All issues

Search results

Capturable grammar hangs decode CUDA-graph capture when capture batch x draft tokens exceeds 128 logits rows (xgrammar, TP=8, B300)

bugSomething isn't workingSomething isn't working

high priority

Status: Open.

#1314 In lightseekorg/tokenspeed;

· abatilo opened on Aug 30, 2026

[Feature] Improve Kimi-K3 EAGLE3 acceptance across representative decode states

Status: Open.

#1305 In lightseekorg/tokenspeed;

· raikonenfnu opened on Aug 30, 2026

Input generators are registered under operator names no kernel uses (quantize.* vs quantization.*, moe.align_block_size vs moe.apply)

Status: Open.

#1244 In lightseekorg/tokenspeed;

· vivekkhandelwal1 opened on Aug 25, 2026

get_available_device() masks the real error behind "Triton is not supported on the current platform"

Status: Open.

#1243 In lightseekorg/tokenspeed;

· vivekkhandelwal1 opened on Aug 25, 2026

[gfx950][k3][dspark] Improve long context decoed cost on low acceptance

Status: Open.

#1132 In lightseekorg/tokenspeed;

· raikonenfnu opened on Aug 18, 2026

TokenSpeed Model Support: Top Open-Source Models & Day 0 Collaboration

high priority

Status: Open.

#965 In lightseekorg/tokenspeed;

· lightseek-bot opened on Aug 6, 2026

TokenSpeed Project Ethos and External Contributions

high priority

Status: Open.

#149 In lightseekorg/tokenspeed;

· lightseek-bot opened on May 14, 2026