTokenizer: <s>[INST] U1[/INST] A1</s>[INST] U2[/INST] A2</s>
     startOfString="<s>"
+    template0="[INST]{system}\n  [/INST] </s>"  
+    template1="[INST] {message}[/INST]"
+    template2=" {response}</s>"
+  if("Mixtral-8x7b-instruct" in modelPath): # https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1
+    startOfString="<s>"                     # AutoTokenzizer: <s> [INST] U1 [/INST]A1</s> [INST] U2 [/INST]A2</s>
     template0=" [INST]{system}\n  [/INST] </s>"  
     template1=" [INST] {message} [/INST]"
     template2=" {response}</s>"
@@ -160,7 +170,11 @@ def extend_prompt(message="", history=None, system=None, RAGAddon=None, system2=
   if(("Discolm_german_7b" in modelPath) or ("SauerkrautLM-7b-HerO" in modelPath)):  #https://huggingface.co/VAGOsolutions/SauerkrautLM-7b-HerO
     template0="<|im_start|>system\n{system}<|im_end|>\n"
     template1="<|im_start|>user\n{message}<|im_end|>\n<|im_start|>assistant\n"
-    template2="{response}<|im_end|>\n"
+    template2="{response}<|im_end|>\n"    
+  if("Llama-3-SauerkrautLM-8b-Instruct" in modelPath):  #https://huggingface.co/VAGOsolutions/SauerkrautLM-7b-HerO
+    template0="<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n{system}<|eot_id|>"
+    template1="<|start_header_id|>user<|end_header_id|>\n\n{message}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
+    template2="{response}<|eot_id|>\n"        
   if("WizardLM-13B-V1.2" in modelPath): #https://huggingface.co/WizardLM/WizardLM-13B-V1.2
     template0="{system} " #<s>
     template1="USER: {message} ASSISTANT: "
@@ -178,7 +192,7 @@ def extend_prompt(message="", history=None, system=None, RAGAddon=None, system2=
     for user_message, bot_response in history[-historylimit:]:
       if user_message is None: user_message = "" 
       if bot_response is None: bot_response = ""
-      bot_response = re.sub("\n\n<details>((.|\n)*?)</details>","", bot_response) # remove RAG-compontents
+      bot_response = re.sub("\n\n<details>.*?</details>","", bot_response, flags=re.DOTALL) # remove RAG-compontents
       if removeHTML==True: bot_response = re.sub("<(.*?)>","\n", bot_response) # remove HTML-components in general (may cause bugs with markdown-rendering)
       if user_message is not None: prompt += template1.format(message=user_message[:zeichenlimit])  
       if bot_response is not None: prompt += template2.format(response=bot_response[:zeichenlimit])