Fix exo bench again again (#1750)

b713889f73c895c35b8ca824085f84f7b864c7d4 · 2026-03-17 13:47:24 +0000 · rltakashige

Mb premature auto merge

Files touched

M bench/exo_bench.py

Diff

commit b713889f73c895c35b8ca824085f84f7b864c7d4
Author: rltakashige <rl.takashige@gmail.com>
Date:   Tue Mar 17 13:47:24 2026 +0000

    Fix exo bench again again (#1750)
    
    Mb premature auto merge
---
 bench/exo_bench.py | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/bench/exo_bench.py b/bench/exo_bench.py
index 13e59f8d..20b0a546 100644
--- a/bench/exo_bench.py
+++ b/bench/exo_bench.py
@@ -496,20 +496,28 @@ def main() -> int:
                                 all_rows.append(row)
 
                             if batch_results:
-                                agg_gen_tps = sum(
+                                valid_gen_tps = [
                                     x["stats"]["generation_tps"]
                                     for x, _ in batch_results
                                     if x["stats"]["generation_tps"] > 0
+                                ]
+                                agg_gen_tps = (
+                                    mean(valid_gen_tps) if valid_gen_tps else 0.0
                                 )
+                                gen_tps = agg_gen_tps / concurrency
                                 logger.info(
                                     f"[concurrent {concurrency}x]  "
                                     f"agg_gen_tps={agg_gen_tps:.2f}  "
+                                    f"gen_tps={gen_tps:.2f}  "
                                     f"errors={batch_errors}"
                                 )
 
                     if runs:
                         prompt_tps = mean(x["stats"]["prompt_tps"] for x in runs)
-                        gen_tps = mean(x["stats"]["generation_tps"] for x in runs)
+                        gen_tps = mean(
+                            x["stats"]["generation_tps"] / x["concurrency"]
+                            for x in runs
+                        )
                         ptok = mean(x["stats"]["prompt_tokens"] for x in runs)
                         gtok = mean(x["stats"]["generation_tokens"] for x in runs)
                         peak = mean(

← 6ee67314 Fix exo bench prefill and decode tps (#1749) · back to Exo · Update mflux to 0.16.9 (#1751) a6519ba0 →