From e8f4ea93b2379162e62014413390d3395539e950 Mon Sep 17 00:00:00 2001
From: Hirokazu Kiyomaru <h.kiyomaru@gmail.com>
Date: Tue, 19 Dec 2023 13:29:10 +0900
Subject: [PATCH] include version in the project name by default

---
 README.md                  | 2 +-
 llm_judge/upload_result.py | 5 ++++-
 2 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index b71d662..384d565 100644
--- a/README.md
+++ b/README.md
@@ -129,7 +129,7 @@ WANDB_ENTITY=<USER-NAME or ORGANIZATION-NAME> WANDB_PROJECT=<PROJECT-NAME> pytho
     [--model-list <LIST-OF-MODEL-IDS>]
 ```
 
-By default, the entity is configured to use your username, and the project name is set to `ja-vicuna-qa-benchmark-dev`.
+By default, the entity is configured to use your username, and the project name is set to `ja-vicuna-qa-benchmark-dev-<VERSION>`.
 
 ## Sample Outputs
 
diff --git a/llm_judge/upload_result.py b/llm_judge/upload_result.py
index bdbf096..c574d21 100644
--- a/llm_judge/upload_result.py
+++ b/llm_judge/upload_result.py
@@ -1,4 +1,5 @@
 import argparse
+import importlib.metadata
 import logging
 import os
 
@@ -17,6 +18,8 @@
 
 logger = logging.getLogger(__name__)
 
+VERSION = importlib.metadata.version("ja-vicuna-qa-benchmark")
+
 
 def get_run_config_from_result(mode: str, result: dict) -> dict:
     if mode == "single":
@@ -59,7 +62,7 @@ def upload_results(
         results: A list of results.
         baseline_model: Baseline model name. Only used in `pairwise-baseline` mode.
     """
-    project = os.getenv("WANDB_PROJECT", "ja-vicuna-qa-benchmark-dev")
+    project = os.getenv("WANDB_PROJECT", f"ja-vicuna-qa-benchmark-dev-{VERSION}")
     if len(results) == 0:
         logger.warning(f"No results found for {result_id}")
         return