Merge branch 'GrandeCapo93-develop'

hyde.py

@@ -1,24 +1,86 @@
-import json
-import langchain
-from cat.mad_hatter.decorators import hook
+from langchain.chains import LLMChain
+from langchain.prompts import PromptTemplate
+
 from cat.log import log
+from cat.mad_hatter.decorators import hook
 
 
-with open("cat/plugins/ccat_hyde/settings.json", "r") as json_file:
-    settings = json.load(json_file)
+# Keys
+HYDE_ANSWER       = "hyde_answer"
+AVERAGE_EMBEDDING = "average_embedding"
 
 
 @hook(priority=1)
 def cat_recall_query(user_message, cat):
 
+    # Acquire settings
+    settings = cat.mad_hatter.get_plugin().load_settings()
+    log.debug(f" --------- ACQUIRE SETTINGS ---------")
+    log.debug(f"settings: {settings}")
+
     # Make a prompt from template
-    hypothesis_prompt = langchain.PromptTemplate(
+    hypothesis_prompt = PromptTemplate(
         input_variables=["input"],
         template=settings["hyde_prompt"]
     )
 
     # Run a LLM chain with the user message as input
-    hypothesis_chain = langchain.chains.LLMChain(prompt=hypothesis_prompt, llm=cat._llm)
+    hypothesis_chain = LLMChain(prompt=hypothesis_prompt, llm=cat._llm)
     answer = hypothesis_chain(user_message)
-    log(answer, "INFO")
-    return answer["text"]
+
+    # Save HyDE answer in working memory
+    cat.working_memory[HYDE_ANSWER] = answer["text"]
+
+    log.debug("------------- HYDE -------------")
+    log.debug(f"user message: {user_message}")
+    log.debug(f"hyde answer: {answer['text']}")
+
+    return user_message
+
+
+# Calculates the average between the user's message embedding and the Hyde response embedding
+def _calculate_vector_average(config: dict, cat):
+
+    # If hyde answer exists, calculate and set average embedding
+    if HYDE_ANSWER in cat.working_memory.keys():
+
+       # Get user message embedding
+        user_embedding = config['embedding']
+
+        # Calculate hyde embedding from hyde answer
+        hyde_answer = cat.working_memory[HYDE_ANSWER]
+        hyde_embedding = cat.embedder.embed_query(hyde_answer)
+
+        # Calculate average embedding and stores it into a working memory
+        average_embedding = [(x + y)/2 for x, y in zip(user_embedding, hyde_embedding)]
+        cat.working_memory[AVERAGE_EMBEDDING] = average_embedding
+
+        log.debug(f" --------- CALCULATE AVERAGE ---------")
+        log.debug(f"hyde answer:       {hyde_answer}")
+        log.debug(f"user_embedding:    {user_embedding}")
+        log.debug(f"hyde_embedding:    {hyde_embedding}")
+        log.debug(f"average_embedding: {average_embedding}")
+
+        # Delete Hyde Answer from working memory
+        del cat.working_memory[HYDE_ANSWER]
+
+    # If average embedding exists, set the embedding
+    if AVERAGE_EMBEDDING in cat.working_memory.keys():
+        average_embedding = cat.working_memory[AVERAGE_EMBEDDING]
+        config['embedding'] = average_embedding
+
+        log.debug(f" --------- SET EMBEDDING ---------")
+        log.debug(f"average_embedding: {average_embedding}")
+
+
+@hook(priority=1)
+def before_cat_recalls_episodic_memories(config: dict, cat):
+    _calculate_vector_average(config, cat)
+
+@hook(priority=1)
+def before_cat_recalls_declarative_memories(config: dict, cat):
+    _calculate_vector_average(config, cat)
+
+@hook(priority=1)
+def before_cat_recalls_procedural_memories(config: dict, cat):
+    _calculate_vector_average(config, cat)

plugin.json

@@ -1,8 +1,8 @@
 {
     "name": "Hypothetical Document Embedding",
-    "version": "0.0.4",
+    "version": "0.0.5",
     "description": "Official plugin of the Cheshire Cat to add the Hypothetical Document Embedding (HyDE) technique",
-    "author_name": "Nicola Corbellini",
+    "author_name": "Nicola Corbellini, Salvatore Mirlocca, Massimiliano D'Amico",
     "author_url": "",
     "plugin_url": "https://github.com/Furrmidable-Crew/ccat_hyde",
     "tags": "hyde, llm, cheshire-cat, embedding",

settings.json

@@ -0,0 +1,3 @@
+{
+  "hyde_prompt": "You will be given a sentence.\n    If the sentence is a question, convert it to a plausible answer. If the sentence does not contain a question, \n    just repeat the sentence as is without adding anything to it.\n\n    Examples:\n    - what furniture there is in my room? --> In my room there is a bed, a wardrobe and a desk with my computer\n    - where did you go today --> today I was at school\n    - I like ice cream --> I like ice cream\n    - how old is Jack --> Jack is 20 years old\n\n    Sentence:\n    - {input} -->"
+}

settings.py

@@ -1,5 +1,5 @@
-from cat.mad_hatter.decorators import plugin
 from pydantic import BaseModel, Field
+from cat.mad_hatter.decorators import plugin
 
 
 class MySettings(BaseModel):
@@ -20,7 +20,6 @@ class MySettings(BaseModel):
                 extra={"type": "TextArea"}
         )
 
-
 @plugin
 def settings_schema():
     return MySettings.schema()