updated files

app.py

@@ -283,29 +283,32 @@ def generate_inter_domain_comparison(metric='f1_score'):
         fig_global = None
     
     return comp_df, fig_global
-    
-    if best_results:
-        best_df = pd.DataFrame(best_results)
-        fig_global = px.bar(
-            best_df, x="Domain", y="Max F1 Score", 
-            color="Domain", 
-            text_auto='.4f',
-            hover_data=["Best Config"],
-            title="Peak Performance per Domain (Max F1 Score)"
-        )
-        fig_global.update_traces(textposition='outside')
-    else:
-        fig_global = None
-    
-    return comp_df, fig_global
 
 # --- 3. UI ---
-APP_VERSION = "v2.1.0-fixed"  # Version stamp to verify code is updated
+APP_VERSION = "v2.2.0"
+
+# Global constants used across all experiments
+GLOBAL_CONSTANTS = """
+**Global Constants (Applied to All Domains):**
+- **Summarization Model:** fangyuan/nq_abstractive_compressor
+- **Generator Model:** llama-3.1-8b-instant
+- **Generator Max Tokens:** 512
+- **Generator Temperature:** 0.2
+- **Generator API Provider:** Groq
+- **Generation LLM Context Budget:** 2000
+- **Judge Model:** llama-3.3-70b-versatile
+- **Judge Max Tokens:** 1024
+- **Judge Temperature:** 0
+- **Judge Sentence Attribution:** ENABLED
+"""
 
 with gr.Blocks(title="RAG Analytics Pro") as demo:
     gr.Markdown("## RAG Pipeline Analytics")
     gr.Markdown(f"**Data Source:** `{DATA_FOLDER}` | **Version:** {APP_VERSION}")
     
+    with gr.Accordion("Global Experiment Configuration", open=False):
+        gr.Markdown(GLOBAL_CONSTANTS)
+    
     with gr.Row():
         refresh_data_btn = gr.Button("Load/Refresh Data", variant="primary")
         status = gr.Textbox(label="Status (Check here for debug info)", interactive=False, scale=3)
@@ -400,6 +403,5 @@ print(f"Loading data from {DATA_FOLDER}...")
 startup_status = load_data()
 print(startup_status)
 
-# Launch Gradio app
-if __name__ == "__main__":
-    demo.launch(ssr_mode=False)
+# Launch Gradio app (for Hugging Face Spaces, this runs on import)
+demo.launch(ssr_mode=False)

config.py

@@ -43,6 +43,7 @@ COLUMN_MAP = {
     'embeddingmodel': 'embedding_model',
     'rerankermodel': 'reranker_model',
     'summarizationmodel': 'summarization_model',
+    'summarization': 'summarization',  # New column name (enabled/disabled)
     'chunkingstrategy': 'chunking_strategy',
     'chunksize': 'chunk_size',
     'overlap': 'overlap',
@@ -69,7 +70,7 @@ COLUMN_MAP = {
 # Metadata columns (excluded from constant/variable analysis)
 METADATA_COLUMNS = [
     'rmse_relevance', 'rmse_utilization', 'rmse_completeness',
-    'aucroc', 'f1_score', 'failed_samples', 
+    'aucroc', 'f1_score',
     'test_id', 'config_purpose', 'dataset_name'
 ]
 

data/Biomedical-pubmedqa.csv

@@ -1,7 +1,7 @@
-Test #,Configuration purpose,Subset(s),Embedding Model,Reranker Model,Summarization Model,Chunking Strategy,Chunk size,Overlap,Stride,Retreival Strategy,Alpha,Retr. K,Final K,Repacking,Summ. Max,Summ. Min,8B GPT Label,RMSE=trace relevance,RMSE=trace utilization,RMSE=trace completeness,AUCROC,F1-score,# Failed/Total Samples
-1,Efficiency Baseline,pubmedqa,NeuML/pubmedbert-base-embeddings,cross-encoder/ms-marco-MiniLM-L-6-v2,None,hard_cut,256,50,N/A,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3677,0.3011,0.5556,0.604,0.5049,32/100
-2,Chunking Proof,pubmedqa,NeuML/pubmedbert-base-embeddings,cross-encoder/ms-marco-MiniLM-L-6-v2,None,sliding_window,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3632,0.2886,0.5074,0.604,0.5049,29/100
-3,Reranking proof,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,None,sliding_window,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,long,0.3289,0.2663,0.6015,0.482,0.38,8/100
-4,Repacking proof,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,None,hard_cut,256,50,206,Hybrid,0.8,50,5,reverse,N/A,N/A,long,0.2752,0.252,0.6246,0.5951,0.449,8/100
-5,Prove Summarization,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,fangyuan/nq_abstractive_compressor,sliding_window,256,50,206,Hybrid,0.8,50,5,reverse,150,20,long_cot,0.4934,1.0537,0.5161,cannot compute,0,9/100
-6,Optimal Medical Hybrid,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,N/A,sliding_window,256,50,206,Hybrid,0.8,50,5,reverse,N/A,N/A,long,0.3223,0.2733,0.6561,0.5053,0.3542,13/100
+Test #,Configuration purpose,Subset(s),Embedding Model,Reranker Model,Summarization,Chunking Strategy,Chunk size,Overlap,Stride,Retreival Strategy,Alpha,Retr. K,Final K,Repacking,Summ. Max,Summ. Min,8B GPT Label,RMSE=trace relevance,RMSE=trace utilization,RMSE=trace completeness,AUCROC,F1-score,# Failed/Total Samples
+1,Efficiency Baseline,pubmedqa,NeuML/pubmedbert-base-embeddings,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,hard_cut,256,50,N/A,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3677,0.3011,0.5556,0.604,0.5049,32/100
+2,Chunking Proof,pubmedqa,NeuML/pubmedbert-base-embeddings,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,sliding_window,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3632,0.2886,0.5074,0.604,0.5049,29/100
+3,Reranking proof,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,disabled,sliding_window,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,long,0.3289,0.2663,0.6015,0.482,0.38,8/100
+4,Repacking proof,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,disabled,hard_cut,256,50,206,Hybrid,0.8,50,5,reverse,N/A,N/A,long,0.2752,0.252,0.6246,0.5951,0.449,8/100
+5,Prove Summarization,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,enabled,sliding_window,256,50,206,Hybrid,0.8,50,5,reverse,150,20,long_cot,0.4934,1.0537,0.5161,cannot compute,0,9/100
+6,Optimal Medical Hybrid,pubmedqa,NeuML/pubmedbert-base-embeddings,BAAI/bge-reranker-base,disabled,sliding_window,256,50,206,Hybrid,0.8,50,5,reverse,N/A,N/A,long,0.3223,0.2733,0.6561,0.5053,0.3542,13/100

data/Finance-finqa.csv

@@ -1,7 +1,7 @@
 Test #,Configuration purpose,Subset(s),Embedding Model,Reranker Model,Summarization Model,Chunking Strategy,Chunk size,Overlap,Stride,Retreival Strategy,Alpha,Retr. K,Final K,Repacking,Summ. Max,Summ. Min,8B GPT Label,RMSE=trace relevance,RMSE=trace utilization,RMSE=trace completeness,AUCROC,F1-score,# Failed/Total Samples
-1,Efficiency Baseline,finqa,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,N/A,hard-cut,512,50,N/A,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.1409,0.0831,0.6365,0.4263,0.099,18/100
-2,Prove Chunking,finqa,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,N/A,sliding-window,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.1667,0.1188,0.6431,0.4316,0.1176,23/100
-3,Prove Hybrid/Rerank,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,N/A,sliding-window,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,long,0.1316,0.0693,0.6763,0.4263,0.099,11/100
-4,Max Raw Context,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,N/A,hard-cut,512,100,412,Hybrid,0.6,50,5,reverse,N/A,N/A,long,0.1947,0.0795,0.7239,0.4316,0.1176,14/100
-5,Golden Setup,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,fangyuan/nq_abstractive_compressor,sliding-window,512,100,412,Hybrid,0.6,50,5,reverse,200,20,long_cot,0.4158,0.8363,0.7073,Cannot compute (insufficient class variance),0,4/100
-6,Optimized Financial,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,N/A,sliding-window,512,100,412,Hybrid,0.8,50,3,forward,N/A,N/A,long,0.2468,0.1679,0.6177,0.5474,0.1731,6/100
+1,Efficiency Baseline,finqa,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,sentence-level,512,50,N/A,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.1409,0.0831,0.6365,0.4263,0.099,18/100
+2,Prove Chunking,finqa,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,token-level,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.1667,0.1188,0.6431,0.4316,0.1176,23/100
+3,Prove Hybrid/Rerank,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,disabled,token-level,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,long,0.1316,0.0693,0.6763,0.4263,0.099,11/100
+4,Max Raw Context,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,disabled,sentence-level,512,100,412,Hybrid,0.6,50,5,reverse,N/A,N/A,long,0.1947,0.0795,0.7239,0.4316,0.1176,14/100
+5,Golden Setup,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,enabled,token-level,512,100,412,Hybrid,0.6,50,5,reverse,200,20,long_cot,0.4158,0.8363,0.7073,Cannot compute (insufficient class variance),0,4/100
+6,Optimized Financial,finqa,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,disabled,token-level,512,100,412,Hybrid,0.8,50,3,forward,N/A,N/A,long,0.2468,0.1679,0.6177,0.5474,0.1731,6/100

data/General-msmarco.csv

@@ -1,7 +1,7 @@
-Test #,Configuration purpose,Subset(s),Embedding Model,Reranker Model,Summarization Model,Chunking Strategy,Chunk size,Overlap,Stride,Retreival Strategy,Alpha,Retr. K,Final K,Repacking,Summ. Max,Summ. Min,8B GPT Label,RMSE=trace relevance,RMSE=trace utilization,RMSE=trace completeness,AUCROC,F1-score,# Failed/Total Samples
-1,Efficiency Baseline,msmarco,BAAI/bge-base-en-v1.5,cross-encoder/ms-marco-MiniLM-L-6-v2,N/A,hard-cut,256,50,N/A,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3252,0.1998,0.4362,0.5125,0.5954,23/100
-2,Prove Chunking,msmarco,BAAI/bge-base-en-v1.5,cross-encoder/ms-marco-MiniLM-L-6-v2,N/A,sliding-window,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3449,0.1947,0.4248,0.495,0.5625,30/100
-3,Prove Hybrid/Rerank,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,N/A,sliding-window,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3183,0.1793,0.407,0.5183,0.6061,22/100
-4,Prove Repacking,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,N/A,hard-cut,256,50,206,Hybrid,0.8,50,5,reverse,N/A,N/A,long,0.3416,0.1837,0.4491,0.559,0.6763,11/100
-5,Prove Summarization,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,fangyuan/nq_abstractive_compressor,sliding-window,256,50,206,Hybrid,0.8,50,5,reverse,150,20,long_cot,0.5066,0.8781,0.5049,N/A,0,3/100
-6,Optimized Hybrid,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,N/A,hard-cut,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,long,0.3292,0.1754,0.5477,0.4842,0.4706,0/100
+Test #,Configuration purpose,Subset(s),Embedding Model,Reranker Model,Summarization,Chunking Strategy,Chunk size,Overlap,Stride,Retreival Strategy,Alpha,Retr. K,Final K,Repacking,Summ. Max,Summ. Min,8B GPT Label,RMSE=trace relevance,RMSE=trace utilization,RMSE=trace completeness,AUCROC,F1-score,# Failed/Total Samples
+1,Efficiency Baseline,msmarco,BAAI/bge-base-en-v1.5,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,sentence-level,256,50,N/A,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3252,0.1998,0.4362,0.5125,0.5954,23/100
+2,Prove Chunking,msmarco,BAAI/bge-base-en-v1.5,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,token-level,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3449,0.1947,0.4248,0.495,0.5625,30/100
+3,Prove Hybrid/Rerank,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,disabled,token-level,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,short,0.3183,0.1793,0.407,0.5183,0.6061,22/100
+4,Prove Repacking,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,disabled,sentence-level,256,50,206,Hybrid,0.8,50,5,reverse,N/A,N/A,long,0.3416,0.1837,0.4491,0.559,0.6763,11/100
+5,Prove Summarization,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,enabled,token-level,256,50,206,Hybrid,0.8,50,5,reverse,150,20,long_cot,0.5066,0.8781,0.5049,N/A,0,3/100
+6,Optimized Hybrid,msmarco,BAAI/bge-base-en-v1.5,BAAI/bge-reranker-base,disabled,sentence-level,256,50,206,Hybrid,0.8,50,5,forward,N/A,N/A,long,0.3292,0.1754,0.5477,0.4842,0.4706,0/100

data/Legal-cuad.csv

@@ -1,6 +1,6 @@
 Test #,Configuration purpose,Subset(s),Embedding Model,Reranker Model,Summarization Model,Chunking Strategy,Chunk size,Overlap,Stride,Retreival Strategy,Alpha,Retr. K,Final K,Repacking,Summ. Max,Summ. Min,8B GPT Label,RMSE=trace relevance,RMSE=trace utilization,RMSE=trace completeness,AUCROC,F1-score,% Failed Sample
-1,Efficiency Baseline,cuad,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,N/A,hard-cut/ token aware chunking ,512,100,N/A,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.2951,0.1697,0.6225,0.4321,0.3761,35.00%
-2,Prove Chunking,cuad,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,N/A,sliding-window,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.2927,0.1623,0.5612,0.4065,0.2609,32.00%
-3,Prove Hybrid/Rerank,cuad,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,N/A,sliding-window,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,long,0.3087,0.1296,0.5315,0.5197,0.5543,15.00%
-4,Max Raw Context,cuad,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,N/A,hard-cut/ token aware chunking ,512,100,412,Hybrid,0.6,50,5,reverse,N/A,N/A,long,0.3287,0.1429,0.6583,0.4132,0.3859,17.00%
-5,Golden Setup,cuad,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,fangyuan/nq_abstractive_compressor,sliding-window,512,100,412,Hybrid,0.6,50,5,reverse,250,50,long_cot,0.5048,0.7648,0.4832,0.0215,0.5054,17.00%
+1,Efficiency Baseline,cuad,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,sentence-level ,512,100,N/A,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.2951,0.1697,0.6225,0.4321,0.3761,35.00%
+2,Prove Chunking,cuad,BAAI/bge-m3,cross-encoder/ms-marco-MiniLM-L-6-v2,disabled,token-level,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,short,0.2927,0.1623,0.5612,0.4065,0.2609,32.00%
+3,Prove Hybrid/Rerank,cuad,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,disabled,token-level,512,100,412,Hybrid,0.6,50,5,forward,N/A,N/A,long,0.3087,0.1296,0.5315,0.5197,0.5543,15.00%
+4,Max Raw Context,cuad,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,disabled,sentence-level,512,100,412,Hybrid,0.6,50,5,reverse,N/A,N/A,long,0.3287,0.1429,0.6583,0.4132,0.3859,17.00%
+5,Golden Setup,cuad,BAAI/bge-m3,BAAI/bge-reranker-v2-m3,enabled,token-level,512,100,412,Hybrid,0.6,50,5,reverse,250,50,long_cot,0.5048,0.7648,0.4832,0.0215,0.5054,17.00%