Enhanced code and added new tools

configs/config.yaml

@@ -1,20 +1,19 @@
 # =================== Lord Of Large Language Models Configuration file =========================== 
-version: 27
+version: 30
 binding_name: null
 model_name: null
 
-# Enables gpu usage
-enable_gpu: true
+
 
 # Host information
 host: localhost
-port: 9601
+port: 9600
 
 # Genreration parameters 
 discussion_prompt_separator: "!@>"
 seed: -1
 n_predict: 1024
-ctx_size: 4048
+ctx_size: 4084
 min_n_predict: 256
 temperature: 0.9
 top_k: 50
@@ -29,25 +28,43 @@ personalities: ["generic/lollms"]
 active_personality_id: 0
 override_personality_model_parameters: false #if true the personality parameters are overriden by those of the configuration (may affect personality behaviour) 
 
-# Extensions
 extensions: []
 
-# User infos
 user_name: user
 user_description: ""
 use_user_name_in_discussions: false
-user_avatar: default_user
+user_avatar: default_user.svg
+use_user_informations_in_discussion: false
+
+# UI parameters
+db_path: database.db
 
-# Automatic update
-auto_update: false
+# Automatic updates
+debug: False
+auto_update: true
 auto_save: true
-debug: false
+auto_title: false
+# Enables gpu usage
+enable_gpu: true
+# Automatically open the browser
+auto_show_browser: true
+
+# Audio
+audio_in_language: 'en-US'
+audio_out_voice: null
+auto_speak: false
+audio_pitch: 1
+audio_auto_send_input: true
+audio_silenceTimer: 5000
 
 # Data vectorization
+use_discussions_history: false # Activate vectorizing previous conversations
+summerize_discussion: false # activate discussion summary (better but adds computation time)
+max_summary_size: 512 # in tokens
+data_vectorization_visualize_on_vectorization: false
 use_files: true # Activate using files
-data_vectorization_visualize_on_vectorization: false # if active, the vectorizer wil lshow data whenever index is called
 data_vectorization_activate: true # To activate/deactivate data vectorization
-data_vectorization_method: "ftidf_vectorizer" #"model_embedding" or "ftidf_vectorizer"
+data_vectorization_method: "tfidf_vectorizer" #"model_embedding" or "tfidf_vectorizer"
 data_visualization_method: "PCA" #"PCA" or "TSNE"
 data_vectorization_save_db: False # For each new session, new files
 data_vectorization_chunk_size: 512 # chunk size

lollms/app.py

@@ -11,9 +11,12 @@
 from typing import Callable
 from ascii_colors import ASCIIColors
 from pathlib import Path
-
+from datetime import datetime
+from functools import partial
 import subprocess
 import importlib
+import sys
+
 class LollmsApplication:
     def __init__(
                     self, 
@@ -121,10 +124,65 @@ def build_long_term_skills_memory(self):
             )
         return self.long_term_memory
 
-    def learn_from_discussion(self, discussion):        
-        learned_skill = self.model.generate(f"{discussion}\n\n!@>What should we learn from this discussion?!@>Summerizer: Here is a summary of the most important informations extracted from the discussion:\n",)
-        return learned_skill
-
+    def process_chunk(
+                        self, 
+                        chunk:str, 
+                        message_type,
+                        parameters:dict=None, 
+                        metadata:list=None, 
+                        personality=None
+                    ):
+
+        pass
+
+    def default_callback(self, chunk, type, generation_infos:dict):
+        if generation_infos["nb_received_tokens"]==0:
+            self.start_time = datetime.now()
+        dt =(datetime.now() - self.start_time).seconds
+        if dt==0:
+            dt=1
+        spd = generation_infos["nb_received_tokens"]/dt
+        ASCIIColors.green(f"Received {generation_infos['nb_received_tokens']} tokens (speed: {spd:.2f}t/s)              ",end="\r",flush=True) 
+        sys.stdout = sys.__stdout__
+        sys.stdout.flush()
+        if chunk:
+            generation_infos["generated_text"] += chunk
+        antiprompt = self.personality.detect_antiprompt(generation_infos["generated_text"])
+        if antiprompt:
+            ASCIIColors.warning(f"\nDetected hallucination with antiprompt: {antiprompt}")
+            generation_infos["generated_text"] = self.remove_text_from_string(generation_infos["generated_text"],antiprompt)
+            return False
+        else:
+            generation_infos["nb_received_tokens"] += 1
+            generation_infos["first_chunk"]=False
+            # if stop generation is detected then stop
+            if not self.cancel_gen:
+                return True
+            else:
+                self.cancel_gen = False
+                ASCIIColors.warning("Generation canceled")
+                return False
+
+    def learn_from_discussion(self, title, discussion, nb_gen=None, callback=None):
+        if self.config.summerize_discussion:
+            prompt = f"!@>discussion:\n--\n!@>title:{title}!@>content:\n{discussion}\n--\n!@>question:What should we learn from this discussion?!@>Summerizer: Here is a summary of the most important informations extracted from the discussion:\n"
+            if nb_gen is None:
+                nb_gen = self.config.max_summary_size
+            generated = ""
+            gen_infos={
+                "nb_received_tokens":0,
+                "generated_text":"",
+                "processing":False,
+                "first_chunk": True,
+            }
+            if callback is None:
+                callback = partial(self.default_callback, generation_infos=gen_infos)
+            self.model.generate(prompt, nb_gen, callback)
+            if self.config.debug:
+                ASCIIColors.yellow(gen_infos["generated_text"])
+            return gen_infos["generated_text"]
+        else:
+            return  discussion
 
     def remove_text_from_string(self, string, text_to_find):
         """

lollms/binding.py

@@ -392,6 +392,7 @@ def generate(self,
             verbose (bool, optional): If true, the code will spit many informations about the generation process. Defaults to False.
         """
         pass
+
     def tokenize(self, prompt:str):
         """
         Tokenizes the given prompt using the model's tokenizer.
@@ -489,9 +490,9 @@ def clear_cuda():
         print(LLMBinding.vram_usage())
         try:
             torch.cuda.empty_cache()
+            ASCIIColors.green("Cleared cache")
         except Exception as ex:
             ASCIIColors.error("Couldn't clear cuda memory")
-        ASCIIColors.red("Cleared cache")
         ASCIIColors.red("*-*-*-*-*-*-*-*")
         ASCIIColors.red("Cuda VRAM usage")
         ASCIIColors.red("*-*-*-*-*-*-*-*")

lollms/configs/config.yaml

@@ -1,5 +1,5 @@
 # =================== Lord Of Large Language Models Configuration file =========================== 
-version: 28
+version: 30
 binding_name: null
 model_name: null
 
@@ -59,6 +59,8 @@ audio_silenceTimer: 5000
 
 # Data vectorization
 use_discussions_history: false # Activate vectorizing previous conversations
+summerize_discussion: false # activate discussion summary (better but adds computation time)
+max_summary_size: 512 # in tokens
 data_vectorization_visualize_on_vectorization: false
 use_files: true # Activate using files
 data_vectorization_activate: true # To activate/deactivate data vectorization

lollms/personality.py

@@ -445,7 +445,7 @@ def generate_with_images(self, prompt, images, max_size, temperature = None, top
                                 ).strip()    
         return self.bot_says
 
-    def generate(self, prompt, max_size, temperature = None, top_k = None, top_p=None, repeat_penalty=None, callback=None, debug=False, show_progress=False ):
+    def generate(self, prompt, max_size, temperature = None, top_k = None, top_p=None, repeat_penalty=None, repeat_last_n=None, callback=None, debug=False, show_progress=False ):
         ASCIIColors.info("Text generation started: Warming up")
         self.nb_received_tokens = 0
         self.bot_says = ""
@@ -460,6 +460,7 @@ def generate(self, prompt, max_size, temperature = None, top_k = None, top_p=Non
                                 top_k=self.model_top_k if top_k is None else top_k,
                                 top_p=self.model_top_p if top_p is None else top_p,
                                 repeat_penalty=self.model_repeat_penalty if repeat_penalty is None else repeat_penalty,
+                                repeat_last_n = self.model_repeat_last_n if repeat_last_n is None else repeat_last_n,
                                 ).strip()    
         return self.bot_says
 
@@ -669,14 +670,13 @@ def add_file(self, path, callback=None):
         if path.suffix in [".png",".jpg",".gif",".bmp",".webp"]:
             if self.callback:
                 try:
-                    if callback:
-                        pth = str(path).replace("\\","/").split('/')
-                        if "uploads" in pth:
-                            idx = pth.index("uploads")
-                            pth = "/".join(pth[idx:])
-                            self.new_message("",MSG_TYPE.MSG_TYPE_FULL)
-                            output = f'<img src="{pth}" width="300">\n\n'
-                            self.full(output)
+                    pth = str(path).replace("\\","/").split('/')
+                    if "uploads" in pth:
+                        idx = pth.index("uploads")
+                        pth = "/".join(pth[idx:])
+                        self.new_message("",MSG_TYPE.MSG_TYPE_FULL)
+                        output = f'<img src="{pth}" width="300">\n\n'
+                        self.full(output)
 
                     if self.model.binding_type not in [BindingType.TEXT_IMAGE, BindingType.TEXT_IMAGE_VIDEO]:
                         self.step_start("Understanding image (please wait)")
@@ -693,7 +693,6 @@ def add_file(self, path, callback=None):
                     else:
                         self.step_start("Importing image (please wait)")
                         self.step_end("Importing image (please wait)")
-                        self.full(output)
 
                 except Exception as ex:
                     trace_exception(ex)
@@ -1684,11 +1683,11 @@ def save_config_file(self, path, data):
         with open(path, 'w') as file:
             yaml.dump(data, file)
 
-    def generate_with_images(self, prompt, images, max_size, temperature = None, top_k = None, top_p=None, repeat_penalty=None, callback=None, debug=False ):
-        return self.personality.generate_with_images(prompt, images, max_size, temperature, top_k, top_p, repeat_penalty, callback, debug=debug)
+    def generate_with_images(self, prompt, images, max_size, temperature = None, top_k = None, top_p=None, repeat_penalty=None, repeat_last_n=None, callback=None, debug=False ):
+        return self.personality.generate_with_images(prompt, images, max_size, temperature, top_k, top_p, repeat_penalty, repeat_last_n, callback, debug=debug)
 
-    def generate(self, prompt, max_size, temperature = None, top_k = None, top_p=None, repeat_penalty=None, callback=None, debug=False ):
-        return self.personality.generate(prompt, max_size, temperature, top_k, top_p, repeat_penalty, callback, debug=debug)
+    def generate(self, prompt, max_size, temperature = None, top_k = None, top_p=None, repeat_penalty=None, repeat_last_n=None, callback=None, debug=False ):
+        return self.personality.generate(prompt, max_size, temperature, top_k, top_p, repeat_penalty, repeat_last_n, callback, debug=debug)
 
     def run_workflow(self, prompt:str, previous_discussion_text:str="", callback: Callable[[str, MSG_TYPE, dict, list], bool]=None):
         """
@@ -2133,7 +2132,7 @@ def yes_no(self, question: str, context:str="", max_answer_length: int = 50) ->
         Returns:
             bool: True if the user prompt is asking to generate an image, False otherwise.
         """
-        return self.multichoice_question(question, ["no","yes"],context, max_answer_length)==1
+        return self.multichoice_question(question, ["no","yes"], context, max_answer_length)>0
 
     def multichoice_question(self, question: str, possible_answers:list, context:str = "", max_answer_length: int = 50) -> int:
         """
@@ -2148,34 +2147,39 @@ def multichoice_question(self, question: str, possible_answers:list, context:str
             int: Index of the selected option within the possible_ansers list. Or -1 if there was not match found among any of them.
         """
         if context!="":
-            template = """!@>instruction:
-Act as prompt classifier, a tool capable of classifying the user prompt.
-!@>Context:                               
+            template = """!@>instructions:
+Answer this multi choices question.
+Answer with an id from the possible answers.
+Do not answer with an id outside this possible answers.
+!@>question: {{question}}
+!@>possible answers:
+{{choices}}
+!@>Context:
 {{context}}
+!@>answer:"""
+        else:
+            template = """!@>instructions:
+Answer this multi choices question.
+Answer with an id from the possible answers.
+Do not answer with an id outside this possible answers.
 !@>question: {{question}}
 !@>choices:
-{{choices}} 
-!@>prompt analyzer: After analyzing the user prompt, the most suitable choice is choice number """
-        else:
-            template = """!@>Instructions:
-Act as  prompt analyzer, a tool capable of analyzing the user prompt and answering yes or no this prompt asking to generate an image.
-!@>instruction: {{question}}
-!@>choices: {{choices}} 
-!@>prompt analyzer: After analyzing the user prompt, the most suitable choice is choice number """
+{{choices}}
+!@>answer:"""
 
         pr  = PromptReshaper(template)
         prompt = pr.build({
                 "context":context,
                 "question":question,
-                "choices":"\n".join([f"{i} - {possible_answer}" for i,possible_answer in enumerate(possible_answers)])
+                "choices":"\n".join([f"{i}. {possible_answer}" for i, possible_answer in enumerate(possible_answers)])
                 }, 
                 self.personality.model.tokenize, 
                 self.personality.model.detokenize, 
                 self.personality.model.config.ctx_size,
                 ["previous_discussion"]
                 )
-        gen = self.generate(prompt, max_answer_length).strip().replace("</s>","").replace("<s>","")
-        selection = gen.split()[0]
+        gen = self.generate(prompt, max_answer_length, temperature=0.1, top_k=50, top_p=0.9, repeat_penalty=1.0, repeat_last_n=50).strip().replace("</s>","").replace("<s>","")
+        selection = gen.strip().split()[0].replace(",","")
         self.print_prompt("Multi choice selection",prompt+gen)
         try:
             return int(selection)