v1-fix

agent/agents.py

@@ -33,8 +33,6 @@ def cohere_llm():
         model="command-r-plus",
         max_tokens=2048,
         temperature=os.environ["temperature"],
-        # p=os.environ["top_p"],
-        # frequency_penalty=os.environ["frequency_penalty"],
     )
 
 

rag.py

@@ -35,34 +35,6 @@ def replace_nl(input: str) -> str:
     return input.replace('\r\n', '<br>').replace('\n', '<br>').replace('\r', '<br>')
 
 
-def rag(agent: Agent, q: str, retrieve_document_count: int):
-    k = retrieve_document_count
-
-    context_doc = retrieve(agent.embedding, q, k)
-
-    prompt_template = PromptTemplate(
-        input_variables=["context", "question"],
-        template=os.environ["RAG_TEMPLATE"]
-    )
-
-    llm_chain = LLMChain(
-        llm=agent.llm,
-        prompt=prompt_template,
-        verbose=False
-    )
-
-    # llm_chain = prompt_template | agent.llm
-
-    result: dict[str, Any] = llm_chain.invoke(
-        input={
-            "question": q,
-            "context": context_doc
-        }
-    )
-
-    return result["text"]
-
-
 def rewrite(agent: Agent, q: str, prompt: str) -> list[str]:
     prompt_template = PromptTemplate(
         input_variables=["question"],

rag_langchain.py

@@ -5,7 +5,7 @@ from gptcache import Cache
 from gptcache.manager.factory import manager_factory
 from gptcache.processor.pre import get_prompt
 from langchain.retrievers import ContextualCompressionRetriever
-from langchain_cohere import CohereRerank
+from langchain_cohere import CohereRerank, CohereEmbeddings
 from langchain_community.cache import GPTCache
 from langchain_core.language_models import BaseChatModel
 from langchain_core.prompts import PromptTemplate
@@ -17,7 +17,7 @@ from agent.Agent import Agent
 from agent.agents import deepinfra_chat, \
     together_ai_chat, groq_chat, cohere_llm
 from emdedd.Embedding import Embedding
-from emdedd.embeddings import chroma_embedding, cohere_embeddings
+from emdedd.MongoEmbedding import EmbeddingDbConnection, MongoEmbedding
 from prompt.prompt_store import PromptStore
 from rag import rag_chain
 
@@ -25,7 +25,7 @@ load_dotenv()
 
 
 class LangChainRAG:
-    embedding: tuple[Embedding]
+    embedding: Embedding
     llms: dict[str, BaseChatModel]
     retriever: BaseRetriever
     prompt_template: PromptTemplate

task_splitting.py

@@ -1,101 +0,0 @@
-import datetime
-from time import sleep
-
-from langchain.chains import LLMChain
-from langchain_core.prompts import PromptTemplate
-
-from agent.Agent import Agent
-from agent.agents import chat_groq_llama3_70
-from emdedd.embeddings import cohere_embeddings, chroma_embedding, embed_zakonnik_prace
-from promts import for_tree_llama3_rag_sub, for_tree_llama3_rag_tree, for_tree_llama3_rag_group
-from retrieval import retrieve_with_rerank
-from questions import questions
-
-
-def rag_tree(agent: Agent, q: str, retrieve_document_count: int) -> str:
-    tree_template = PromptTemplate(
-        input_variables=["context", "question"],
-        template=for_tree_llama3_rag_tree
-    )
-
-    context_doc = retrieve_with_rerank(agent.embedding, q, retrieve_document_count * 2)
-
-    sub_qs = LLMChain(
-        llm=agent.llm,
-        prompt=tree_template,
-        verbose=False
-    ).invoke(
-        input={
-            "question": q,
-            "context": context_doc
-        }
-    )["text"]
-
-    print(sub_qs)
-    sleep(60)
-
-    print("_________")
-
-    sub_template = PromptTemplate(
-        input_variables=["context", "question"],
-        template=for_tree_llama3_rag_sub
-    )
-
-    sub_answers: dict[str, str] = {}
-
-    for sub_q in sub_qs.splitlines():
-        if "?" not in sub_q: continue
-        print(sub_q)
-        sub_answers[sub_q] = LLMChain(
-            llm=agent.llm,
-            prompt=sub_template,
-            verbose=False
-        ).invoke(
-            input={
-                "question": sub_q,
-                "context": retrieve_with_rerank(agent.embedding, sub_q, retrieve_document_count)
-            }
-        )["text"]
-        print(sub_answers[sub_q])
-        sleep(60)
-
-
-    final_template = PromptTemplate(
-        input_variables=["context", "question", "subs"],
-        template=for_tree_llama3_rag_group
-    )
-
-    result = LLMChain(
-        llm=agent.llm,
-        prompt=final_template,
-        verbose=True
-    ).invoke(
-        input={
-            "question": q,
-            "context": context_doc,
-            "subs": sub_answers.items()
-        }
-    )
-
-    return result["text"]
-
-
-def tree_of_thought(name: str, agent: Agent, emded: bool = False, retrieve_document_count=5):
-    try:
-        result_file = open(name + "_test.md", "a")
-        if emded:
-            embed_zakonnik_prace(agent.embedding)
-        for q in questions:
-            print("--- Q: " + q)
-            result_file.write("\n\n| " + name + str(datetime.datetime.now()) + " | " + q + " |")
-            result_file.write("\n|-------|-----------|")
-
-            answer = rag_tree(agent, q, retrieve_document_count)
-            print(answer)
-            result_file.write(
-                "\n| tree | " + answer.replace('\r\n', '<br>').replace('\n', '<br>').replace('\r', '<br>') + " |")
-            sleep(60)
-    finally:
-        result_file.write("\n\n")
-        result_file.flush()
-        result_file.close()