MCPcopy Create free account
hub / github.com/OpenNMT/CTranslate2 / set_decoder

Method set_decoder

python/ctranslate2/converters/transformers.py:690–702  ·  view source on GitHub ↗
(self, spec, module)

Source from the content-addressed store, hash-verified

688 config.unk_token = tokenizer.unk_token
689
690 def set_decoder(self, spec, module):
691 spec.scale_embeddings = False
692 self.set_embeddings(spec.embeddings, module.wte)
693 self.set_position_encodings(spec.position_encodings, module.wpe)
694 self.set_layer_norm(spec.layer_norm, module.ln_f)
695
696 for layer_spec, layer in zip(spec.layer, module.h):
697 self.set_layer_norm(layer_spec.self_attention.layer_norm, layer.ln_1)
698 self.set_linear(layer_spec.self_attention.linear[0], layer.attn.c_attn)
699 self.set_linear(layer_spec.self_attention.linear[1], layer.attn.c_proj)
700 self.set_layer_norm(layer_spec.ffn.layer_norm, layer.ln_2)
701 self.set_linear(layer_spec.ffn.linear_0, layer.mlp.c_fc)
702 self.set_linear(layer_spec.ffn.linear_1, layer.mlp.c_proj)
703
704
705@register_loader("GPTJConfig")

Callers 1

get_model_specMethod · 0.95

Calls 4

set_embeddingsMethod · 0.45
set_layer_normMethod · 0.45
set_linearMethod · 0.45

Tested by

no test coverage detected