Repository navigation

All pull requests

[https://nvbugs/6658015][fix] Pin uvloop<0.23.0 to restore trtllm-serve streaming throughput
#19992chenfeiz0326 opened Oct 8, 2026Collaborator1 of 2 tasks
1
[None][feat] Kimi K3 on modeling_v2: MXFP4 target, MNNVL and collective decode kernels, DSpark drafter
#19991vsabavat opened Oct 8, 2026Collaborator1 task done
[None][feat] modeling_v2 catalog: Kimi K3 decode, KDA / MLA and generic-path kernels as entries (sm_100)
#19990vsabavat opened Oct 8, 2026Collaborator1 task done
[None][perf] One-model speculative decoding: host work between steps, worker-produced target logits, KDA drafts' records
#19989vsabavat opened Oct 8, 2026Collaborator1 task done
[None][fix] Kimi K3 / KDA / MNNVL / DFlash: fixes for defects on main
#19988vsabavat opened Oct 8, 2026Collaborator1 task done
[https://nvbugs/16812332][fix] Reshape the KDA decode fallback core to match prefill
#19987yifeizhang-c opened Oct 8, 2026Collaborator1 task done
3
[TRTLLM-16619][doc] Remove stale execution backend guidance
#19986Funatiq opened Oct 8, 2026Collaborator1 task
5
[https://nvbugs/6884443][fix] Release failed-test memory pinned by pytest-rerunfailures in unittests
#19985xxi-nv opened Oct 8, 2026Collaborator1 task done
6
[https://nvbugs/6782263][fix] Cherry-pick #19553 to release/1.3: Serialize HF remote-code loading when building tokenizers
#19983tianyuxbear opened Oct 8, 2026Collaborator1 task done
8
[None][fix] Reclaim idle eager CUTLASS MoE workspace
#19981zhaoyangwang-nvidia opened Oct 8, 2026Collaborator1 task done
[None][chore] Sync Claude Code agents and skills from trtllm-agent-toolkit
#19980kaiyux opened Oct 8, 2026Member1 task done
12
[https://nvbugs/6818106][fix] Cherry-pick fix for disaggregated logprobs computation #19928 to release/1.3
#19978yifeizhang-c opened Oct 8, 2026Collaborator1 task done
4
[TRTLLM-16304][feat] Upgrade the infrastructure of modeling V2
#19977WeiHaocheng opened Oct 8, 2026Collaborator
17
[None][perf] Port the DKG paged MQA logits kernel improvements and default the DSA indexer to CuTe DSL
#19976dc3671 opened Oct 8, 2026Collaborator4 of 5 tasks
[None][perf] MiniMax-M3: split the MSA indexer's prefill proxy along KV
#19975zheyuf opened Oct 8, 2026Collaborator1 task done
[None][perf] MiniMax-M3: remove NVFP4 KV cache host overhead
#19974zheyuf opened Oct 8, 2026Collaborator1 task done
[None][fix] Sweep orphaned GPU worker processes after pytest-timeout
#19973HuiGao-NV opened Oct 8, 2026Collaborator1 task done
10
[https://nvbugs/6859731][test] Make speculative D2H sync checks deterministic
#19972zhaoyangwang-nvidia opened Oct 8, 2026Collaborator1 task done
5
[https://nvbugs/6810945][fix] Backport Kimi FP8 test fixes to release/1.3
#19971jiaganc opened Oct 8, 2026Collaborator1 task done
10
[None][feat] Add external benchmark tools to Agent Flow
#19970GuanhuaWang2001 opened Oct 8, 2026Collaborator1 task done
4
[TRTLLM-16628][refactor] Remove unused LlmRequest fields
#19967Funatiq opened Oct 8, 2026Collaborator1 task
3
[None][perf] Default TRTLLM_KV_TRANSFER_NUM_THREADS to 4 and fix sender worker races
#19966chuangz0 opened Oct 8, 2026Collaborator1 task done
10
[https://nvbugs/6843918][fix] Cherry-pick #19700, #19697, #19692 to release/1.3: perf-sanity gen_only metric/env fixes, orphaned perf worker cleanup, DSv4 FP8 startup timeout
#19965chenfeiz0326 opened Oct 8, 2026Collaborator2 tasks done
7
[None][feat] merge two renders into a standalone serve/render for dynamo
#19964hchings opened Oct 8, 2026Collaborator1 task
[https://nvbugs/6812424][fix] Enable CUTLASS MXFP8 MoE on SM107 for release/1.3
#19962peaceh-nv opened Oct 8, 2026Collaborator
5

Pull requests