chore(memory): snapshot auto-learned memory and renamed workstation skills

--home-blank-.omp--: MEMORY.md +140 / learned.md / memory_summary.md / raw_memories.md +1002 rewritten by auto-learn after the model rename, frank-medium split-balance and bottleneck studies. Skills renamed to single-topic buckets: dotfile-git-workflow -> ghostty-text-aa + niri-display-rotation + noctalia-theme, handy-service-mgmt -> handy-flm-inference, inference-runtime -> llama-cpp-parallelism. --home-blank-Projects-serverAppConfig--: 38 new rollout summaries, raw_memories.md/MEMORY.md refreshed, skills re-bucketed (bench-protocol, llm-fleet-model-swap, rootless-quadlet-log-retrieval, orchestration-quadlet). Per-worktree bucket (wt-20261006-122204) is now gitignored, not committed.
2026-10-09 13:17:53 -07:00
1 parent a34322aeb0
commit 5b08896e4e
@@ -0,0 +1,4 @@
+thread_id: 01a0e091-9152-75fc-b81b-8e311474eea8
+updated_at: 1790624016
+
+Transitioned frank-primary to Qwen3.8-Flash-Next-GGUF at Q_8. Configured with Q_8 drafter, vision enabled, and layered speculative decoding using mtp and ngram-mod.
@@ -0,0 +1,4 @@
+thread_id: 01a0e983-6c65-77e6-813a-c655e46ce06f
+updated_at: 1790628098
+
+Deployed prefix caching and integrated autoresearch branch updates (PLE timeout 1800s, bench matrix runbook, index/doc-link fixes) for the frank-primary model on R9V hardware.
@@ -0,0 +1,4 @@
+thread_id: 01a0ea08-9e09-7608-8003-81ab447e507e
+updated_at: 1790641878
+
+Purged all documentation mentions of 'Dankware' (retired workstation) from both `~/.omp` and `serverAppConfigs`. Functional artifacts `packages/dankware.pkgs` and `scripts/install.sh` were preserved as the name is load-bearing for host lookups.
@@ -0,0 +1,4 @@
+thread_id: 01a0eb77-a0b2-7748-8c3c-c2eee29d34d4
+updated_at: 1790724034
+
+Replace the current llama.cpp-based runtime of frank-large with the gufo inference engine.
@@ -0,0 +1,4 @@
+thread_id: 01a0f190-490b-765e-b62d-21e28c3c0ec0
+updated_at: 1790760416
+
+Deep research on DeepSeek 4.1 Flash quantizations for the current hardware (2x R9700 gfx1201 + 246 GiB DDR5 RAM). Identified `vcruz305 Q2_K` as the optimal fit for the backbone (~191 GiB) and `GSQ-RCO 3.0-bit` as the high-quality alternative. Key bottleneck is RAM bandwidth (42 GB/s), limiting decode to ~6-11 tok/s. Next step: Build native HIP version from the fork.
@@ -0,0 +1,4 @@
+thread_id: 01a0fb3e-f1d3-7330-bd66-2b997c1a0264
+updated_at: 1790938297
+
+Reconfigured frank-small to llama.cpp MiMo-V2.6-Distill-Qwen-9B-GGUF:Q5_K_M with 3 contexts (180,224 tok/slot) + vision on RTX 3090; frank-large to Qwen3.8-27B-ROCmFP4-STRIX-MTP-GGUF on gfx1151 iGPU. Key: n_ctx_slot=CTX_SIZE/PARALLEL formula, vision bills ~1.4 GiB, legacy images field ignored, compaction overflow risk.
@@ -0,0 +1,4 @@
+thread_id: 01a0fc98-aecb-778a-bad8-f3d0b1e54464
+updated_at: 1790958824
+
+Autoresearch harness for frank-primary (vLLM on Dankserver 2×R9700 TP=2 port 7770) speed optimization. Harness: dev-box orchestrator → SSH → frank-primary-window.sh on Dankserver. Discipline: stop prod unit, boot candidate AS service (same script/port, no scratch ports), in-container measure.py, EXIT trap unconditional restore+health verify. Boot cost ~7min. Staging dir /home/dank/.autoresearch-frank-primary/ holds candidate env/serve/measure/run.json. AGENTS.md has contradiction between topology table (no MTP, 2×262144, bf16 KV) and detail section (MTP4, 200k×3, fp8 KV) — authoritative source is actual live scripts.
@@ -0,0 +1,4 @@
+thread_id: 01a0fd77-162f-7451-9285-566736703023
+updated_at: 1790972101
+
+Queried frank-primary live config; discovered AGENTS.md is stale. Actual candidate: MAXLEN=200704, SEQ=3, KV_BYTES=2.2e9, MNBT=2048. Shell escaping pitfall in bash tool (command substitution mangled). frank-primary subsequently entered repeated socket-disconnect crash loop after config probe.
@@ -0,0 +1,4 @@
+thread_id: 01a0fe18-c297-766a-b631-5889c2444dc2
+updated_at: 1790971702
+
+Determined log access paths for frank-primary on dankserver: engine logs via podman logs (container-lifetime only) or file log at /home/dank/logs/frank_primary.log; journalctl -u frank-primary only has script-level lines since container runs detached.
@@ -0,0 +1,4 @@
+thread_id: 01a0feba-078d-70b7-bdf0-aacc26bf5078
+updated_at: 1791150294
+
+Comprehensive survey of tcclaviger vLLM fork logging levers on frank-primary: standard vLLM env vars (VLLM_LOGGING_LEVEL, VLLM_LOG_STATS_INTERVAL, VLLM_LOGGING_CONFIG_PATH, VLLM_DEBUG_WORKSPACE, VLLM_TRACE_FUNCTION, AMD_LOG_LEVEL) are the only surface; no R4D_*/CLAV_* debug switches exist in r4d.so/parohip/fp8hip. Log file trap: `podman run -d > "$LOG"` captures only container ID, not engine output.
@@ -0,0 +1,4 @@
+thread_id: 01a100c6-1871-7112-b7c8-08ecfdfa896f
+updated_at: 1791176840
+
+Swap frank-large to Uncensored Q6_0_ROCMFPX_AGENT (23.15 GiB, ftype 114) from kingjones777/Qwen3.8-27B-Uncensored-ROCmFP4-STRIX-MTP-GGUF; extend frank-small context to 200700/stream; download MiMo Q8_0 to host.
@@ -0,0 +1,4 @@
+thread_id: 01a100ed-f985-75f9-b64b-22c927319ec5
+updated_at: 1791176847
+
+Frank-primary (Qwen3.8-Flash-Next-Uncensored-FP8 on tcclaviger RDNA4 vLLM fork, dankserver:7770) memory planning analysis. Shape E: fp8 KV, MAXLEN 200704, SEQS 3, MNBT 2048, pool 3.15e9, expert-precise-memory ON. Key findings: arena lever is max_model_len not kv-cache-memory-bytes; SEQS 2 cannot boot (torch.fx ConstraintViolationError); KV price is shape-dependent (9190 B/tok @262k, 7892 B/tok @200k); margin is fossilized estimator padding not live headroom; proposed registry patch to reclaim ~3.5 GiB into expert arena.
@@ -0,0 +1,4 @@
+thread_id: 01a108e0-8f6c-703e-9ba5-a77404744afc
+updated_at: 1791160220
+
+ASUS ProArt B850-Creator WIFI NEO: BIOS reset via CLRTC header short (pins 1-2, 5-10s) or F5 Load Optimized Defaults; max DDR5 spec is 8400+ MT/s (not marketing 9000); 4-DIMM derates below printed ceiling; MT/s = MHz×2; user hit 4600 MHz (9200 MT/s) before failure — well above spec; DIMMs do not need removal for CMOS clear; CMOS clear does not affect module SPD.
@@ -0,0 +1,4 @@
+thread_id: 01a10978-0a99-75e0-877e-8d062824fbb7
+updated_at: 1791160751
+
+Fleet token-throughput benchmark across three llama.cpp endpoints (frank-primary, frank-large, frank-small). Discovered `bc` is not installed on the host. All endpoints healthy (200 OK, <10ms health checks). Sequential 256-token completions completed in ~19.5s total wall time across all three. Recent project work: Q6_K quantization cutover, arena/KV shape tuning.
@@ -0,0 +1,4 @@
+thread_id: 01a1097b-bb20-77d1-b95f-64c86e9e8c0e
+updated_at: 1791160787
+
+User requested a 3000-word creative story about a Soviet pizza analog. Assistant incorrectly evaluated the html-change-report skill as a trigger for creative writing, then was interrupted before completing the story. No output was delivered.
@@ -0,0 +1,4 @@
+thread_id: 01a10981-24c4-75c2-b330-2af61176e8fd
+updated_at: 1791171874
+
+Frank (Chief Designer) operates a multi-model local AI serving stack (frank-primary/large/small) on dankserver; frank-primary is an MoE Flash-Next tower on R9700 DDR5 with MNBT 2048 and 7,762-slot arena, frank-large is a 27B MoE on iGPU with MTP4, frank-small is a 9B hybrid on 3090. Token generation throughput test confirmed ~38 t/s on frank-primary with ~1.8s TTFT.
@@ -0,0 +1,4 @@
+thread_id: 01a10981-39c9-7454-b724-44156879c4d6
+updated_at: 1791160938
+
+Retrieved frank-primary live config (shape G, 2026-10-03): vLLM tcclaviger RDNA4 fork, Qwen3.8-Flash-Next-Uncensored-FP8, TP=2, MAXLEN 200704, SEQS 3, MNBT 2048, UTIL 0.982, fp8 KV 3.1665e9 B/rank, expert-precise-memory ON, 17.77 GiB arena / 7,762 slots, 401,408 KV tokens / 2.00x concurrency, CUDA-graph term 0.07 GiB, gather 32x64, spec decode none.
@@ -0,0 +1,4 @@
+thread_id: 01a10985-1eed-7260-ad2d-49a2af9c96c4
+updated_at: 1791176837
+
+Documented expert arena definition levers in frank-primary vLLM fork: arena = budget - non-expert - KV_reserve(9190B/token × max_model_len) - graphs - overhead. Key levers: max_model_len (lower=increase), --expert-precise-memory (reduces overhead 7.60→5.25 GiB), VLLM_CUDAGRAPH_* env vars (graph term), gpu_memory_utilization (budget). Pool/KV_BYTES is NOT an arena lever. --expert-offload-mem controls host offload (120 GiB). Current shape G: 17.77 GiB / 7762 slots, UTIL must stay 0.982.
@@ -0,0 +1,4 @@
+thread_id: 01a10a47-6406-7086-a869-d8f92b6dc5a8
+updated_at: 1791185010
+
+Added Authentik user `ridley` (ridley@franksdank.cloud) with user-level-only access via Django ORM in `ak shell` on dankclaw. Key workflow: SSH to dankclaw, run `ak shell`, create User with set_password(), assign groups (Non-Admin Users, jellyfin_users, openwebui_users). Verified auth via authenticate() ORM call. Browser SPA interaction is unreliable for Authentik flows—use ORM verification instead. Group membership is the sole gate for app access in this deployment.
@@ -0,0 +1,4 @@
+thread_id: 01a10af2-ce97-7257-9335-d246213db0e5
+updated_at: 1791256146
+
+Diagnosed frank-primary (tcclaviger vLLM fork, Qwen3.8-Flash-Next FP8, 2×R9700, Dankserver) repetition/corruption after ~40k tokens. Top hypotheses: fp8 KV precision degradation at long context, hybrid model (DeltaNet/Gated Attention) recurrent state not correctly checkpointed at preemption/recompute boundaries, and prefix-caching eviction triggering non-identical state recompute. Also recovered frank-large from a rogue-session git clobber (nofile ulimit regression, half-finished Ornith swap) and verified endpoint contract.
@@ -0,0 +1,4 @@
+thread_id: 01a10dd0-8791-71ff-993e-1979264c9275
+updated_at: 1791233132
+
+Diagnosed frank-primary ~40k-token output corruption (repetition/garbage) on dankserver. Exhaustive software hypothesis sweep (fp8 KV degradation, prefix caching eviction, expert offload LRU, RoPE, chunked prefill boundaries, CUDA graph, block table overflow) all failed to reproduce or fix. Root cause: marginal DDR5 at 4800 MT/s on 4 DIMMs silently corrupting data; fix was dropping to 4600 MT/s. Key methodological lesson: consumer non-ECC RAM produces no MCE for silent flips, short pattern checks miss intermittent corruption, and software reverts tested under the same marginal hardware are confounded.
@@ -0,0 +1,4 @@
+thread_id: 01a10dd2-1b9e-7100-bdfe-2dd120a34962
+updated_at: 1791256148
+
+Ornith-1.5-9B-Q8_0 swap on dankserver (AMD/ROCm) replacing MiMo: f16 KV, ubatch 1024, single 262,144 slot. Deploy blocked by tree-clean gate due to parallel docs writer.
@@ -0,0 +1,4 @@
+thread_id: 01a10e21-a0a6-7161-92a2-18c1cf17d7ed
+updated_at: 1791260816
+
+Context bloat diagnosis and skill-library distillation: measured context components (AGENTS.md, memory_summary.md, 175 managed-skills descriptions = major bloat source), distilled skills into vault reference pages on Gitea, staged deletion to ~/.AGENTTRASH/, purged after explicit user approval.
@@ -0,0 +1,4 @@
+thread_id: 01a10f30-003b-74c6-a60f-ae34c9eac5fa
+updated_at: 1791316694
+
+Replaced frank-primary from tcclaviger vLLM fork to radiance (StillDeadcode v1.0.7) runtime serving Qwen/Qwen3.8-Flash-Next-FP8 at native FP8 (no requantization) with TP=2 on 2x R9700, MTP depth 3, expert_tiered placement, and fp8 KV cache. Native FP8 checkpoint at its own precision requires no quantization step or container.
@@ -0,0 +1,4 @@
+thread_id: 01a10f44-6c75-7548-86b1-a7a3603ca8af
+updated_at: 1791270185
+
+ComfyUI workflow goal setup on Strix Halo iGPU (114 GiB GTT, 3090 occupied by frank-small); FP8/Q6 quantization required; ghostty kitty-protocol image display verified as in-spec zero-install path; harness read tool image behavior documented.
@@ -0,0 +1,4 @@
+thread_id: 01a10fad-1e4d-71ba-95a6-376974fd706a
+updated_at: 1791280376
+
+Attempted frank-small swap to ukisai/Swift-1.5-Qwen3.8-27B-GSQ-RCO-GGUF IQ3_S with MTP+vision+FP8 KV on RTX 3090; rollback ordered mid-flight. Final live: Swift RCO-IQ3_XXS + draft-mtp 2×131072. Bonsai2-CRACK + llama-prism parked on disk, not deployed.
@@ -0,0 +1,4 @@
+thread_id: 01a10fc3-5009-7470-980a-fb58b594886f
+updated_at: 1791422424
+
+Authentik user lifecycle: template (ridley shape), audit-before-delete, denormalized-event quirk, Jellyfin DN preservation on recreation
@@ -0,0 +1,4 @@
+thread_id: 01a11006-6c8c-7288-9049-6d53c24be957
+updated_at: 1791283043
+
+frank-large cutover from halogen 0.1.4 (qwen3.8-27b-p1w4d-d2) to halogen-flash-server v2 (Swift1.5 Abliterated) on Dankclaw iGPU. Result: 210 t/s prefill, 30.6 t/s decode, spec-decode draft_n:12/accepted:9. Repo commit 31b1ec3+2d89bf9 pushed to Gitea.
@@ -0,0 +1,4 @@
+thread_id: 01a11414-45b0-744e-894d-0751abafe264
+updated_at: 1791390151
+
+frank-large (halogen-flash-server 0.16.2, Qwen3.8-Flash-Next v2 Swift1.5 abliterated) 400 error when max_tokens exceeds remaining context due to harness token-estimate under-run on code-heavy prompts
@@ -0,0 +1,4 @@
+thread_id: 01a11a28-0cdb-70df-9edd-7d10adfcd213
+updated_at: 1791447271
+
+Fixed halogen 400 max_tokens-does-not-fit error by raising HALOGEN_CTX from 200192 to 262144 (native Qwen3.8-Flash-Next window) on dankclaw frank-large service, enabling 180k+ token sessions. Harness contextWindow set to 217088 = 262144 - 45056. KV pool expanded to 524288 positions (13.3 GiB).