Merge pull request #1 from argmaxinc/SW-202-remove-duplicates-from-multilingual-benchmarks

main.py

@@ -113,26 +113,15 @@ model_to_multilingual_wer = dict(
     zip(multilingual_df["Model"], multilingual_df["Average WER"])
 )
 
-# Copy over the multilingual WER to matching models
-multilingual_models = {}
-for multilingual_model, multilingual_wer in model_to_multilingual_wer.items():
-    for model in benchmark_df["model"].unique().tolist():
-        if model in model_to_multilingual_wer:
-            continue
-        if model == "openai/whisper-large-v3-v20240930/turbo/632MB":
-            multilingual_models[model] = model_to_multilingual_wer["openai/whisper-large-v3-v20240930"]
-        if model.endswith("/turbo") and model.replace("/turbo", "") in model_to_multilingual_wer:
-            multilingual_models[model] = model_to_multilingual_wer[model.replace("/turbo", "")]
-        elif multilingual_model in model and not model.endswith("en"):
-            multilingual_models[model] = multilingual_wer
-
-# Update the dictionary with turbo models
-model_to_multilingual_wer.update(multilingual_models)
-
 # Add English WER and Multilingual WER to performance_df
 benchmark_df["english_wer"] = benchmark_df["model"].map(model_to_english_wer)
 benchmark_df["multilingual_wer"] = benchmark_df["model"].map(model_to_multilingual_wer)
-benchmark_df.fillna({"multilingual_wer": "English-only model"}, inplace=True)
+benchmark_df.fillna({"multilingual_wer": "N/A"}, inplace=True) # Mark all untested models as N/A
+
+# Mark English-only models
+english_only_mask = benchmark_df["model"].str.contains(r"\.en$|distil-whisper", case=False, na=False)
+benchmark_df.loc[english_only_mask, "multilingual_wer"] = "English-only model"
+
 benchmark_df["multilingual_wer"] = benchmark_df["multilingual_wer"].astype(str)
 
 sorted_performance_df = (

utils.py

@@ -909,7 +909,7 @@ strong, b {
 }
 
 #release-dropdown {
-    width: 14%;
+    width: 17%;
     margin-left: 0px;
     margin-right: auto;
 }