fixed tiny evaluation bug

backend/config.json

@@ -17,7 +17,7 @@
     "search_type" : "similarity",
     "reranking" : false,
     "long_context_reorder" : false,
-    "structured_query": false,
+    "structured_query": true,
     "use_chroma_for_saving_metadata": false,
     "chunk_size": 1000,
     "chroma_metadata_dir": "../data/chroma_db_metadata"

evaluation/evaluation_results.csv

@@ -5,6 +5,9 @@ Snowflake/snowflake-arctic-embed-l,phi3,No LLM filtering,,0.6475524475524476,1.0
 Snowflake/snowflake-arctic-embed-l,llama3,False,,0.6503587153587154,0.9997150997150998,0.8818890986108248
 Snowflake/snowflake-arctic-embed-l,llama3,True,,0.6503587153587154,0.9997150997150998,0.8818890986108248
 Snowflake/snowflake-arctic-embed-l,llama3,No LLM filtering,,0.6475524475524476,1.0,0.8036501822692299
+Alibaba-NLP/gte-large-en-v1.5,llama3,False,,0.6481719248764702,0.9935147834769047,0.8124944306054139
+Alibaba-NLP/gte-large-en-v1.5,llama3,True,,0.6481719248764702,0.9935147834769047,0.8124944306054139
+Alibaba-NLP/gte-large-en-v1.5,llama3,No LLM filtering,,0.6481719248764702,0.9935147834769047,0.8124944306054139
 BAAI/bge-large-en-v1.5,llama3,False,512_chunk,0.6547227323742475,0.9979275073972045,0.8259522762251688
 BAAI/bge-large-en-v1.5,llama3,False,similarity_score_threshold_search,0.7046654902068915,0.9655886894421927,0.8152488113948533
 BAAI/bge-large-en-v1.5,llama3,False,mmr_search,0.6659358738904193,0.9911131517192123,0.8500771034849167

evaluation/run_all_training.py

@@ -30,6 +30,7 @@
         "BAAI/bge-large-en-v1.5",
         "BAAI/bge-base-en-v1.5",
         "Snowflake/snowflake-arctic-embed-l",
+        "Alibaba-NLP/gte-large-en-v1.5"
     ]
     list_of_llm_models = ["llama3", "phi3"]
 

evaluation/training_utils.py

@@ -175,7 +175,8 @@ def run_experiments(self):
             client = chromadb.PersistentClient(path=self.config["persist_dir"])
 
             # Note : I was not sure how to move this to the next loop, we need the QA setup going forward..
-            if os.path.exists(self.config["persist_dir"]):
+            # Check if the chroma db as well as metadata files exist.
+            if os.path.exists(self.config["persist_dir"]) and os.path.exists(main_experiment_directory/"all_dataset_description.csv"):
                 # load the qa from the persistent database if it exists. Disabling training does this for us.
                 self.config["training"] = False
 
@@ -189,6 +190,7 @@ def run_experiments(self):
                 qa_dataset, _ = qa_dataset_handler.setup_vector_db_and_qa()
                 self.config["training"] = True
             else:
+                self.config["training"] = True
                 qa_dataset_handler = QASetup(
                     config=self.config,
                     data_type=self.config["type_of_data"],

ollama/get_ollama.sh

@@ -7,5 +7,5 @@ while [ "$(ollama list | grep 'NAME')" == "" ]; do
 done
 
 
-ollama run llama3
+ollama run llama3.1
 tail -f /dev/null