push copy to disk (#12348)

2026-01-10 07:28:15 -05:00 · 2025-09-29 21:55:05 -07:00
parent 881709cd33
commit 2a0caa09c2
3 changed files with 19 additions and 4 deletions
--- a/examples/gpt2.py
+++ b/examples/gpt2.py
@@ -134,8 +134,7 @@ class GPT2:
    transposed = ('attn.c_attn.weight', 'attn.c_proj.weight', 'mlp.c_fc.weight', 'mlp.c_proj.weight')
    for k in weights:
      if k.endswith(transposed):
-        # TODO: it should not silently break without that .to(None)
-        weights[k] = weights[k].to(None).T
+        weights[k] = weights[k].T
    # lm head and wte are tied
    weights['lm_head.weight'] = weights['wte.weight']