Add final layer norm to OPT
This commit is contained in:
parent
a10446f258
commit
a7e3ef71aa
|
@ -12,6 +12,8 @@
|
|||
"decoder.embed_tokens.weight": {"mtj": {"module": "embedding_shard/~/linear", "param": "w", "transforms": ["no_transpose", "vocab_pad"]}},
|
||||
"decoder.project_in.weight": {"mtj": {"module": "embedding_shard", "param": "project_in"}},
|
||||
"decoder.embed_positions.weight": {"mtj": {"module": "embedding_shard", "param": "pos_embs", "transforms": ["no_transpose", "remove_first_two_rows"]}},
|
||||
"decoder.final_layer_norm.weight": {"mtj": {"module": "projection_shard/~/replicated_layer_norm", "param": "scale"}},
|
||||
"decoder.final_layer_norm.bias": {"mtj": {"module": "projection_shard/~/replicated_layer_norm", "param": "offset"}},
|
||||
"decoder.project_out.weight": {"mtj": {"module": "projection_shard", "param": "project_out"}}
|
||||
},
|
||||
"layer_weights": {
|
||||
|
|
Loading…
Reference in New Issue