Skip to content

Pull requests: lightseekorg/tokenspeed

New pull request New
36 Open 240 Closed
36 Open 240 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

fix(deps): pin flashinfer-jit-cache==0.6.11 in cuda requirements
#284 opened May 27, 2026 by qywu Collaborator • Draft
7
Update tokenspeed-smg* pins to 1.4.1.post20260527 / 0.4.8.post20260527 / 0.5.3.post20260527
#283 opened May 27, 2026 by lightseek-bot Contributor Loading…
perf: fuse kv write for decode only
#281 opened May 27, 2026 by borontion Contributor Loading…
Replace sampling FlashInfer backend with TokenSpeed Triton
#280 opened May 27, 2026 by FlamingoPg Contributor Loading…
14
feat(memory-saver): optional CPU staging for round-trip weight preservation
#275 opened May 27, 2026 by qywu Collaborator • Draft
3 of 5 tasks
feat(memory-saver): wrap CUDA graphs and attention workspaces in saver.region()
#274 opened May 27, 2026 by qywu Collaborator • Draft
3 of 5 tasks
feat: expose POST /release_memory_occupation and /resume_memory_occupation
#272 opened May 27, 2026 by qywu Collaborator Loading…
6 tasks done
8
feat(engine): add pause_generation / continue_generation API
#270 opened May 26, 2026 by qywu Collaborator Loading…
5 tasks
17
feat(scheduler): per-adapter KV prefix-cache namespace + max_loras batch cap
#268 opened May 26, 2026 by qywu Collaborator Loading…
5 tasks
38
fix(vlm): clamp gpu_memory_utilization conditionally and add --language-model-only
#266 opened May 26, 2026 by chenht2022 Contributor Loading…
4 tasks
6
fix(vlm): substitute mm pad ids in drafter before draft embed_tokens
#265 opened May 26, 2026 by chenht2022 Contributor Loading…
2 tasks
6
[WIP] ci(perf): add 1m perf bench for qwen3.5
#264 opened May 26, 2026 by minedec Contributor Loading…
4
feat(spec-decode): add native DFlash support
#263 opened May 26, 2026 by mesaleh Loading…
23
perf(sampling): fuse logits fp32 cast to argmax or softmax
#262 opened May 26, 2026 by syuoni Member Loading…
4
fix(deepseek): guard missing quant weight_block_size
#261 opened May 26, 2026 by mesaleh Loading…
fix(trtllm-mla): make spec-decode CUDA graph capture causal
#260 opened May 26, 2026 by mesaleh Loading…
6
ci(perf): add qwen3.5 agentic perf ci bs16 case
#257 opened May 26, 2026 by minedec Contributor Loading…
chore(logging): suppress noisy third-party server-launch warnings
#256 opened May 26, 2026 by zhyncs Member Loading…
8
perf: TokenSpeed MLA decode kernel optimization for num_heads=16
#255 opened May 26, 2026 by dishengbin Contributor Loading…
3
ci(eval): add Kimi-K2.5-NVFP4 ocr_bench task
#253 opened May 26, 2026 by zhyncs Member • Draft
feat(eplb): eplb support high priority
#251 opened May 26, 2026 by XucSh Contributor Loading…
11
refactor(scheduler): unify hybrid cache model boundaries.
#243 opened May 25, 2026 by SimonCqk Contributor Loading…
@wangbo981016
13
perf(deepseek-v4): vectorize read_deepseek_v4_indexer_fp8_cache
#238 opened May 24, 2026 by yuanqingz Loading…
9
Support DP sampling for spec decode
#232 opened May 24, 2026 by yubofredwang Contributor Loading…
31
Use operand format signatures for kernel selection
#230 opened May 23, 2026 by antiagainst Member Loading…
12
ProTip! Mix and match filters to narrow down what you’re looking for.