Better .env with more settings

example.env

@@ -1,6 +1,25 @@
+#       DIRECTORY SETTINGS      #
+
 PERSIST_DIRECTORY=db
 DOCUMENTS_DIRECTORY=source_documents
-LLAMA_EMBEDDINGS_MODEL=models/ggml-model-q4_0.bin
+
+#       LLM MODEL SETTINGS      #
+
+# Your LLM type (GPT4All or LlamaCpp)
 MODEL_TYPE=GPT4All
+
+# Absolute path to your llama supported embeddings model.
+LLAMA_EMBEDDINGS_MODEL=models/ggml-model-q4_0.bin
+
+# Absolute path to your GPT4All or LlamaCpp model
 MODEL_PATH=models/ggml-gpt4all-j-v1.3-groovy.bin
-MODEL_N_CTX=1000
+
+# Context size for both the vector datbase and the llm seperately in one value
+# Double this value if you are getting context size errors
+MODEL_N_CTX=1024
+# Temperature range of 0=Logical to 1=Creative 
+MODEL_TEMP=0.8
+
+# Stop based on certain characters or strings.
+# \n is for new lines and \t is for tabs.
+MODEL_STOP=\n,\t

startLLM.py

@@ -12,6 +12,8 @@
 model_type = os.environ.get('MODEL_TYPE')
 model_path = os.environ.get('MODEL_PATH')
 model_n_ctx = os.environ.get('MODEL_N_CTX')
+model_temp = os.environ.get('MODEL_TEMP')
+model_stop = os.environ.get('MODEL_STOP').split(",")
 
 def main():
     # Load stored vectorstore
@@ -32,7 +34,7 @@ def main():
     match model_type:
         case "LlamaCpp":
             from langchain.llms import LlamaCpp
-            llm = LlamaCpp(model_path=local_path, n_ctx=model_n_ctx, callbacks=callbacks, verbose=True)
+            llm = LlamaCpp(model_path=local_path, n_ctx=model_n_ctx, temperature=model_temp, stop=model_stop, callbacks=callbacks, verbose=True)
         case "GPT4All":
             from langchain.llms import GPT4All
             llm = GPT4All(model=local_path, n_ctx=model_n_ctx, callbacks=callbacks, verbose=True, backend='gptj')