dank/serverAppConfigs
1
0
Fork 0
Code Issues 6 Pull Requests Actions Packages Projects Releases Wiki Activity

Commit Graph

  • 95219cf4c1 swift15: release the numpy view before closing the PLE audit mmap Frank 2026-10-06 20:35:51 -07:00
  • fee66187a1 swift15: PLE fixed-scale audit (streaming BF16 absmax vs 448*scale cap) Frank 2026-10-06 20:33:55 -07:00
  • 0258ae0897 frank-primary: pin radiance v1.2.0, translate expert-vs-cache-ratio to vram knobs Frank 2026-10-06 20:31:46 -07:00
  • 9aee00f2b3 swift15: fetch script + w4dn8 expert recipe for the frank-primary campaign Frank 2026-10-06 19:19:40 -07:00
  • 41d2f56269 docs: frank-secondary at ubatch 2048 — Measured section, rollback ladder, fleet docs wt/20261006-122204 Frank 2026-10-06 18:40:26 -07:00
  • d8da7a9eb7 frank-secondary: ubatch 2048 (the second rung inside the 112k headroom) Frank 2026-10-06 18:30:35 -07:00
  • 44e7a8dfaf docs: frank-secondary at 2x114688/ubatch 1024 — runbook renamed + rewritten, fleet docs updated Frank 2026-10-06 18:13:05 -07:00
  • f2575a1018 frank-secondary: 2×114688 ctx (112k per-slot), ubatch 1024 Frank 2026-10-06 16:38:58 -07:00
  • 87721ab0fe rename frank-small → frank-secondary (Dankclaw llama.cpp :7771) Frank 2026-10-06 15:37:13 -07:00
  • c71a3066c0 frank-large: host-RAM pin-gate preflight in serve-frank-large.sh Frank 2026-10-06 12:58:34 -07:00
  • df6b5bb6e8 MAX_SEQS 2->3: decode 33.06 t/s (+2.0% over depth-2 32.40; +4.2% over baseline). Bigger batch headroom co-schedules concurrent decode streams (decode is weight-streaming bound, so a fuller batch amortizes the weight read). KV pool 600k tokens (3x200k), above the 400k floor. Draft acceptance unchanged (0.755). Frank 2026-10-06 04:18:53 -07:00
  • 95923d3a98 SPEC_TOKENS 3->2: decode 32.40 t/s (+2.1% over 31.72). Acceptance 0.755 (between depth-3 0.831 and depth-4 0.497) — shallower drafts verify cheaper, net positive. Depth 2 is the decode optimum. Prefill/TTFT shifts are cold-run noise (prefill is orthogonal to spec depth). Frank 2026-10-06 04:09:39 -07:00
  • 81e6122891 frank-primary: autoresearch bench harness (decode t/s @32k) Frank 2026-10-06 03:53:55 -07:00
  • 2d89bf97cc docs: frank-large flash-server v2 cutover — runbook, index, AGENTS.md Frank 2026-10-06 03:35:27 -07:00
  • 20d7182f4b docs index: link frank-large flash-server v2 page Frank 2026-10-06 03:26:43 -07:00
  • 0f9faf05ea radiance runbook: MNBT A/B verdict + rad-tune tune-dir bind-mount Frank 2026-10-06 03:08:30 -07:00
  • 31b1ec397b frank-large: cutover to flash-server v2 (abliterated Swift1.5) Frank 2026-10-06 03:07:40 -07:00
  • 7fef3cb907 frank-primary: persist rad-tune kernel cache on a host mount Frank 2026-10-06 03:02:51 -07:00
  • bb3532dfec log: 2026-10-06 frank-small wave (two reshapes; RCO-IQ3_XXS + MTP-on, 489 MiB free, draft ctx ~2.4 GiB measured) Frank 2026-10-06 01:41:23 -07:00
  • 0f7274e55a frank-small docs: runbook renamed + rewritten for the RCO-IQ3_XXS MTP-on shape (489 MiB free, 63.3 t/s recipe, draft ctx ~2.4 GiB measured); AGENTS, vault indexes, models.yml comment, MiMo cross-link retargeted Frank 2026-10-06 01:33:57 -07:00
  • da36fa32cf frank-small: Swift-1.5-27B Uncensored RCO-IQ3_XXS with MTP back on (two 131072 slots, ubatch 512, q8_0 KV, publisher's BF16 mmproj) Frank 2026-10-06 01:10:43 -07:00
  • 713b126639 frank-primary: cutover measured — docs reflect the live service Frank 2026-10-06 01:06:11 -07:00
  • 2d9cecae6d frank-primary: pre-create the SSD prefix tier path Frank 2026-10-06 01:00:54 -07:00
  • fadad18da1 docs: frank-large FA_TQ A/B rejected at 80k; deep roles rerouted to frank-primary Frank 2026-10-06 00:59:14 -07:00
  • 306ff32427 frank-primary: rad floor under the measured payload (177109 MiB) Frank 2026-10-06 00:55:30 -07:00
  • d39424ffba docs: frank-large cache pin verified warm (16x prefill) + depth law Frank 2026-10-06 00:34:59 -07:00
  • 3eb4fa2b62 frank-primary: rad floor gate to the measured 172.76 GiB container Frank 2026-10-06 00:29:23 -07:00
  • ba67ca5c9e frank-large: pin the prompt cache (16 GB) and capture the engine ledger on disk Frank 2026-10-06 00:26:28 -07:00
  • 8343397584 frank-primary: merged PLE file must sort before the raw shards Frank 2026-10-06 00:21:40 -07:00
  • 5e166543fa docs: record the two-slot no-spec frank-small shape (measured 1,397 MiB free, 48.0 t/s) Frank 2026-10-06 00:15:55 -07:00
  • 812d9ebff5 frank-small: two 128k slots, no spec, ubatch 512 (Frank's tight-fit order, 2026-10-06) Frank 2026-10-06 00:07:03 -07:00
  • fb1a4fd150 frank-primary: never write the stage index through the raw symlink Frank 2026-10-06 00:03:58 -07:00
  • 90a08ab4b5 comfyui: two proven Qwen-Image 2.1 workflows (art + legible text) + payload docs and next-model recommendations Frank 2026-10-05 23:59:29 -07:00
  • 272acb4c7f frank-primary: two stacking fixes from the first live run Frank 2026-10-05 23:56:36 -07:00
  • 13112a93f5 frank-small: record measured cutover values; unit Description follows the Swift swap Frank 2026-10-05 23:41:05 -07:00
  • baf33d016f frank-small: Swift-1.5 IQ3_S-mtp swap (parallel session) + fleet docs index Frank 2026-10-05 23:26:59 -07:00
  • 1919ad7f31 frank-primary: radiance native-FP8 TP=2 cutover (engine, payload build, rollback) Frank 2026-10-05 23:26:51 -07:00
  • 8617b79a18 docs: index frank-primary-freetoken-pr587; mark the r9v q8 runbook retired frank-archon-freetoken-pr587-20261005 Frank 2026-10-05 21:20:05 -07:00
  • 33eae71d33 docs: distill the managed skill library into the vault (waves 1-7) Frank 2026-10-05 21:17:56 -07:00
  • 8618ec6dbe prometheus: drop the frank-primary scrape job (FreeToken exports no /metrics) Frank 2026-10-05 19:38:14 -07:00
  • e2308dff71 docs: frank-primary runs FreeToken (PR587 ROCm) — runbook, index, AGENTS, log Frank 2026-10-05 19:35:31 -07:00
  • 6ba19f6f48 frank-primary: deploy health gate reads the body (FreeToken 200-loading) Frank 2026-10-05 19:27:21 -07:00
  • 0ff016a6b2 frank-primary: KV pool snapped to the qsa_sparse page-64 multiple (299968) Frank 2026-10-05 19:19:24 -07:00
  • bd5c28c4e9 frank-primary: FreeToken (PR587 ROCm) cutover — trio, seam, deployer Frank 2026-10-05 19:17:09 -07:00
  • 4a02dc0980 frank-primary: digest-pin FreeToken base image, symlink libhiprtc dev name frank-archon-vllm-shapeG-20261003 Frank 2026-10-05 19:11:02 -07:00
  • 1f6bdac759 frank-primary: add FreeToken PR587 ROCm serving image recipe Frank 2026-10-05 18:58:57 -07:00
  • a6ad9af708 docs: vault link integrity restored after rogue-session doc WIP landed Frank 2026-10-05 18:40:31 -07:00
  • 380aafa1cc frank-small: Frank overrules q8_0 V — f16 K+V, ubatch 1024 first, ONE native 262144 slot (the only f16/f16 shape that fits Q8 weights on 24 GiB) Frank 2026-10-05 18:31:50 -07:00
  • 5573ca171f frank-small: seam follows the f16 K + q8_0 V contract from 1862dc3 Frank 2026-10-05 18:23:44 -07:00
  • 1c725ea024 e2e: run.sh keeps playwright arg passthrough with the default project set Frank 2026-10-05 18:18:21 -07:00
  • 1862dc3c0f frank-small: env arithmetic for the Ornith Q8_0 rung (f16 K + q8_0 V) Frank 2026-10-05 18:18:06 -07:00
  • 2930431052 frank-small: repoint the deployer payload gate and unit seam at the Ornith-1.5-9B payload Frank 2026-10-05 18:13:01 -07:00
  • 8f53df7b4f frank-small: complete the Ornith swap in seam + deployer twins Frank 2026-10-05 18:10:59 -07:00
  • 9c508218e9 e2e: scope the FrankBook-Pro leg to its own trigger-gated Playwright project Frank 2026-10-05 18:04:02 -07:00
  • 77decb545e frank-small: swap the payload to Ornith-1.5-9B Q8_0 + BF16 mmproj Frank 2026-10-05 17:47:48 -07:00
  • 69f10a17bb frank-small: swap the payload to Ornith-1.5-9B Q8_0 Frank 2026-10-05 17:46:45 -07:00
  • be840b4881 frank-large: halogen container cutover, rootless rlimit law, and the DFlash2 drafter shape Frank 2026-10-05 17:35:05 -07:00
  • b0b840b149 frank-primary: document FRANK_PRIMARY_MNBT=2048 and the chunked-prefill flag surface Frank 2026-10-05 17:32:22 -07:00
  • bdbe814fac frank-large: correct the halogen payload contract to the Hub's truth Frank 2026-10-05 16:45:12 -07:00
  • 87695c5126 download-frank-large-halogen: resume against the resolved CDN URL Frank 2026-10-05 16:31:35 -07:00
  • 0900551317 docs: frank-large halogen runbook, AGENTS.md + indices for the swap Frank 2026-10-05 16:29:11 -07:00
  • fece3ffdd0 frank-large: cut over to halogen container (qwen3.8-27b .hgn + DFlash2) Frank 2026-10-05 16:29:11 -07:00
  • aa9908257d docs(index): frank-small summaries carried the superseded 2,849 MiB free figure Frank 2026-10-05 16:03:18 -07:00
  • 67141c8c02 docs(frank-small): record the batch doubling and scope the ubatch law to its backend Frank 2026-10-05 15:53:06 -07:00
  • e5893dc30d frank-small: double the compute-buffer pair to 4096/2048 Frank 2026-10-05 15:10:15 -07:00
  • b06b47c7b4 docs(fleet): radiance standalone engine probe - qwen4exp refuses fp8 input, quant happens at convert Frank 2026-10-05 14:11:47 -07:00
  • 2e64f66b82 docs(frank-primary): RESOLVED - the ~40k corruption was marginal DDR5 at 4800, fixed at 4600 Frank 2026-10-05 13:54:06 -07:00
  • 35a45b8a66 frank-primary: prefix caching back ON — the OFF experiment did not fix the corruption Frank 2026-10-05 13:13:26 -07:00
  • 1fee99c8ce docs(frank-primary): record the verified 2026-10-05 prefix-cache-off deploy Frank 2026-10-05 05:55:39 -07:00
  • b820d341f5 frank-primary: fix the prefix-cache flag render (127 crash loop) Frank 2026-10-05 05:18:53 -07:00
  • 9d8b18553d frank-primary: prefix caching OFF — the fork's QFN align-mode GDN state fault Frank 2026-10-05 03:45:26 -07:00
  • a1c4701d7e docs(frank-small): Q8_0 rung evaluated and closed — floor break plus -24% decode Frank 2026-10-04 18:17:43 -07:00
  • c4603b130c frank-primary: pin the expert host region to 120 GiB (--expert-offload-mem) Frank 2026-10-04 18:16:38 -07:00
  • 9a144f779e docs(frank-primary): record VLLM_CUDAGRAPH_MEM_ESTIMATE_MB, the graph-term outright pin Frank 2026-10-04 17:51:27 -07:00
  • 50459e4e31 docs: frank-small Q6_K cutover runbook + measured post-boot numbers Frank 2026-10-03 11:20:33 -07:00
  • ff8c0ee264 frank-small: unit Description names the Q6_K rung Frank 2026-10-03 11:20:10 -07:00
  • 6ec0b1a85d frank-small: Q5_K_M -> Q6_K rung at the held 2x200704 f16-KV shape Frank 2026-10-03 11:00:19 -07:00
  • b80886e318 docs: frank-large Q6 uncensored cutover + frank-small 2x200704 f16-KV reslice Frank 2026-10-03 08:01:44 -07:00
  • 47a2c1287b frank-primary: shape G measured — arena 17.77 GiB / 7,762 slots, KV 401,408 tokens / 2.00x Frank 2026-10-03 03:21:55 -07:00
  • c67a2aeb80 frank-primary: shape G v2 — UTIL back to the 0.982 law; the 1.00 force was unbootable Frank 2026-10-03 03:01:55 -07:00
  • 0a4ff30db2 frank-primary: shape G — force the arena to the flag ceiling, KV to 400,700 tokens Frank 2026-10-03 02:49:38 -07:00
  • c682c6329c frank-primary: shape F — MNBT 2048, util 0.982, CUDA-graph plan term pinned to the measured capture Frank 2026-10-03 02:22:49 -07:00
  • 6584c059d6 frank-small: two wide unquantized contexts (2 x 200,704) on the 3090 Frank 2026-10-03 02:04:54 -07:00
  • 7d3dd33de0 frank-large: swap to the uncensored Q6_0_ROCMFPX_AGENT payload Frank 2026-10-03 02:04:22 -07:00
  • 0ae923b879 autoresearch arms: add D/E context arms (KV 4.05e9 = 2.00x under MTP3; E adds MNBT 4096 + util 0.99) Frank 2026-10-02 23:52:04 -07:00
  • b7c3bdb2be autoresearch: temp sweep, engine evidence, env overlays, arms driver Frank 2026-10-02 16:38:17 -07:00
  • d57174c8d3 frank-primary: env-driven spec/compilation/sampling knobs Frank 2026-10-02 16:37:52 -07:00
  • ea1132aa0a docs: record the unmeasured CPU-side recipe-diff candidates (HSA knobs, OMP_NUM_THREADS) Frank 2026-10-02 15:15:13 -07:00
  • 57f33e5289 frank-primary: record the measured shape-E and compile-cache numbers Frank 2026-10-02 15:05:34 -07:00
  • 1ac4d48371 frank-primary: shape E — --expert-precise-memory (arena 6,496->7,522 slots), GPU_MAX_HW_QUEUES=2, and a mounted /root/.cache Frank 2026-10-02 14:37:57 -07:00
  • 812193aa63 docs: the bare-deploy drain law + the plex gate fix in the work log Frank 2026-10-02 12:26:55 -07:00
  • 39f0e0d8d4 plex: probe the readiness gate over HTTPS — PMS drops plain HTTP off-loopback Frank 2026-10-02 12:23:56 -07:00
  • 656012a0fa docs: LLM index + work log for frank-primary shape D, harness window Frank 2026-10-02 12:05:39 -07:00
  • 295d5784d3 frank-primary: record the measured shape-D numbers (398,658 tokens, 1.99x, 6,496 slots) Frank 2026-10-02 11:48:09 -07:00
  • c769ff2f63 frank-primary: 3 slots x 200704, fp8, pool sized for 2 contexts Frank 2026-10-02 11:33:27 -07:00
  • e460bbbab5 frank-primary: revert to known-good fp8 3x262144 Frank 2026-10-02 10:46:01 -07:00
  • 30904bac1b frank-primary: 2 x 200704 fp8, pool cut for the expert arena, MNBT 1024 autoresearch/optimize-the-speed-of-frank-primary-make-it-fast-20261002 Comrade Chief Designer 2026-10-02 09:54:08 -07:00
  • f8f845d86b autoresearch: harness setup Frank 2026-10-02 07:50:50 -07:00
  • 934285c2b2 docs(frank-small): record the 3 x 180,224 + vision shape and what it cost Frank 2026-10-02 03:46:50 -07:00
  • 7fe881de01 frank-small: pool 540672 (3 x 180224) so the card keeps ~2.7 GiB free Frank 2026-10-02 03:17:35 -07:00