td-builder commited on
Commit
896436f
·
verified ·
1 Parent(s): 129299a

Upload 141 files

Browse files
Files changed (1) hide show
  1. hugging/td_fuse/merge.py +11 -7
hugging/td_fuse/merge.py CHANGED
@@ -945,13 +945,17 @@ def run_single_merge(
945
  step_t = time.time()
946
  if residual_bank is not None:
947
  print(f"\n[merge] Saving residuals for {stage_name}...")
948
- residual_bank.save_residuals(
949
- stage_name=stage_name,
950
- pre_merge_target_state=pre_merge_state,
951
- source_state=source_state_cpu, # Already on CPU from step 5.7
952
- post_merge_state={k: v.cpu() for k, v in target_model.state_dict().items()},
953
- source_config=source_config,
954
- )
 
 
 
 
955
 
956
  print(f"[merge] Step 9/10 done in {time.time()-step_t:.0f}s"); sys.stdout.flush()
957
 
 
945
  step_t = time.time()
946
  if residual_bank is not None:
947
  print(f"\n[merge] Saving residuals for {stage_name}...")
948
+ try:
949
+ residual_bank.save_residuals(
950
+ stage_name=stage_name,
951
+ pre_merge_target_state=pre_merge_state,
952
+ source_state=source_state_cpu, # Already on CPU from step 5.7
953
+ post_merge_state={k: v.cpu() for k, v in target_model.state_dict().items()},
954
+ source_config=source_config,
955
+ )
956
+ except Exception as e:
957
+ print(f"[merge] WARNING: Residual save failed ({e}) — continuing without residuals")
958
+ print(f"[merge] This is non-fatal, merge is still valid")
959
 
960
  print(f"[merge] Step 9/10 done in {time.time()-step_t:.0f}s"); sys.stdout.flush()
961