sync

configs/config.yaml

@@ -1,5 +1,5 @@
 # =================== Lord Of Large Language Multimodal Systems Configuration file =========================== 
-version: 120
+version: 121
 binding_name: null
 model_name: null
 model_variant: null
@@ -251,6 +251,7 @@ rag_check_new_files_at_startup: false #if true, the vectorizer will automaticall
 rag_preprocess_chunks: false #if true, an LLM will preprocess the content of the chunk before writing it in a simple format
 rag_activate_multi_hops: false #if true, we use multi hops algorithm to do multiple researches until the AI has enough data
 rag_min_nb_tokens_in_chunk: 10 #this removed any useless junk ith less than x tokens
+rag_max_n_hops: 3 #We set the maximum number of hop in multi hops rag
 
 activate_skills_lib: false # Activate vectorizing previous conversations
 skills_lib_database_name: "default" # Default skills database

lollms/app.py

@@ -1053,10 +1053,21 @@ class LollmsApplication(LoLLMsCom):
                             f"{start_header_id_template}Documentation{end_header_id_template}"])
                         documentation += f"{separator_template}"
                     results = []
-                    for db in self.active_rag_dbs:
-                        v = db["vectorizer"]
-                        r=v.search(query)
-                        results+=r
+                    recovered_ids=[[]*len(self.active_rag_dbs)]
+                    i=0
+                    hop_id = 0
+                    while( len(results)<self.config.rag_n_chunks and hop_id<self.config.rag_max_n_hops):
+                        hop_id +=1
+                        for db in self.active_rag_dbs:
+                            v = db["vectorizer"]
+                            r=v.search(query, self.config.rag_n_chunks, recovered_ids[i])
+                            recovered_ids[i].append([rg.chunk_id for rg in r])
+                            if self.config.rag_activate_multi_hops:
+                                r = [rg for rg in r if self.personality.verify_rag_entry(query, rg.content)]
+                            results+=r
+                            i+=1
+                        if len(results)>=self.config.rag_n_chunks:
+                            break
                     n_neighbors = self.active_rag_dbs[0]["vectorizer"].n_neighbors
                     sorted_results = sorted(results, key=lambda x: x.distance)[:n_neighbors]
 
@@ -1321,7 +1332,7 @@ class LollmsApplication(LoLLMsCom):
             "extra":""
         }    
         if self.config.debug:
-            ASCIIColors.hilight(documentation,"source_document_title", ASCIIColors.color_yellow, ASCIIColors.color_red, False)
+            ASCIIColors.highlight(documentation,"source_document_title", ASCIIColors.color_yellow, ASCIIColors.color_red, False)
         # Return the prepared query, original message content, and tokenized query
         return prompt_data, current_message.content, tokens, context_details, internet_search_infos                
 

lollms/configs/config.yaml

@@ -1,5 +1,5 @@
 # =================== Lord Of Large Language Multimodal Systems Configuration file =========================== 
-version: 120
+version: 121
 binding_name: null
 model_name: null
 model_variant: null
@@ -251,6 +251,7 @@ rag_check_new_files_at_startup: false #if true, the vectorizer will automaticall
 rag_preprocess_chunks: false #if true, an LLM will preprocess the content of the chunk before writing it in a simple format
 rag_activate_multi_hops: false #if true, we use multi hops algorithm to do multiple researches until the AI has enough data
 rag_min_nb_tokens_in_chunk: 10 #this removed any useless junk ith less than x tokens
+rag_max_n_hops: 3 #We set the maximum number of hop in multi hops rag
 
 activate_skills_lib: false # Activate vectorizing previous conversations
 skills_lib_database_name: "default" # Default skills database

lollms/personality.py

@@ -3798,10 +3798,14 @@ fetch('/open_file', {
     def compress_html(self, code):
         return compress_html(code)
 
+
+
+
 # ===========================================================
     def select_model(self, binding_name, model_name):
         self.personality.app.select_model(binding_name, model_name)
-
+    def verify_rag_entry(self, query, rag_entry):
+        return self.yes_no("Does the text entry contain the answer to the query?", self.system_custom_header("Query")+query+"\n"+self.system_custom_header("text entry")+":\n"+rag_entry)
     # Properties ===============================================
     @property
     def start_header_id_template(self) -> str:

lollms/services/diffusers/lollms_diffusers.py

@@ -33,7 +33,9 @@ import shutil
 from tqdm import tqdm
 import threading
 
-
+def adjust_dimensions(value: int) -> int:
+    """Adjusts the given value to be divisible by 8."""
+    return (value // 8) * 8
 
 def download_file(url, folder_path, local_filename):
     # Make sure 'folder_path' exists
@@ -215,7 +217,8 @@ class LollmsDiffusers(LollmsTTI):
             sc = self.get_scheduler_by_name(sampler_name)
             if sc:
                 self.model.scheduler = sc
-
+        width = adjust_dimensions(width)
+        height = adjust_dimensions(height)
         if output_path is None:
             output_path = self.output_dir
         if seed!=-1: