update

src/llm_perf.py

@@ -110,10 +110,12 @@ def get_llm_perf_df(machine: str = "hf-dgx-01"):
     llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(
         lambda x: x.replace("float16+", "").replace("float32+", "").replace("bfloat16+", "") if "bit" in x else x
     )
-    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("awq-4bit", "awq"))
-    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("gptq-4bit", "gptq"))
-    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("bettertransformer", "sdpa"))
-    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("flash-attention-v2", "FA2"))
+    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("bnb-4bit", "BnB-4bit"))
+    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("bnb-8bit", "BnB-8bit"))
+    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("awq-4bit", "AWQ-4bit"))
+    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("gptq-4bit", "GPTQ-4bit"))
+    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("bettertransformer", "SDPA"))
+    llm_perf_df["experiment_name"] = llm_perf_df["experiment_name"].apply(lambda x: x.replace("flash-attention-v2", "FA-v2"))
     # add arch
     llm_perf_df["Arch"] = llm_perf_df["Arch"].apply(process_arch)
     # filter columns