[Model] Add Inkling MTP=1 support [3/N] (#48869)

Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Giancarlo Delfin <32987265+TheEpicDolphin@users.noreply.github.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Isotr0py <mozf@inferact.ai>
Co-authored-by: Jee Jee Li <jeejeelee@inferact.ai>
Co-authored-by: Roger Wang <hey@rogerw.io>
Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Woosuk Kwon
2026-07-16 13:27:21 -07:00
committed by GitHub
co-authored by Bugen Zhao Giancarlo Delfin Isotr0py Isotr0py Jee Jee Li Roger Wang Yifan Qiao Claude Fable 5 OpenAI Codex
parent 971dac2caa
commit fb5ec0dc9e
11 changed files with 710 additions and 6 deletions
+6
View File
@@ -61,6 +61,8 @@ class InklingModelConfig(PretrainedConfig):
logits_mup_width_multiplier: float | None = None,
final_logit_softcapping: float | None = None,
tie_word_embeddings: bool = False,
num_nextn_predict_layers: int = 0,
chain_hidden_post_norm: bool = False,
**kwargs: Any,
) -> None:
if head_dim is None:
@@ -135,6 +137,10 @@ class InklingModelConfig(PretrainedConfig):
self.unpadded_vocab_size = self.vocab_size
self.logits_mup_width_multiplier = logits_mup_width_multiplier
self.final_logit_softcapping = final_logit_softcapping
# MTP (multi-token prediction) draft head: number of depth layers in the
# checkpoint (0 if absent); chain_norm applied after each depth.
self.num_nextn_predict_layers = num_nextn_predict_layers
self.chain_hidden_post_norm = chain_hidden_post_norm
super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)