Frank 9a144f779e docs(frank-primary): record VLLM_CUDAGRAPH_MEM_ESTIMATE_MB, the graph-term outright pin ...

Verified in the shipped image (v1/worker/gpu/cudagraph_mem.py):
estimate_cudagraph_memory() checks VLLM_CUDAGRAPH_MEM_ESTIMATE_MB FIRST and
returns that MiB total outright, ignoring the BASE/PER_GRAPH pair — the only
way to state the CUDA-graph plan term directly instead of deriving it. We
keep shipping the pair (0/12) so a future spec-decode shape re-prices
automatically; documented the switch condition.

Also corrected two stale shape-G claims in the runbook that predated the
1.00 revert: the deployed shape block and the lever table both still said
util 1.0 / 'le=1 ceiling', while shape G ships 0.982.

2026-10-04 17:51:27 -07:00

..

Pipeline Parallelism

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

agentionai-fast-imatrix.md

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

ciru-activefpx-promptforge.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Dankclaw Improvements.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Dankclaw System Specs.md

frank-large: replace stock llama.cpp with gufo (gufo-org/gufo) on Dankclaw

2026-09-28 22:08:29 -07:00

dankclaw-iommu-translation-off.md

feat(llm): swap frank-large to halogen-flash-server, retire frank-npu

2026-09-11 14:48:28 -07:00

Dankserver System Specs.md

docs: frontier-MoE feasibility on Dankserver (V4.1 / MiMo / GLM-5.3)

2026-09-28 14:58:59 -07:00

dankserver-frontier-moe-feasibility.md

docs: GLM-5.3-Flash readiness revised — tcclaviger fork has native MXFP4/DSFP4

2026-09-29 23:28:04 -07:00

dankserver-qwen38-fp8-expert-offload-tuning.md

docs(frank-primary): record VLLM_CUDAGRAPH_MEM_ESTIMATE_MB, the graph-term outright pin

2026-10-04 17:51:27 -07:00

dankserver-vram-growth.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

dealign-ornith-crack.md

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

deployment.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

dflash2-large-vulkan.md

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

DFlash DSpark Speculative Decoding.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Dual-R9700-Cooperation.md

feat(monitoring,harness): follow frank-medium onto vllm:* metrics and fix red-by-construction targets

2026-09-20 19:35:56 -07:00

eemin-rocmi4-mtp.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

FastFlowLM NPU Inference.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-large-benchmark.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-large-ciru-orca.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-large-gufo-qwen38-27b.md

docs: record the 2026-10-02 swaps — frank-large on ROCmFPX, frank-small on Q5_K_M

2026-10-02 01:51:11 -07:00

frank-large-ik-uncensored.md

docs: the vault and AGENTS.md describe two Dankclaw endpoints

2026-09-16 14:52:51 -07:00

frank-large-kairic-edge-qwen38-27b.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-large-rocmfpx-q6agent-uncensored-mtp.md

docs: frank-large Q6 uncensored cutover + frank-small 2x200704 f16-KV reslice

2026-10-03 08:01:44 -07:00

frank-large-rocmfpx-q8agent-mtp.md

docs: frank-large Q6 uncensored cutover + frank-small 2x200704 f16-KV reslice

2026-10-03 08:01:44 -07:00

frank-large-unsloth-qwen38-27b-q8xl.md

frank-large: replace stock llama.cpp with gufo (gufo-org/gufo) on Dankclaw

2026-09-28 22:08:29 -07:00

frank-large-unsloth-qwen38-flash-next.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-large-vllm-mimo-v26-bf16.md

docs+scripts: reconcile payload inventory after both swaps

2026-10-02 02:31:39 -07:00

frank-medium-batch-ubatch.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-medium-llamacpp-dflash2.md

feat(monitoring,harness): follow frank-medium onto vllm:* metrics and fix red-by-construction targets

2026-09-20 19:35:56 -07:00

frank-medium-model-swap.md

test(frank-medium): A/B DFlash2 vs MTP on FP8 — DFlash2 wins on decode

2026-09-20 22:20:53 -07:00

frank-medium-radiance-mxfp4.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-medium-rocmfpx.md

docs: frank-medium is llama.cpp with DFlash2 on both cards

2026-09-16 15:15:02 -07:00

frank-npu.md

feat(llm): swap frank-large to halogen-flash-server, retire frank-npu

2026-09-11 14:48:28 -07:00

frank-primary-glm53-flash-w4a16.md

frank-primary: qwen cutover docs + deployer identity + seam re-pin

2026-09-30 20:24:49 -07:00

frank-primary-llamacpp-qwen38-flash-next-q8.md

frank-primary: measure expert routing; block choice is immaterial

2026-09-26 21:58:03 -07:00

frank-primary-lucebox-ds4.md

frank-primary: swap Dankserver to llama.cpp Qwen3.8-Flash-Next Q8_0

2026-09-26 19:45:28 -07:00

frank-primary-r9v-radiance-qwen38-q8.md

docs: prefix caching deployed on frank-primary (E8-qualified, 9x/63x reuse TTFT)

2026-09-28 12:33:28 -07:00

frank-primary-r9v-radiance-qwen38-uncensored.md

docs: frank-primary q8 cutover — new runbook, uncensored retired as rollback, fleet table/index/log synced

2026-09-27 22:53:19 -07:00

frank-redo-2026-08-17.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-small-benchmark.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-small-gemma4-12b.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-small-llama-mimo-q5km.md

docs: frank-small Q6_K cutover runbook + measured post-boot numbers

2026-10-03 11:20:33 -07:00

frank-small-llama-mimo-q6k-2x200704.md

docs: frank-small Q6_K cutover runbook + measured post-boot numbers

2026-10-03 11:20:33 -07:00

frank-small-llama-mimo-q6k.md

docs: record the 2026-10-02 swaps — frank-large on ROCmFPX, frank-small on Q5_K_M

2026-10-02 01:51:11 -07:00

frank-small-qwen38-gsq-rco.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

frank-small-sglang-mimo-v26.md

docs(frank-small): record the 3 x 180,224 + vision shape and what it cost

2026-10-02 03:46:50 -07:00

frank-small-vllm-gemma4.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

freetoken-benchmark-sweep.md

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

freetoken-frank-small.md

docs: frank-medium runbook, fleet topology and client cutover

2026-09-15 19:27:04 -07:00

halogen-flash-server.md

frank-small: SGLang MiMo-V2.6-Distill-Qwen-9B INT8 (qwen3_5 hybrid, RTX 3090)

2026-10-01 06:55:48 -07:00

Hybrid CPU GPU Processing Dankserver.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

ikllama-ornith-dankserver.md

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

index.md

docs: frank-small Q6_K cutover runbook + measured post-boot numbers

2026-10-03 11:20:33 -07:00

kairic-edge.md

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

kingjones-q8agent-rocmfpx.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

kingjones-strix-mtp.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

kingjones-strix-rocmfp4.md

feat(llm): swap frank-large to halogen-flash-server, retire frank-npu

2026-09-11 14:48:28 -07:00

ling-3.0-flash-trial.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

ling-3.0-flash-vllm-plan.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Quantization by Hardware Architecture.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

qwen3.8-spec-matrix.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

qwen3.8-strix-optimizations.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

RAM CPU Offloading in VLLM.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Running Qwen3.6 27B FP8 on Dankclaw.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Running Qwen3.6 35B A3B FP8 on Dankserver.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Shared Memory Architecture Dankclaw.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

ubatch-benchmark-large.md

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

VLLM Index.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

VLLM Overview.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

vllm_research_findings.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

x4-pair-experiment.md

the x4 link, measured: 6.72 GB/s — which decides the streaming question

2026-09-16 23:25:06 -07:00