Frank 53ada02a63 frank-default-large: PARALLEL=1, CTX 262144, split 0.50 (max 3090 weight) ...

Single full-context slot instead of two half-context slots. The 3090 takes
~33 of 66 layers (up from 26): the VRAM budget frees up because the f16 KV
pool shrinks 401400 -> 262144. 33 layers is the max with a safety margin
(34 -> 244 MiB headroom, 35 -> OOM). Verified math: /tmp/vram-math.py from
live journal buffers.

2026-08-16 23:53:09 -07:00

..

analyze-spec-matrix.py

docs: Qwen3.8 spec-matrix benchmark page; add matrix tools; final env n_max=5

2026-08-14 14:28:51 -07:00

bench-frank-large.sh

feat: frank-default-large → llama.cpp Qwen3.6-27B Q8_K_P + DFlash draft + ngram-mod (bench pending)

2026-08-11 13:47:37 -07:00

bench-high-ctx.sh

dankclaw: switch backend to Vulkan0 (48% faster decode at 262k ctx); add bench-high-ctx.sh

2026-08-15 17:03:01 -07:00

bench-pipeline-64k.sh

bench: 4k pipeline harness w/ early crash check; parametrized spec-type

2026-08-11 16:14:44 -07:00

deploy.sh

deploy.sh: per-target health wait; 20 min for Dankclaw cold loads

2026-08-13 21:01:01 -07:00

frank-default-large.env

frank-default-large: PARALLEL=1, CTX 262144, split 0.50 (max 3090 weight)

2026-08-16 23:53:09 -07:00

frank-default-large.ling.env

ling B-test: spec off (--spec-type none), keep parallel=2 - isolate spec path as freeze trigger

2026-08-14 03:39:15 -07:00

frank-default-large.service

rename frank-pp-head to frank-default-large: PP split is the new production large endpoint (2026-08-16)

2026-08-16 13:44:31 -07:00

frank-default-small.env

frank-default-small: swap to qwen3.6-moe:35b-a3b at 262144 ctx (native max), keep turbo+asr

2026-08-16 14:48:33 -07:00

frank-default-small.service

promote NPU endpoint to frank-default-small: FLM serve unit + wrapper + env (gemma4-it:e4b default), LimitMEMLOCK=infinity (2026-08-16)

2026-08-16 14:19:37 -07:00

frank-large-ling.service

dankclaw: Ling-3.0-flash Q6_K trial config (Vulkan+GTT+draft-mtp)

2026-08-14 01:20:30 -07:00

frank-pp-worker.env

PP kit + Strix research: Qwen3.8-27B RPC split (Dankclaw head / Dankserver worker)

2026-08-16 05:40:27 -07:00

frank-pp-worker.service

PP kit + Strix research: Qwen3.8-27B RPC split (Dankclaw head / Dankserver worker)

2026-08-16 05:40:27 -07:00

llama-server.service

sync scripts from remotes: ROCmFPX llama-server.service (dankclaw), dankserver vLLM serve/validate; drop failed-attempt scripts

2026-08-10 19:48:46 -07:00

pp-bench.py

PP kit + Strix research: Qwen3.8-27B RPC split (Dankclaw head / Dankserver worker)

2026-08-16 05:40:27 -07:00

pp-test.sh

pp-test: restore drops caches before production load; all-mode EXIT trap auto-restores; ssh ConnectTimeout

2026-08-16 10:17:01 -07:00

serve-flm-small.sh

frank-default-small: pmode turbo, ASR (Whisper-V3-Turbo) on, ctx-len 131072 (max verified)

2026-08-16 14:37:32 -07:00

serve-ling3-vllm.sh

feat: raise Ling-3 GPU memory utilization to 0.95

2026-08-10 23:25:02 -07:00

serve-ling-flash.sh

ling: batch 8192 (ubatch 3072); add -lv verbosity + log file for stop-investigation

2026-08-14 02:12:35 -07:00

serve-llama-frank-large.sh

dankclaw: add MTP=1 toggle - prepend draft-mtp before --spec-default ngram-mod fallback

2026-08-15 22:31:29 -07:00

serve-llama-frank-small.sh

frank-default-small: KV K+V to f16, add verbosity tunable (F16 KV diag)

2026-08-14 12:50:59 -07:00

serve-pp-head.sh

rename frank-pp-head to frank-default-large: PP split is the new production large endpoint (2026-08-16)

2026-08-16 13:44:31 -07:00

serve-pp-worker.sh

PP kit + Strix research: Qwen3.8-27B RPC split (Dankclaw head / Dankserver worker)

2026-08-16 05:40:27 -07:00

serve-vllm-qwen36-dflash.sh

sync scripts from remotes: ROCmFPX llama-server.service (dankclaw), dankserver vLLM serve/validate; drop failed-attempt scripts

2026-08-10 19:48:46 -07:00

spec-matrix-bench.sh

docs: Qwen3.8 spec-matrix benchmark page; add matrix tools; final env n_max=5

2026-08-14 14:28:51 -07:00

test_dspark_stream.py

dankclaw: add accurate DSpark stream/queue test (PARALLEL=1) - reports true prompt_tokens; single+concurrent 16k pass

2026-08-15 14:32:21 -07:00

validate-vllm.sh

sync scripts from remotes: ROCmFPX llama-server.service (dankclaw), dankserver vLLM serve/validate; drop failed-attempt scripts

2026-08-10 19:48:46 -07:00