fix: add global step number in hf moe chpt (#643)

YashasviChaurasia · web-flow · commit facbce801d77 · 2025-11-25T02:48:27.000+05:30
Signed-off-by: yashasvi &lt;yashasvi@ibm.com&gt;
diff --git a/tuning/config/acceleration_configs/fast_moe.py b/tuning/config/acceleration_configs/fast_moe.py
@@ -99,7 +99,7 @@ def on_save(
                 def checkpoint(checkpoint_dir, save_dir, is_intermediate: bool = True):
                     if is_intermediate:
                         hf_converted_output_dir = os.path.join(
-                            save_dir, "hf_converted_checkpoint"
+                            save_dir, f"safetensors-{state.global_step}"
                         )
                     else:
                         hf_converted_output_dir = save_dir
diff --git a/tuning/trainercontroller/callback.py b/tuning/trainercontroller/callback.py
@@ -587,7 +587,7 @@ def on_save(
             kwargs["is_final"] = False
 
         base_path = kwargs["path"]
-        hf_converted_path = os.path.join(base_path, "hf_converted_checkpoint")
+        hf_converted_path = os.path.join(base_path, f"safetensors-{state.global_step}")
 
         if os.path.isdir(hf_converted_path):
             kwargs["hf_path"] = hf_converted_path

Original file line number	Diff line number	Diff line change
`@@ -99,7 +99,7 @@ def on_save(`
`99`	`99`	`def checkpoint(checkpoint_dir, save_dir, is_intermediate: bool = True):`
`100`	`100`	`if is_intermediate:`
`101`	`101`	`hf_converted_output_dir = os.path.join(`
`102`		`- save_dir, "hf_converted_checkpoint"`
	`102`	`+ save_dir, f"safetensors-{state.global_step}"`
`103`	`103`	`)`
`104`	`104`	`else:`
`105`	`105`	`hf_converted_output_dir = save_dir`