Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Pull requests: gittensor-ai-lab/sparkinfer

New pull request New
17 Open 164 Closed
17 Open 164 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

perf(attn): trim int8 mma flash-decode shared round-trips (+3.9% at 16k)
#221 opened Jul 4, 2026 by fansilas Contributor Loading…
1 task done
fix(runtime): handle tied embeddings in convert_qwen35.py (fixes #219)
#220 opened Jul 4, 2026 by minion1227 Loading…
1 task
1
fix(runtime): guarantee forward progress in Scheduler::schedule() (fixes #217)
#218 opened Jul 4, 2026 by minion1227 Loading…
1 task
1
perf(decode): lower adaptive 256-split threshold for long-context flash decode area:kernels subsystem (emission weight 0.42) area:runtime subsystem (emission weight 0.26) needs-benchmark Box ticked but decode before/after not filled with a real improvement — not evaluated
#194 opened Jul 3, 2026 by thomasalvaedison7777-lgtm Loading…
7 of 8 tasks
1
perf(decode): default gate/up to 1-row mmvq2_qwen + rmsnorm load elimination (sm_120) area:kernels subsystem (emission weight 0.42) not-tested Awaiting maintainer approval to run on RTX 5090; not evaluated
#188 opened Jul 3, 2026 by thomasalvaedison7777-lgtm Loading…
5 tasks done
perf(attn): cp.async double-buffered KV staging in GQA flash-decode split area:kernels subsystem (emission weight 0.42) not-tested Awaiting maintainer approval to run on RTX 5090; not evaluated
#187 opened Jul 3, 2026 by real-venus Loading…
1 task
perf(decode): pack2 MMVQ for attention Q4_K Wq and O projections 128-context UI-only: strongest measured context in sparkinfer eval area:kernels subsystem (emission weight 0.42) eval:none sparkinfer auto-eval verdict: none test-on-5090 Maintainer-approved to evaluate on RTX 5090 (greenlight)
#142 opened Jul 3, 2026 by jony376 Loading…
1 task done
1
1
fix(kv-cache): grow sequences by block delta, not append — prevents block leak + device block-table overrun (RTX 5090 verified) area:runtime subsystem (emission weight 0.26) needs-benchmark Box ticked but decode before/after not filled with a real improvement — not evaluated
#135 opened Jul 2, 2026 by milosde111 Loading…
1 task done
2
perf(decode): occupancy-first adaptive n_splits — 64-split mid-context plateau (skip 128) area:runtime subsystem (emission weight 0.26) needs-benchmark Box ticked but decode before/after not filled with a real improvement — not evaluated
#134 opened Jul 2, 2026 by milosde111 Loading…
1 task done
1
perf(attn): 128-bit uint4 KV staging in 16k GQA flash-decode split (+6% decode) area:kernels subsystem (emission weight 0.42) eval:none sparkinfer auto-eval verdict: none re-evaluate Winner merged — rebase onto main; bot re-evaluates on push test-on-5090 Maintainer-approved to evaluate on RTX 5090 (greenlight)
#131 opened Jul 2, 2026 by galuis116 Contributor Loading…
1 task done
4
perf(attn): bf16 GQA smem, uint4 ldg KV loads, tile14 + adaptive combine area:kernels subsystem (emission weight 0.42) not-tested Awaiting maintainer approval to run on RTX 5090; not evaluated
#129 opened Jul 2, 2026 by jimcody1995 Contributor • Draft
3 of 4 tasks
perf(gemv): specialize q4 mmvq common K area:kernels subsystem (emission weight 0.42) area:runtime subsystem (emission weight 0.26) eval:none sparkinfer auto-eval verdict: none re-evaluate Winner merged — rebase onto main; bot re-evaluates on push test-on-5090 Maintainer-approved to evaluate on RTX 5090 (greenlight)
#128 opened Jul 2, 2026 by DragunovX16 Contributor • Draft
1 task done
3
Optimize 256-split long-context flash decode area:kernels subsystem (emission weight 0.42) not-tested Awaiting maintainer approval to run on RTX 5090; not evaluated
#127 opened Jul 2, 2026 by DragunovX16 Contributor • Draft
1 task done
feat(qwen): add Qwen3.6 hybrid decode path
#118 opened Jul 1, 2026 by ai-hpc Member • Draft
fix(runtime): grow-aware KV cache allocate; bound cumulative blocks (fixes #110) area:runtime subsystem (emission weight 0.26) not-tested Awaiting maintainer approval to run on RTX 5090; not evaluated
#111 opened Jun 30, 2026 by ultrahighsuper Loading…
1 task
1
Skip LM head on teacher-forced prompt tokens area:kernels subsystem (emission weight 0.42) area:runtime subsystem (emission weight 0.26) not-tested Awaiting maintainer approval to run on RTX 5090; not evaluated
#107 opened Jun 30, 2026 by claytonlin1110 Loading…
5 tasks
1
runtime: add Gemma 4 26B-A4B end-to-end decode path area:runtime subsystem (emission weight 0.26) not-tested Awaiting maintainer approval to run on RTX 5090; not evaluated
#85 opened Jun 28, 2026 by andriypolanski Loading…
5 tasks done
1
3
ProTip! Mix and match filters to narrow down what you’re looking for.