[Speculative Decoding] Add DFlash speculators config parsing (#38300)
Signed-off-by: Zhanqiu Hu <zhu@redhat.com>
This commit is contained in:
@@ -523,7 +523,14 @@ class DFlashQwen3ForCausalLM(Qwen3ForCausalLM):
|
||||
self.logits_processor = LogitsProcessor(
|
||||
self.config.draft_vocab_size, scale=logit_scale
|
||||
)
|
||||
self.draft_id_to_target_id = None
|
||||
target_vocab_size = vllm_config.model_config.get_vocab_size()
|
||||
if self.config.draft_vocab_size != target_vocab_size:
|
||||
self.draft_id_to_target_id = nn.Parameter(
|
||||
torch.zeros(self.config.draft_vocab_size, dtype=torch.long),
|
||||
requires_grad=False,
|
||||
)
|
||||
else:
|
||||
self.draft_id_to_target_id = None
|
||||
|
||||
def embed_input_ids(
|
||||
self,
|
||||
|
||||
Reference in New Issue
Block a user