Update README.md

rryisthebest · web-flow · commit 5bfd918dafd2 · 2024-08-30T21:17:05.000-05:00
diff --git a/README.md b/README.md
@@ -18,7 +18,7 @@ Before running, do
 export REPO_DIR=<path to this directory e.g. /shared/nas/data/m1/revanth3/exp/prf/ai2_data/workspace/repo/llm-reranker>
 ```
 
-## Retrieval
+## 1. Retrieval
 Please download the precomputed BEIR encodings stored at (Link will be added shortly)
 Run the baseline Contriever retrieval using the precomputed encodings
 
@@ -31,13 +31,13 @@ To get the baseline contriever scores and preprocess datasets, Run:
 bash bash/beir/run_eval.sh rank
 ```
 
-## Reranking
+## 2a. Reranking
 Cross-encoder rerankig config is at `{REPO_DIR}/bash/beir/run_rerank_CE.sh`
 To run the baseline cross encoder re-ranking, run:
 ```
 bash bash/beir/run_rerank.sh
 ```
-### LLM Reranking
+### 2b. LLM Reranking
 LLM results preparation config is at `{REPO_DIR}/bash/beir/run_convert_results.sh`
 To prepare retrieval results for LLM reranking, run:
 
@@ -60,7 +60,8 @@ bash bash/run_eval.sh rerank
 Set flag --suffix to "llm_FIRST_alpha" for FIRST LLM evaluation or "ce" for cross encoder reranker
 ```
 
-### Train
+
+### 3. Train
 
 We support three training objectives: Ranking, Generation, and Combined. The Ranking objective uses a learning-to-rank algorithm to output the logits for the highest-ranked passage ID. The Generation objective follows the principles of Causal Language Modeling, focusing on permutation generation. The Combined objective, which we introduce in our paper, is a novel weighted approach that seamlessly integrates both ranking and generation principles, and is the setting applied to FIRST model.
 
