• Joined on 2026-02-17
grenade opened issue helexa/helexa#125 2026-07-03 19:41:55 +00:00
TP auto-recovery livelocks after watchdog poison: unload "in flight" + reload "already loaded" every 2s, never converges
grenade commented on issue helexa/helexa#123 2026-07-03 18:34:26 +00:00
TP load-on-demand is not disconnect-safe: client timeout kills rank-1 mid-load and leaks the leader shard

Two more facets of load-on-demand fragility observed while retrying the swap (build 1773df3):

  1. Load against occupied VRAM has no admission check: POST /models/load for Coder-Next while…
grenade pushed to main at helexa/helexa 2026-07-03 17:24:42 +00:00
1773df34ad Merge pull request 'fix(neuron): qwen3_next models get snapshots, prefix cache + batch engine (#98)' (#124) from fix/98-qwen3-next-snapshot-gate into main
687324745f fix(neuron): qwen3_next models get snapshots, prefix cache + batch engine (#98)
Compare 2 commits »
grenade merged pull request helexa/helexa#124 2026-07-03 17:24:40 +00:00
fix(neuron): qwen3_next models get snapshots, prefix cache + batch engine (#98)
grenade created pull request helexa/helexa#124 2026-07-03 17:24:32 +00:00
fix(neuron): qwen3_next models get snapshots, prefix cache + batch engine (#98)
grenade pushed to fix/98-qwen3-next-snapshot-gate at helexa/helexa 2026-07-03 17:09:03 +00:00
687324745f fix(neuron): qwen3_next models get snapshots, prefix cache + batch engine (#98)
grenade created branch fix/98-qwen3-next-snapshot-gate in helexa/helexa 2026-07-03 17:09:02 +00:00
grenade opened issue helexa/helexa#123 2026-07-03 17:03:04 +00:00
TP load-on-demand is not disconnect-safe: client timeout kills rank-1 mid-load and leaks the leader shard
grenade pushed to main at helexa/helexa 2026-07-03 14:22:55 +00:00
66632db578 config(beast): enable batched decode engine — max_in_flight 8 (#98)
grenade pushed to main at helexa/helexa 2026-07-03 14:20:33 +00:00
abac5cb303 Merge pull request 'feat(neuron): continuous batching — TP batched decode + engine backend seam (#98)' (#122) from feat/98-tp-batched-decode into main
daeab6be12 fix(neuron): TP unload with Arc-shared pool mutex (#98)
2a366a1c17 feat(neuron): engine backend seam — TP models batch through the engine (#98)
4c3a2735b0 feat(neuron): TP rank plumbing for batched decode (#98)
Compare 4 commits »
grenade merged pull request helexa/helexa#122 2026-07-03 14:20:32 +00:00
feat(neuron): continuous batching — TP batched decode + engine backend seam (#98)
grenade created pull request helexa/helexa#122 2026-07-03 14:20:21 +00:00
feat(neuron): continuous batching — TP batched decode + engine backend seam (#98)
grenade pushed to feat/98-tp-batched-decode at helexa/helexa 2026-07-03 14:04:48 +00:00
daeab6be12 fix(neuron): TP unload with Arc-shared pool mutex (#98)
grenade pushed to feat/98-tp-batched-decode at helexa/helexa 2026-07-03 13:46:04 +00:00
2a366a1c17 feat(neuron): engine backend seam — TP models batch through the engine (#98)
grenade pushed to feat/98-tp-batched-decode at helexa/helexa 2026-07-03 13:23:55 +00:00
4c3a2735b0 feat(neuron): TP rank plumbing for batched decode (#98)
grenade created branch feat/98-tp-batched-decode in helexa/helexa 2026-07-03 13:23:55 +00:00
grenade commented on issue helexa/helexa#98 2026-07-03 12:39:35 +00:00
F4d — perf(neuron): continuous batching for agentic fan-out

Slices 1–3b merged in #121 (single-GPU path complete, default-off):

  • Design: gang batching with lockstep decode + cat-based rebatch. Prefill stays B=1 (chunked prefill + prefix cache reused…
grenade pushed to main at helexa/helexa 2026-07-03 12:39:09 +00:00
7454ec7744 Merge pull request 'feat(neuron): continuous batching — batched decode + engine (#98, slices 1–3b)' (#121) from feat/98-batched-decode into main
d3ef9513f8 fix(neuron): as_deref for Arc'd prefix_cache in cuda-only worker path (#98)
ec7a5b750a feat(neuron): lockstep batched decode engine (#98)
7f2fcc3527 feat(neuron): batch-row extraction for rebatch on join/leave (#98)
9ac1a1586c feat(neuron): device-worker jobs for batched decode (#98)
Compare 6 commits »
grenade merged pull request helexa/helexa#121 2026-07-03 12:39:08 +00:00
feat(neuron): continuous batching — batched decode + engine (#98, slices 1–3b)
grenade created pull request helexa/helexa#121 2026-07-03 12:39:00 +00:00
feat(neuron): continuous batching — batched decode + engine (#98, slices 1–3b)