https://findings.yatabis.dev/ https://findings.yatabis.dev/archive/ https://findings.yatabis.dev/archive/2026-11-09/ https://findings.yatabis.dev/archive/2026-11-02/ https://findings.yatabis.dev/archive/2026-10-26/ https://findings.yatabis.dev/archive/2026-10-19/ https://findings.yatabis.dev/archive/2026-10-05/ https://findings.yatabis.dev/archive/2026-09-28/ https://findings.yatabis.dev/archive/2026-09-21/ https://findings.yatabis.dev/archive/2026-09-14/ https://findings.yatabis.dev/archive/2026-09-07/ https://findings.yatabis.dev/findings/agentic-self-modification-open-weights/ https://findings.yatabis.dev/findings/anyjev-calibrated-logit-readout/ https://findings.yatabis.dev/findings/apple-lensvlm-9b-selective-context-expansion/ https://findings.yatabis.dev/findings/astra-single-forward-pass-reasoning/ https://findings.yatabis.dev/findings/astronomical-c695688f43f9-untied-head-shard-mapping/ https://findings.yatabis.dev/findings/ax-engine-flashnext-prefill-mtp-exactness/ https://findings.yatabis.dev/findings/basalt-0d32bb19447b-quest-sign-split-matmul/ https://findings.yatabis.dev/findings/basert-026-qwen3x-garbage-output-fix/ https://findings.yatabis.dev/findings/batched-mtp-private-row-state-isolation/ https://findings.yatabis.dev/findings/clm-8b-contrastive-system-one/ https://findings.yatabis.dev/findings/copy-draft-sampled-tile-aware/ https://findings.yatabis.dev/findings/deepseek-v4-1-flash-new-architecture/ https://findings.yatabis.dev/findings/deepseek-v4-1-reap-2bit-apple-negative-qualification/ https://findings.yatabis.dev/findings/deepseek-v4-load-metadata-cast-transient/ https://findings.yatabis.dev/findings/deepseek-v41-architecture-fast-path/ https://findings.yatabis.dev/findings/deepseek-v41-ced-prefill-bounded-replay/ https://findings.yatabis.dev/findings/deepseek-v41-engram-ssd-tiering/ https://findings.yatabis.dev/findings/deepseek-v41-omlx-fast-path-27tps/ https://findings.yatabis.dev/findings/difflut-net-learnable-fpga-lut-connectivity/ https://findings.yatabis.dev/findings/dooms-9b63edf6ba7e-quant-decoder-split-attention/ https://findings.yatabis.dev/findings/dotllm-318c5d1c6615-prefill-last-logit-contract/ https://findings.yatabis.dev/findings/draw-things-local-code-public-beta/ https://findings.yatabis.dev/findings/dynamic-logic-kernel-width-scaling/ https://findings.yatabis.dev/findings/ei-ddlgn-tfhe-encrypted-inference/ https://findings.yatabis.dev/findings/gliner25-decide-schema-decision-encoder/ https://findings.yatabis.dev/findings/glm53-owned-mtp-transaction/ https://findings.yatabis.dev/findings/irodori-coreml-v0-2-int8/ https://findings.yatabis.dev/findings/irodori-meanflow-mlx-audio-0-5-5/ https://findings.yatabis.dev/findings/irodori-onnx-no-reference-voice/ https://findings.yatabis.dev/findings/irodori-onnx-packaged-server-models/ https://findings.yatabis.dev/findings/irodori-onnx-rust-meanflow-runtime/ https://findings.yatabis.dev/findings/irodori-server-meanflow-defaults/ https://findings.yatabis.dev/findings/irodori-speech-cpp-layout2-voice-design-inversion/ https://findings.yatabis.dev/findings/irodori-speech-cpp-v0-5-openai-api/ https://findings.yatabis.dev/findings/irodori-speech-cpp-v0-7-single-gguf-api3/ https://findings.yatabis.dev/findings/irodori-v4-1-anime-audiocpp-q8/ https://findings.yatabis.dev/findings/irodori-v4-1-small-mf-meanflow/ https://findings.yatabis.dev/findings/irodori-v4-1-small-mf-mlx-merged/ https://findings.yatabis.dev/findings/irodori-v4-1-small-mf-mlx-port/ https://findings.yatabis.dev/findings/irodori-v4-large-audiocpp-wip/ https://findings.yatabis.dev/findings/irodori-v4-large-forthcoming/ https://findings.yatabis.dev/findings/irodori-v4-large-mlx-conversions/ https://findings.yatabis.dev/findings/irodori-v4-large-released/ https://findings.yatabis.dev/findings/irodori-voicedesign-speaker-embedding-practice/ https://findings.yatabis.dev/findings/kapitsa-9fc7ff3cc6e5-metal31-split-merges/ https://findings.yatabis.dev/findings/lithos-metal-launch-20261008-layer-mixer-megakernels/ https://findings.yatabis.dev/findings/livemem-mlx-0e38b4c0bda3-bf16-gdn2-bounded-kv/ https://findings.yatabis.dev/findings/llama-cpp-0138937e3f98-shared-kv-attention-placement/ https://findings.yatabis.dev/findings/llama-cpp-1b6935b227b9-unified-kv-row-lists/ https://findings.yatabis.dev/findings/llama-cpp-276378515cee-metal-top-k-mask/ https://findings.yatabis.dev/findings/llama-cpp-44f91a02b2df-command-buffer-fusion-boundary/ https://findings.yatabis.dev/findings/llama-cpp-494d0f47bde6-src1-precision-gating/ https://findings.yatabis.dev/findings/llama-cpp-4b46d965385c-tri-select-nonfinite/ https://findings.yatabis.dev/findings/llama-cpp-57d04820dfef-metal-topk-deterministic-ties/ https://findings.yatabis.dev/findings/llama-cpp-6d194531e1e5-metal-mxfp4-dispatch/ https://findings.yatabis.dev/findings/llama-cpp-80d2c0c2c56e-group-norm-batch-strides/ https://findings.yatabis.dev/findings/llama-cpp-9b8446b91434-metal-matmul-offset-width/ https://findings.yatabis.dev/findings/llama-cpp-ab3768cdc246-recurrent-rollback-bounds/ https://findings.yatabis.dev/findings/llama-cpp-e01c9f1875e8-fa-output-init-bounds/ https://findings.yatabis.dev/findings/llama-cpp-ea40a3a1e177-biased-moe-routing-fusion/ https://findings.yatabis.dev/findings/llama-cpp-metal-moe-gdn-fusions-28948/ https://findings.yatabis.dev/findings/llama-metal-fusion-pack-concurrency-regression/ https://findings.yatabis.dev/findings/llama-metal-moe-f16-range-scaling/ https://findings.yatabis.dev/findings/llama-metal-moe-ssm-fusion-patterns/ https://findings.yatabis.dev/findings/llama-qsa-incremental-pooled-key-cache/ https://findings.yatabis.dev/findings/llama-recurrent-rollback-snapshot-provenance/ https://findings.yatabis.dev/findings/llama-speculative-batch-row-identity/ https://findings.yatabis.dev/findings/llamacpp-3f03b17724ad-remove-metal-global-amax/ https://findings.yatabis.dev/findings/llamacpp-5c3114268229-qsa-selection-without-dense-mask/ https://findings.yatabis.dev/findings/llamacpp-5c965fd8ca0b-four-dimensional-quantization/ https://findings.yatabis.dev/findings/llamacpp-6e66cb5d4e9a-metal-bf16-tensor-attention/ https://findings.yatabis.dev/findings/llamacpp-74bca3950054-deepseek41-compressed-kv-sharing/ https://findings.yatabis.dev/findings/llamacpp-8baabd53b77d-expert-token-f16-rescale/ https://findings.yatabis.dev/findings/llamacpp-df11702136d0-metal-argmax-pool-reduction/ https://findings.yatabis.dev/findings/llamacpp-eb182d85c0e3-draft-token-map-gpu-placement/ https://findings.yatabis.dev/findings/llamacpp-fa3645feb5f4-noalloc-recurrent-compressor-state/ https://findings.yatabis.dev/findings/llamacpp-glm53-lightning-indexer-metal-cpu-fallback/ https://findings.yatabis.dev/findings/llamacpp-m5-tensor-api-flash-attention-f16-kv/ https://findings.yatabis.dev/findings/llamacpp-m5-wide-query-flash-attention-tuning/ https://findings.yatabis.dev/findings/llamacpp-metal-fa-library-parallel-compile/ https://findings.yatabis.dev/findings/llamacpp-metal-sparse-fa-index-threadgroup-cache/ https://findings.yatabis.dev/findings/llamacpp-v050-qwen4exp-metal-runtime/ https://findings.yatabis.dev/findings/llamacpp-v060-metal-few-row-mma-dflash/ https://findings.yatabis.dev/findings/long-context-preserve-kv-capacity-and-bound-attention/ https://findings.yatabis.dev/findings/loopspec-pipelined-self-speculative-decoding/ https://findings.yatabis.dev/findings/lumen-rs-5a41406634dc-group-four-code-probe/ https://findings.yatabis.dev/findings/m5-int8-activation-q4q5-prefill/ https://findings.yatabis.dev/findings/m5-metal-wide-f16-fwht/ https://findings.yatabis.dev/findings/memory-guard-separate-physical-cap-from-dynamic-ceiling/ https://findings.yatabis.dev/findings/mercury-2-5-diffusion-llm-production-maturity/ https://findings.yatabis.dev/findings/metal-tensor-slice-2gib-wrap/ https://findings.yatabis.dev/findings/metal-threadgroup-resource-envelope/ https://findings.yatabis.dev/findings/mimo-v2-6-large-scale-agentic-rl/ https://findings.yatabis.dev/findings/mlx-007478f1a8b6-affine-wide-eight-vectors/ https://findings.yatabis.dev/findings/mlx-06122f8bc91f-afc03c51105a-large-view-pad-offsets/ https://findings.yatabis.dev/findings/mlx-065ec4683cce-mid-row-split-kv-attention/ https://findings.yatabis.dev/findings/mlx-303763f1a8f0-propagate-caller-stream/ https://findings.yatabis.dev/findings/mlx-32a183e1211b-negative-affine-groups/ https://findings.yatabis.dev/findings/mlx-339279525ea3-quantized-shard-boundaries/ https://findings.yatabis.dev/findings/mlx-360078a8d695-native-sliding-window-sdpa/ https://findings.yatabis.dev/findings/mlx-36969c094354-metal-fence-deadlock/ https://findings.yatabis.dev/findings/mlx-3d2fbcdf11de-rope-custom-frequencies/ https://findings.yatabis.dev/findings/mlx-48143570bd34-qqmm-shared-dispatch/ https://findings.yatabis.dev/findings/mlx-49366677f1e2-addmm-epilogue-precision/ https://findings.yatabis.dev/findings/mlx-4ca8d5d5a78f-generated-kernel-source-cache/ https://findings.yatabis.dev/findings/mlx-52a6c234ab35-fp4-half-tile-loader/ https://findings.yatabis.dev/findings/mlx-700520ae71ca-metal-expm1-reduction/ https://findings.yatabis.dev/findings/mlx-914732419f25-fp-staged-affine-qmv-nax/ https://findings.yatabis.dev/findings/mlx-93e65c0b13ea-layernorm-tail-fill/ https://findings.yatabis.dev/findings/mlx-946077b5f490-gdn-f32-gates/ https://findings.yatabis.dev/findings/mlx-9d5b21d310ea-sliceupdate-large-output/ https://findings.yatabis.dev/findings/mlx-9e2e8e233187-gdn-channel-gate-index/ https://findings.yatabis.dev/findings/mlx-affine-qmv-fast-rows-tail/ https://findings.yatabis.dev/findings/mlx-b996a43fa7c0-metal-grid-sliced-dispatch/ https://findings.yatabis.dev/findings/mlx-ba8c005c1b68-affine-few-row-simdgroup/ https://findings.yatabis.dev/findings/mlx-bb3e2d37660f-metal-extension-build/ https://findings.yatabis.dev/findings/mlx-bff10220c742-staged-affine-nax-splitk/ https://findings.yatabis.dev/findings/mlx-c5afbf3927a8-strided-scan-bounds/ https://findings.yatabis.dev/findings/mlx-c6abf0dbb4fc-few-row-fp-quant-matrix/ https://findings.yatabis.dev/findings/mlx-core-gated-delta-metal-primitive/ https://findings.yatabis.dev/findings/mlx-custom-kernel-cache-source-hash/ https://findings.yatabis.dev/findings/mlx-custom-kernel-dispatch-full-coverage/ https://findings.yatabis.dev/findings/mlx-d512-nax-head-split/ https://findings.yatabis.dev/findings/mlx-fence-dependency-specific-fast-synch/ https://findings.yatabis.dev/findings/mlx-gated-delta-simd-row-packing/ https://findings.yatabis.dev/findings/mlx-gather-qmm-expert-aware-tile-scheduling/ https://findings.yatabis.dev/findings/mlx-gather-qmv-medium-alignment-path/ https://findings.yatabis.dev/findings/mlx-global-scale-nvfp4-matrix-kernels/ https://findings.yatabis.dev/findings/mlx-kquant-nax-q6k-verify-kernels/ https://findings.yatabis.dev/findings/mlx-large-view-span-indexing-correctness/ https://findings.yatabis.dev/findings/mlx-lm-1f7057ca530f-deepseek32-sparse-prefill/ https://findings.yatabis.dev/findings/mlx-lm-33fdd163aac1-convert-nvfp4-global-scale/ https://findings.yatabis.dev/findings/mlx-lm-3696eb276d9f-nvfp4-expert-global-scale/ https://findings.yatabis.dev/findings/mlx-lm-4cfc6478473b-flint-router-cache-int4-layout/ https://findings.yatabis.dev/findings/mlx-lm-4e9a1de64e2d-dsa-cross-layer-indexer-sharing/ https://findings.yatabis.dev/findings/mlx-lm-56230fa02b8f-fused-sparse-mla-prefill/ https://findings.yatabis.dev/findings/mlx-lm-59995f131cf2-minimax-m31-nvfp4-router/ https://findings.yatabis.dev/findings/mlx-lm-72be8c955d9a-afm7-left-padding-rope/ https://findings.yatabis.dev/findings/mlx-lm-9d8abd94d63a-granite-padded-ssm-lengths/ https://findings.yatabis.dev/findings/mlx-lm-a28cdd0dc266-ernie-correction-bias-routing/ https://findings.yatabis.dev/findings/mlx-lm-context-sharding-local-cluster/ https://findings.yatabis.dev/findings/mlx-lm-dc2012bae73a-glm-blocked-sparse-prefill/ https://findings.yatabis.dev/findings/mlx-lm-f39c4aa2af5d-speculative-processor-history/ https://findings.yatabis.dev/findings/mlx-lm-fused-q4-batched-decode-weight-reuse/ https://findings.yatabis.dev/findings/mlx-lm-logprob-sync-avoid/ https://findings.yatabis.dev/findings/mlx-lm-mimo-mixed-head-dim-flash-attention/ https://findings.yatabis.dev/findings/mlx-lm-quantized-kv-attention-query-row-tiling/ https://findings.yatabis.dev/findings/mlx-m5-d256-attention-phase-routing/ https://findings.yatabis.dev/findings/mlx-m5ultra-expert-aware-gather-mm-tiling/ https://findings.yatabis.dev/findings/mlx-metal-command-buffer-launch-heavy-decode/ https://findings.yatabis.dev/findings/mlx-precompiled-metal-kernel-offline-ofast/ https://findings.yatabis.dev/findings/mlx-qmm-staged-8bit-matrix-lanes/ https://findings.yatabis.dev/findings/mlx-qwen4-native-layout-qsa-long-context/ https://findings.yatabis.dev/findings/mlx-sdpa-vector-four-key-unroll/ https://findings.yatabis.dev/findings/mlx-serve-m5-moe-rowmap-swiglu-gather/ https://findings.yatabis.dev/findings/mlx-serve-m5-nax-prefill-generation-aware-routing/ https://findings.yatabis.dev/findings/mlx-serve-m5-qsa-nax-prefill-occupancy/ https://findings.yatabis.dev/findings/mlx-serve-moe-padding-expert-streaming-threshold/ https://findings.yatabis.dev/findings/mlx-serve-prefill-decode-time-share/ https://findings.yatabis.dev/findings/mlx-serve-prefill-fused-expert-reduction/ https://findings.yatabis.dev/findings/mlx-serve-qsa-verify-fused-m5ultra/ https://findings.yatabis.dev/findings/mlx-serve-sparse-batch-pad-waste-floor/ https://findings.yatabis.dev/findings/mlx-vlm-exact-multitoken-verify-6bit/ https://findings.yatabis.dev/findings/mlx2-3ec53f2e16c5-artifact-int8-prefill-qualification/ https://findings.yatabis.dev/findings/mlxcel-0f3a2cc140ee-gdn-l2norm-epsilon/ https://findings.yatabis.dev/findings/mlxcel-32c489866af8-native-batched-projection-ragged-kv/ https://findings.yatabis.dev/findings/mlxcel-56bd9cd5430d-lora-fusion-base-dtype/ https://findings.yatabis.dev/findings/mlxcel-6759cea0ce8a-recurrent-sequence-owned-batching/ https://findings.yatabis.dev/findings/mlxcel-6a1159166cfa-qwen-next-cache-continuation/ https://findings.yatabis.dev/findings/mlxcel-9bb2ff8f3c76-native-window-prefill-route/ https://findings.yatabis.dev/findings/mlxcel-dd9951eb1bc7-qwen4-exp-state-contract/ https://findings.yatabis.dev/findings/mlxcel-e1e116272f8a-gemma3n-native-batched-decode/ https://findings.yatabis.dev/findings/mlxcel-f4f255a9e615-metal-routed-gate-up/ https://findings.yatabis.dev/findings/mlxcel-looped-model-owned-sequence-state/ https://findings.yatabis.dev/findings/moe-lab-1a67c544bf1f-geglu-shared-expert-contract/ https://findings.yatabis.dev/findings/moe-offload-expert-major-prefill/ https://findings.yatabis.dev/findings/molten-611e704fe9a3-gpu-resident-fused-decode/ https://findings.yatabis.dev/findings/mtp-upstream-rollback-acceptance-regression/ https://findings.yatabis.dev/findings/nuclis-0d3f9b78351b-speech-prefill-time-budget/ https://findings.yatabis.dev/findings/nvidia-pair-local-inference-routing/ https://findings.yatabis.dev/findings/occamy-1-0-qwen36-agent-posttrain/ https://findings.yatabis.dev/findings/omlx-06bd1973b55a-turboquant-attention-sinks/ https://findings.yatabis.dev/findings/omlx-606beb6861d6-lightning-mtp-decode-exact/ https://findings.yatabis.dev/findings/omlx-affine-kv-200k-qwen38/ https://findings.yatabis.dev/findings/omlx-affine8-kv-cache/ https://findings.yatabis.dev/findings/omlx-agent-kv-cache-artifact-export-import/ https://findings.yatabis.dev/findings/omlx-ane-dflash-prefill-rollback-safeguards/ https://findings.yatabis.dev/findings/omlx-batched-qsa-row-banks-sparse-gather/ https://findings.yatabis.dev/findings/omlx-c2aef5f30722-mixed-affine-mxfp4-f16/ https://findings.yatabis.dev/findings/omlx-checkpoint-backed-moe-expert-offload/ https://findings.yatabis.dev/findings/omlx-cluster-v2-distributed-mac-inference/ https://findings.yatabis.dev/findings/omlx-d2ef2e8b8dcc-paced-kv-bank-growth/ https://findings.yatabis.dev/findings/omlx-deterministic-sdpa256-route/ https://findings.yatabis.dev/findings/omlx-dynamic-ssd-expert-streaming/ https://findings.yatabis.dev/findings/omlx-flash-next-prefill-fused-lane-m5-max/ https://findings.yatabis.dev/findings/omlx-gdn-prefill-software-pipeline-m5-ultra/ https://findings.yatabis.dev/findings/omlx-glm5-promoted-boundary-metadata/ https://findings.yatabis.dev/findings/omlx-hybrid-draft-cache-position-and-recurrent-snapshots/ https://findings.yatabis.dev/findings/omlx-long-context-attention-key-range-passes/ https://findings.yatabis.dev/findings/omlx-long-context-kv-corruption-cache-poisoning/ https://findings.yatabis.dev/findings/omlx-m5-pro-ane-prefill-split/ https://findings.yatabis.dev/findings/omlx-m5ultra-oversized-nax-gather-qmm/ https://findings.yatabis.dev/findings/omlx-mimo-expert-streaming-phase-aware-hotset/ https://findings.yatabis.dev/findings/omlx-mimo-mtp-one-pass-kv-verify-attention/ https://findings.yatabis.dev/findings/omlx-mimo-multihead-mtp-sliding-window-rollback/ https://findings.yatabis.dev/findings/omlx-mimo-prefill-block-router-occupancy/ https://findings.yatabis.dev/findings/omlx-model-scoped-batched-prefill-control-plane/ https://findings.yatabis.dev/findings/omlx-moe-prefill-layer-backpressure-reclaim/ https://findings.yatabis.dev/findings/omlx-mrope-specprefill-absolute-position/ https://findings.yatabis.dev/findings/omlx-mtp-late-join-reconcile-budget/ https://findings.yatabis.dev/findings/omlx-mtp-parked-head-priming/ https://findings.yatabis.dev/findings/omlx-multi-engine-metal-submission-gpu-hang/ https://findings.yatabis.dev/findings/omlx-native-moe-slot-offload/ https://findings.yatabis.dev/findings/omlx-paged-ssd-restore-bounded-arrays-state/ https://findings.yatabis.dev/findings/omlx-prefill-chunk-independent-cache-block/ https://findings.yatabis.dev/findings/omlx-prefill-pool-threshold-reclaim/ https://findings.yatabis.dev/findings/omlx-prefix-tail-terminal-block/ https://findings.yatabis.dev/findings/omlx-qwen-vlm-fused-decode-gate-regression/ https://findings.yatabis.dev/findings/omlx-qwen36-fp16-top8-fused-moe-decode/ https://findings.yatabis.dev/findings/omlx-qwen38-coarse-head-exact-rescore-speculation/ https://findings.yatabis.dev/findings/omlx-qwen38-ple-load-shard-lifetime/ https://findings.yatabis.dev/findings/omlx-qwen4-cache-owned-batch-conversion/ https://findings.yatabis.dev/findings/omlx-qwen4-ple-lookahead-prefetch/ https://findings.yatabis.dev/findings/omlx-qwen4-row-exact-fused-decode-deferred-gdn-state/ https://findings.yatabis.dev/findings/omlx-qwen4-ssd-ple-concurrent-pread/ https://findings.yatabis.dev/findings/omlx-qwen4-yarn-rope-long-context-correctness/ https://findings.yatabis.dev/findings/omlx-ragged-batched-mtp/ https://findings.yatabis.dev/findings/omlx-reasoning-output-cache-transcript-identity/ https://findings.yatabis.dev/findings/omlx-resident-speculative-head-under-moe-offload/ https://findings.yatabis.dev/findings/omlx-specprefill-progressive-canonical-recovery/ https://findings.yatabis.dev/findings/omlx-subblock-tail-memory-tier-cache/ https://findings.yatabis.dev/findings/omlx-tail-cache-mtp-history-continuity/ https://findings.yatabis.dev/findings/omlx-turboquant-prefix-cache-retrieval-corruption/ https://findings.yatabis.dev/findings/omlx-workload-aware-paged-cache-block/ https://findings.yatabis.dev/findings/prefill-admission-fairness-under-decode/ https://findings.yatabis.dev/findings/project-obsidian-1a7e533f0dc2-layer-major-expert-union-prefill/ https://findings.yatabis.dev/findings/qwen-image-2-1-comfyui-mps-edit-corruption/ https://findings.yatabis.dev/findings/qwen36-fused-gdn-single-token/ https://findings.yatabis.dev/findings/qwen36-native-hybrid-cache-text-lane/ https://findings.yatabis.dev/findings/qwen38-36-27b-blend-junie-local/ https://findings.yatabis.dev/findings/qwen38-next-architecture-report/ https://findings.yatabis.dev/findings/qwen38-omni-flash-realtime-api-family/ https://findings.yatabis.dev/findings/qwen38-omni-flash-technical-report/ https://findings.yatabis.dev/findings/qwen4-gdn-verify-bit-exact-l2-fusion/ https://findings.yatabis.dev/findings/qwen4-mtp-lazy-speculative-transaction/ https://findings.yatabis.dev/findings/qwengyre-xlong-horizon-agent-rl/ https://findings.yatabis.dev/findings/rapid-mlx-0142-qwen36-compiled-decode-stable/ https://findings.yatabis.dev/findings/rapid-mlx-glm53-mtp-bounded-prefill/ https://findings.yatabis.dev/findings/rapid-mlx-hybrid-recurrent-state-checkpoints/ https://findings.yatabis.dev/findings/rapid-mlx-moe-fusion-reference-lifetime/ https://findings.yatabis.dev/findings/rapid-mlx-parked-mtp-decode-runahead/ https://findings.yatabis.dev/findings/rapid-mlx-prefill-snapshot-tile-alignment/ https://findings.yatabis.dev/findings/rapid-mlx-qwen36-compiled-decode-replay/ https://findings.yatabis.dev/findings/rapid-mlx-qwen36-fused-router-gdn-decode/ https://findings.yatabis.dev/findings/rapid-mlx-singleton-kv-cache-reference-cycle/ https://findings.yatabis.dev/findings/realtime-venus-full-duplex-delegation/ https://findings.yatabis.dev/findings/rmlx-cache-resume-same-split-oracle/ https://findings.yatabis.dev/findings/saragossa-016c0c9ba5c6-skinny-gated-row-oracles/ https://findings.yatabis.dev/findings/saragossa-b80d12a84ab4-u4-affine-least-squares/ https://findings.yatabis.dev/findings/saragossa-resident-metal-prefix-gdn-cache/ https://findings.yatabis.dev/findings/sdpa-transient-aware-memory-planning/ https://findings.yatabis.dev/findings/splash-m4-pro-q6-independent-validation/ https://findings.yatabis.dev/findings/splash-model-specialized-runtime-m5-ultra-paths/ https://findings.yatabis.dev/findings/strands-decider-2b/ https://findings.yatabis.dev/findings/streamoe-1d364eadc28f-prefill-expert-row-budget/ https://findings.yatabis.dev/findings/tensorfold-102-glm-m5ultra-tp2-persistent-prompt/ https://findings.yatabis.dev/findings/tensorfold-1c72d5b21b8a-metal-qmm-tile-unroll/ https://findings.yatabis.dev/findings/tensorfold-2496881f8962-nemotron-optional-packed-metal/ https://findings.yatabis.dev/findings/tensorfold-2e33e297d609-transient-context-pool-queue/ https://findings.yatabis.dev/findings/tensorfold-3207ee374668-qwen35-metal-paged-kv/ https://findings.yatabis.dev/findings/tensorfold-47f397fb29bc-metal-checkpoint-spill/ https://findings.yatabis.dev/findings/tensorfold-490b4c56d517-host-sampling-selection/ https://findings.yatabis.dev/findings/tensorfold-4c8bf272f32e-eac2306459da-prefill-time-share/ https://findings.yatabis.dev/findings/tensorfold-5694dcea9ae4-long-prompt-projection-dispatch/ https://findings.yatabis.dev/findings/tensorfold-623de9687ed9-mamba-stream-pool-wired-budget/ https://findings.yatabis.dev/findings/tensorfold-6a95995b57e5-dflash-contiguous-kv-inplace/ https://findings.yatabis.dev/findings/tensorfold-7520d4c97f1d-deltanet-rollback-bank-allocation/ https://findings.yatabis.dev/findings/tensorfold-754476ab8647-flashnext-image-session-positions/ https://findings.yatabis.dev/findings/tensorfold-755976f06346-macos-memory-pressure-governor/ https://findings.yatabis.dev/findings/tensorfold-922c64d381aa-gemma-masked-nonfinite-contract/ https://findings.yatabis.dev/findings/tensorfold-977b922b07a3-flashnext-metal-rounds-state/ https://findings.yatabis.dev/findings/tensorfold-aa6fbd89030b-metal-dsa-two-key-block/ https://findings.yatabis.dev/findings/tensorfold-bb3741d17a7f-drama-metal-load-admission/ https://findings.yatabis.dev/findings/tensorfold-c20c4dd3a569-packed-mtp-overlap/ https://findings.yatabis.dev/findings/tensorfold-caac8bfbb093-fcfs-prefill-policy/ https://findings.yatabis.dev/findings/tensorfold-e949b9aa4144-drama-native-nvfp4-state/ https://findings.yatabis.dev/findings/veda-sparse-attention-comfyui-apple-silicon/ https://findings.yatabis.dev/findings/veloxquant-13b7e199f1fa-keyformer-absolute-position/ https://findings.yatabis.dev/findings/veloxquant-81b460fac753-single-token-head-batching/ https://findings.yatabis.dev/findings/veloxquant-94bf76f5023a-incremental-age-requantization/ https://findings.yatabis.dev/findings/veloxquant-f0589f21d2c2-d25bcee60ab1-gram-prefill/ https://findings.yatabis.dev/findings/vllm-metal-469aa3ca0e60-mps-single-buffer-budget/ https://findings.yatabis.dev/findings/vllm-metal-8e934fab500d-checkpoint-expert-offload/ https://findings.yatabis.dev/findings/vllm-metal-a236c56aa0d2-kv-override-before-allocation/ https://findings.yatabis.dev/findings/vllm-metal-a3ea0519552a-explicit-kv-byte-budget/ https://findings.yatabis.dev/findings/vllm-metal-cac610b74b84-paged-output-query-dtype/ https://findings.yatabis.dev/findings/vllm-metal-dafa052ff3a0-mixed-decode-work-split/ https://findings.yatabis.dev/findings/vllm-metal-dflash-batch-adaptive-width/ https://findings.yatabis.dev/findings/vllm-metal-dspark-greedy-serving-memory-profile/ https://findings.yatabis.dev/findings/vllm-metal-dspark-scheduler-owned-paged-kv/ https://findings.yatabis.dev/findings/vllm-metal-ec1035607247-packed-gdn-prefill/ https://findings.yatabis.dev/findings/vllm-metal-gqa-multirequest-decode-routing/ https://findings.yatabis.dev/findings/vllm-metal-gqa-p256-p512-long-context-decode/ https://findings.yatabis.dev/findings/vllm-metal-mla-short-context-batch-chunking/ https://findings.yatabis.dev/findings/xing4-0-29b-a4b-mhc-local-scale/