Search results

the speed of speculative decoding

Status: Open.

#42 In feifeibear/LLMSpeculativeSampling;

· YZH0905 opened on Jun 19, 2025

why autoregressive_sampling and speculative_sampling_v2 output different tokens

Status: Open.

#41 In feifeibear/LLMSpeculativeSampling;

· java63940 opened on May 29, 2025

Errors from Tuples and DynamicCache

Status: Open.

#40 In feifeibear/LLMSpeculativeSampling;

· doquangg opened on Apr 1, 2025

Deepmind version's probability distribution

Status: Open.

#36 In feifeibear/LLMSpeculativeSampling;

· YEOMJINSEOP opened on Nov 2, 2024

Bloom's kvcache are both (bs, head, seq, head_dim) in new version of transformers

Status: Open.

#34 In feifeibear/LLMSpeculativeSampling;

· shadow150519 opened on Oct 6, 2024

output logits not match. question about decoding when draft model and target model is the same.

documentationImprovements or additions to documentationImprovements or additions to documentation

Status: Open.

#27 In feifeibear/LLMSpeculativeSampling;

· 66RING opened on Apr 8, 2024