enable batch_size auto for model eval (#2675)

vkuzo · web-flow · commit 18edd011863c · 2025-08-05T07:35:56.000-04:00
* Update

[ghstack-poisoned]

* Update

[ghstack-poisoned]

* Update

[ghstack-poisoned]
diff --git a/benchmarks/_models/eval_hf_models.py b/benchmarks/_models/eval_hf_models.py
@@ -147,7 +147,7 @@ def run(
         "--device", type=str, default="cuda:0", help="Device to run the model on."
     )
     parser.add_argument(
-        "--batch_size", type=int, default=1, help="Batch size for lm_eval."
+        "--batch_size", type=str, default="auto", help="Batch size for lm_eval."
     )
     parser.add_argument(
         "--prompt",

Original file line number	Diff line number	Diff line change
`@@ -147,7 +147,7 @@ def run(`
`147`	`147`	`"--device", type=str, default="cuda:0", help="Device to run the model on."`
`148`	`148`	`)`
`149`	`149`	`parser.add_argument(`
`150`		`- "--batch_size", type=int, default=1, help="Batch size for lm_eval."`
	`150`	`+ "--batch_size", type=str, default="auto", help="Batch size for lm_eval."`
`151`	`151`	`)`
`152`	`152`	`parser.add_argument(`
`153`	`153`	`"--prompt",`