Skip to content

Actions: jamesburton/dotLLM

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
2,500+ workflow runs
2,500+ workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
perf(vulkan): packed Q8_0 shared experts for Qwen3.6 hybrid (#987) CI #737: Pull request #989 opened by jamesburton
In progress issue/986-hybrid-shared-packed
issue/986-hybrid-shared-packed
In progress
perf(cuda): mma decode attention is latency-bound at depth (F16 decode 18.8 -> 14.5 tok/s at 16k) - bigger KV tiles, fewer syncs, vectorized loads Gemini Bot #1793: Issue #988 opened by jamesburton
1s
1s
perf(cuda): mma decode attention is latency-bound at depth (F16 decode 18.8 -> 14.5 tok/s at 16k) - bigger KV tiles, fewer syncs, vectorized loads Claude Code #1793: Issue #988 opened by jamesburton
7s
7s
perf(vulkan): keep Qwen3.6/qwen4exp Q8_0 shared-expert gate/up/down packed (no F16 copies) Claude Code #1792: Issue #987 opened by jamesburton
1s
1s
perf(vulkan): keep Qwen3.6/qwen4exp Q8_0 shared-expert gate/up/down packed (no F16 copies) Gemini Bot #1792: Issue #987 opened by jamesburton
1s
1s
perf(engine/cuda): long prefill chunk size 256 re-converts/dequantizes the whole KV history 4x more than needed - tune chunk size (and expose it) Claude Code #1791: Issue comment #963 (comment) created by jamesburton
1s
1s
perf(engine/cuda): long prefill chunk size 256 re-converts/dequantizes the whole KV history 4x more than needed - tune chunk size (and expose it) Gemini Bot #1791: Issue comment #963 (comment) created by jamesburton
1s
1s
perf(vulkan): dp4a grouped MMQ for MoE prefill (Qwen3.6 at 6% of llama.cpp) Gemini Bot #1790: Issue #986 opened by jamesburton
2s
2s
perf(vulkan): dp4a grouped MMQ for MoE prefill (Qwen3.6 at 6% of llama.cpp) Claude Code #1790: Issue #986 opened by jamesburton
2s
2s
perf(cpu): default decode thread cap of 8 leaves 30-50% of CPU decode throughput on the table (dense AND MoE) Claude Code #1789: Issue comment #984 (comment) created by jamesburton
1s
1s
perf(cpu): default decode thread cap of 8 leaves 30-50% of CPU decode throughput on the table (dense AND MoE) Gemini Bot #1789: Issue comment #984 (comment) created by jamesburton
1s
1s
perf(cpu): MoE decode (n=1) routed experts run on k single-thread Parallel.For tasks - two pool dispatches per layer instead Gemini Bot #1788: Issue comment #981 (comment) created by jamesburton
2s
2s
perf(cpu): MoE decode (n=1) routed experts run on k single-thread Parallel.For tasks - two pool dispatches per layer instead Claude Code #1788: Issue comment #981 (comment) created by jamesburton
1s
1s
Merge pull request #985 from jamesburton/issue/984-decode-thread-default CI #736: Commit 801d532 pushed by jamesburton
12m 6s dev
dev
12m 6s
Gemma 4 MTP speculative decoding (gemma4-assistant drafter) Gemini Bot #1787: Issue comment #959 (comment) created by jamesburton
2s
2s
Gemma 4 MTP speculative decoding (gemma4-assistant drafter) Claude Code #1787: Issue comment #959 (comment) created by jamesburton
3s
3s
Merge pull request #982 from jamesburton/issue/981-moe-decode-n1 CI #735: Commit 7e0d7f4 pushed by jamesburton
12m 30s dev
dev
12m 30s
perf(cpu): default decode thread count 8 -> threads-max(2,threads/8) (#984) CI #734: Pull request #985 opened by jamesburton
7m 48s issue/984-decode-thread-default
issue/984-decode-thread-default
7m 48s
perf(cpu): default decode thread cap of 8 leaves 30-50% of CPU decode throughput on the table (dense AND MoE) Gemini Bot #1786: Issue #984 opened by jamesburton
1s
1s
perf(cpu): default decode thread cap of 8 leaves 30-50% of CPU decode throughput on the table (dense AND MoE) Claude Code #1786: Issue #984 opened by jamesburton
7s
7s
feat(vulkan): Gemma 4 MTP speculative decoding Claude Code #1785: Issue comment #983 (comment) created by jamesburton
1s
1s
feat(vulkan): Gemma 4 MTP speculative decoding Gemini Bot #1785: Issue comment #983 (comment) created by jamesburton
7s
7s
feat(vulkan): Gemma 4 MTP speculative decoding CI #733: Pull request #983 opened by jamesburton
6m 10s issue/959-gemma4-mtp
issue/959-gemma4-mtp
6m 10s
perf(cpu): MoE single-token decode - two pool dispatches per layer instead of k single-thread tasks (#981) CI #732: Pull request #982 opened by jamesburton
7m 45s issue/981-moe-decode-n1
issue/981-moe-decode-n1
7m 45s
merge: resolve Architecture.cs conflict (keep both MiMoV2 and Gemma4A… CI #731: Commit b17a80f pushed by jamesburton
7m 42s dev
dev
7m 42s