- app.py: thin FastAPI wrapper exposing OpenAI /v1/audio/speech (+ /v1/audio/voices,
/healthz) around OmniVoice's Python API; precomputes a voice-clone prompt per voice
at startup (loaded Whisper auto-transcribes each reference). Replaces the Gradio demo.
- Dockerfile/compose: run the uvicorn wrapper, /healthz healthcheck, project name pinned
to "omnivoice" so the asset-engine liveness probe matches.
- deploy-omnivoice.yaml: stage chatterbox /refs/*.wav as clone voices (skip _* artifacts)
+ verify the API surface.
- services.yaml: catalog entry (id omnivoice, :8199/v1/audio/speech, voice list sourced
live from /v1/audio/voices) + reproducibility_audit row.
Verified live on irv-ml1: /healthz ok, 33 voices loaded, test synth -> 24kHz PCM_16 WAV.
Zero-shot, massively-multilingual (600+ language) voice-cloning + voice-design
TTS (diffusion-LM, Apache-2.0). No official image, so a thin CUDA container
around the pip package running upstream's own Gradio demo (no FastAPI wrapper).
Pinned to GPU 0 (3090) — the A6000 is ComfyUI-exclusive — port 8199. Built +
verified live on irv-ml1 (Gradio 200, container healthy). Surface is the Gradio
UI + Gradio API, NOT OpenAI-compat /v1/audio/speech (wrap later if asset-engine
should consume it). deploy-omnivoice.yaml builds local + verifies.
ComfyUI 0.24.1 added native attention selection; the node-based
BlehGlobalSageAttention errors "does not support the new ComfyUI attention
changes". Add --use-sage-attention to COMFY_CMDLINE_EXTRA so the in-image
sageattention v2.2.0 sm_86 build (rebuilt vs pinned torch 2.12.1) binds via
the native path. OOM flags preserved. Deployed to irv-ml1 + recreated; log
confirms "Using sage attention", container healthy, serving 200.
(comfy-dev request, thread 01KVE89T2DKC)
Operator consolidation (2026-06-18): give ComfyUI the full 48 GB A6000 and move the
audio/TTS zoo to the 3090.
- comfyui: NVIDIA_VISIBLE_DEVICES all -> 1 (A6000 only), + DISABLE_UPGRADES=true to
pin torch at 2.12.1+cu129 so the mmartial boot script stops auto-upgrading it and
the compiled SageAttention kernels stay matched (comfy-dev torch-pin, approved).
- parakeet: NVIDIA_VISIBLE_DEVICES all -> 0 (3090).
Other GPU reassignments are deployment-side (not repo compose): chatterbox-fast via
its .env CBF_GPU_DEVICES=0; vibevoice device_ids ["1"]->["0"] (deployed from
/worktank/vibevoice/build); yt-voice-clipper worker via its override. dia2-2b,
ace-step, csm-expressiva downed (stale/unused).
Result: A6000 = ComfyUI alone (48.3 GB free); 3090 = chatterbox + parakeet + the
on-demand audio (vibevoice/ytvc/kokoro). SageAttention rebuilt against the pinned
torch; OOM cmdline (COMFY_CMDLINE_EXTRA) preserved; /object_info still lists the 9
acceleration nodes.
Serves the in-house abliterated Mistral Small 4 (heretic NVFP4, vision-intact)
under --served-model-name mistral-small-4 on ana-ml2 GPU0:8010 — a true drop-in
for the official mistral-small-4 backend. Both litellm entries (mistral-small-4 +
mistral-small-4-reasoning) route here with no litellm change.
GPU0 fits one mistral-class model, so this is a backend swap, not a co-tenant:
bring up after downing the official stack; revert by downing this and up-ing the
official. Verified live through the gateway: standard returns clean answers,
reasoning populates reasoning_content (the [THINK] split). Checkpoint built per
tools/mistral-small4-nvfp4/.
Quantize a HF-format Mistral Small 4 (Mistral3ForConditionalGeneration MoE) to
NVFP4 with the vision tower intact, then convert HF NVFP4 -> Mistral native so
vLLM can serve it (there is no HF Mistral4 serving path in any vLLM version).
Built + validated end-to-end on ana-ml2 for the abliterated character-model
successor (darkc0de/Mistral-Small-4-119B-2603-heretic): quant -> dry-run (clean
vs the official native NVFP4 reference) -> convert -> serve-test (loads on the
native loader, correct text, vision functional).
Converter scaffold came from worldtree-codex (bf16 bin maps + fused-expert
split); fixed here: NVFP4 layer regexes (keep the `model.` prefix) + non-mmap
shard reads (ZFS large-mmap ENOMEM). nvfp4_quant.py is local. README documents
the pipeline + every gotcha that cost a failed run. Homed here per operator
direction (not Worldtree).
ana-ml2 ran overcommit_memory=0 with zero swap, capping the CommitLimit at
~RAM/2 (~283 GB of 566 GB). The resident vLLM services commit ~224 GB, so a
large model-file mmap (the 50 GB NVFP4 shard during HF->native conversion, or
a vLLM model load) failed with ENOMEM despite ~393 GB of RAM actually free.
overcommit_memory=1 is the conventional setting for ML hosts that mmap large
files. A drop-in under /etc/sysctl.d/ makes it reboot-durable. Operator-directed
permanent (2026-06-17). Idempotent via when:; sudo tee for the root-owned path
(elway runs steps as the SSH user, so a shell > redirect can't write there).
NH3 manager/external-dev box at 10.100.50.42 (Debian 13 VM on nh3-pve),
successor to the retired nh3-ansible. infra-ops identity here is sudo-LESS
by operator decision (2026-06-17): key-only, no NOPASSWD, not in docker
group — user-level management only. Adds servers/nh3-extdev/{README,
ssh-target,system-details.txt}, the CLAUDE.md inventory row, and a local
ssh alias (nh3-extdev -> infra-ops@10.100.50.42, infra-ops key). Login +
sudo-less posture verified.
GLM 5.2 released ~2026-06; confirmed reachable with our existing
Z_AI_API_KEY (z.ai /models lists glm-5.2; a live completion returned
clean). Added two model_list entries mirroring the glm-5.1 pattern:
glm-5.2 (thinking DISABLED by default, per the 2026-06-11 operator call)
and glm-5.2-reasoning (thinking ENABLED, opt-in). Deployed to
ana-docker /opt/docker/conf/litellm/config.yaml, litellm restarted,
both verified through the gateway (disabled -> reasoning_tokens 0;
reasoning -> 234).
Proven 2026-06-16 diagnosing arbo run #5/task 1175: the RUN-level
/runs/{id}/logs 404s, but the per-JOB endpoint
GET /api/v1/repos/{o}/{r}/actions/jobs/{job_id}/logs returns the full
plain-text log (claude-bot basic-auth, internal :3000) — no UI needed.
Also noted gitea's misleading per-step conclusions (every step shows
failure once any fails; trust the log + timestamps).
/healthz catalog_version read the BAKED package version (importlib.metadata),
so a catalog/frontend-only webhook deploy (no image rebuild) left it stale —
v0.12.4 data went live but /healthz still reported 0.12.3. comfy-dev's v0.12.5
reads comfy_catalog.__version__ from the repo-root pyproject.toml; mounting it
on the same checkout mount (catalog/graphs/frontend) makes /healthz report the
MOUNTED version after a catalog-restart. Falls back cleanly if absent.
Pushed to irv-ml1's host compose + validated via `docker compose config`
(bind -> /app/pyproject.toml:ro resolves). Recreate deferred to comfy-dev's
imminent v0.12.5 rebuild (the mount is inert for v0.12.5 itself, which bakes
its own version — it only matters for subsequent catalog-only deploys — so no
separate prod blip). Requested by comfy-dev (engine owner), althing thread
01KV95R88A3Y.
The irv-ml1-arbo Gitea Actions runner (host-executor as lkraven under
systemd) gets the bare service PATH (/usr/local/bin:/usr/bin:/bin), which
omits ~/.local/bin — so the CI uv bootstrap failed with "uv: not found".
Symlink uv/uvx into /usr/local/bin (on the systemd PATH) to fix it and
retire the per-run curl|sh bootstrap. Idempotent (creates: guard);
re-applies cleanly after a runner rebuild. Authorized by comfy-dev
(engine owner) per althing thread 01KV94VTS27B.
Zero-dependency, zero-backend HTML chat UI for the LiteLLM gateway. The
browser talks straight to :4000 (gateway CORS is open), so it's just one
file you open — no container, no stack. System-prompt textarea, model
datalist, streaming SSE, renders reasoning_content for the -thinking/
-reasoning models, settings persist in localStorage.
Deliberately never sends a `tools` field, sidestepping the vLLM "tools
must not be an empty array" bug that breaks the LiteLLM admin UI
playground for vLLM-backed models (litellm #6228; the gateway's
strip_empty_tools hook can't reach the UI's in-process completion call).
Verified against the live gateway: streams + parses a real completion
with no tools sent.
vLLM's OpenAI server 400s on an empty tools array ("tools must not be an
empty array"), which broke every gateway call carrying tools:[] (clients
that send it to mean "no tools" -- OpenAI tolerates it, vLLM does not).
drop_params doesn't help: it drops unsupported PARAMS, not empty VALUES.
Add a CustomLogger async_pre_call_hook (conf/strip_empty_tools.py) that
pops an empty/None tools field (+ orphaned tool_choice) before forwarding,
registered globally via litellm_settings.callbacks so it covers every
vLLM-backed model, not just mistral-small-4. Mounted at
/app/strip_empty_tools.py beside config.yaml (LiteLLM resolves callbacks
relative to the config dir). Surgical: only fires when tools is present
and empty; real tools pass through untouched.
Verified on live gateway (1.87.0): mistral-small-4 and granite-4.1-8b
with tools:[] now 200 (were 400); no-tools baseline unchanged; a real
tool still passes through.
Append 10.100.10.50:8391 to BIFROST_CLIENT_ALLOWED_HOSTS on the personal
Worldtree (.env) so the consumer may bind the memory provider at session-create
(affect :8390 was already listed; the url-guard 422s un-allowlisted endpoints).
Idempotent elway playbook; surgical worldtree-api recreate that auto-derives the
image pin from the matrix sibling to avoid the stale-:latest crash-block footgun.
Repoint servers/corviduo-dev/ssh-target to infra-ops (operator granted durable
NOPASSWD admin on corviduo-dev 2026-06-15).
Put GPU1's idle ~11 GB buffer to work on the two KV-bound models that gained
live consumers from the worldtree migration (granite + the pooling models
under-use their util, so growing them is wasted):
- selene 0.13→0.17: KV 2.53→6.33 GiB, concurrency 1.27x→3.16x @32K (Domari judge)
- qwen36 0.32→0.34: KV 7.73→9.63 GiB, concurrency 2.92x→3.64x @131K (arbo judge +
worldtree actor/echo + gateway)
GPU1 free now ~5.6 GB (safe floor for single-service recreates).
selene: AtlaAI Selene-1-Mini-Llama-3.1-8B judge restored on vLLM after the
llama-swap teardown took its Q6_K GGUF offline. FP8 (dynamic --quantization
fp8; FP8 >= the validated Q6_K fidelity, and text-only Llama so no vision-
tower-noise risk; NVFP4's W4A4 too aggressive for a precision judge). GPU1
util 0.13 (8.51 GiB weights + 2.53 GiB KV, 32K ctx, 1.27x concurrency),
~11 GB GPU1 buffer left. Gateway selene-1-mini-8b → :8011 (shadows the *
wildcard that used to reach it via llama-swap). Judge smoke: scored an
unfaithful claim 1/5 correctly.
mistral-small-4: max-model-len 131072 → 262144 (full native 256K) for
novel-length consistency-checking. KV pool is util-bound (~862K tokens), so
256K costs no extra VRAM — max concurrency just drops to 3.29x at full length.
max-num-seqs 64 → 32 keeps the warmup transient flat (scales with seqs × len),
so it fits the tight GPU0 (free unchanged at 5.2 GB). Verified loaded + healthy.
Operator needs a verified-working vision tower as the abliteration/tuning
baseline. vLLM 0.23.0 crashes Mistral multimodal at startup (#44911
fetch_images regression, ~0.22.1+). Pinned the Mistral container to
v0.22.0 — the last pre-regression release — which loads the NVFP4
(compressed-tensors) AND serves vision: verified a half-blue/half-red
image read correctly ('left blue, right red'). Dropped --limit-mm
(vision re-enabled). qwen36 stays on 0.23.0 (separate container; needs it
for its ModelOpt NVFP4).
- gateway: add mistral-small-4-reasoning. Operator asked for effort=medium
but Mistral's reasoning_effort is BINARY (none/high only — medium 400s);
set to 'high' (sole reasoning-ON level). NOTE: reasoning fires but
reasoning_content-splitting is unreliable on v0.22.0 (lands in content);
clean split would need 0.23.0, which breaks vision — vision prioritized.
- mistral-small-4 (instant) + mistral-small-4-reasoning both gateway-live.
Mistral-Small-4-119B-2603-NVFP4 (119B/6.5B-active MoE, 65.3 GiB) on the
freed GPU 0 (dedicated 96 GB Blackwell), vLLM 0.23.0, :8010. NVFP4 is the
only variant that fits one card (FP8 ~119 GB / bf16 ~238 GB need 2 GPUs).
- TEXT-ONLY: vLLM 0.23.0's Mistral multimodal processor crashes at startup
(fetch_images bug); loaded with --limit-mm-per-prompt image/video=0.
Remove the flag to restore vision once vLLM patches it.
- MLA attn (TRITON_MLA), mistral tool-call + reasoning parsers, util 0.93,
max-len 131072 (capped from native 256K), image pinned by 0.23.0 digest.
- litellm: mistral-small-4 → :8010, shadows the * wildcard.
- GPU 0 reassigned from the (now-offline) llama-swap zoo per operator.
The qwen3.6-35b-a3b VL checkpoint is a single hybrid model with a per-
request enable_thinking switch (Qwen3-style), defaulting thinking ON.
Make the default non-thinking and add an opt-in reasoning variant,
mirroring the existing glm-5.1 / glm-5.1-reasoning gateway split.
- qwen36-vl compose: add --reasoning-parser qwen3 (model-matched) so the
single :8007 endpoint splits <think> into reasoning_content when on and
routes all output to content when off — serving both modes cleanly.
- litellm gateway: base qwen3.6-35b-a3b pins chat_template_kwargs
enable_thinking=false (non-thinking default); new qwen3.6-35b-a3b-thinking
pins enable_thinking=true (opt-in reasoning). Same upstream checkpoint,
no extra VRAM/container.
Deployed + verified on ana-ml2 (vLLM recreated, healthy) and ana-docker
(litellm reloaded): default returns a direct answer with no reasoning_content;
-thinking returns cleanly-separated reasoning_content, no raw tag leak.
Retire qwen35-vl (Qwen3.5-9B); add qwen36-vl serving the official FP8
Qwen3.6-35B-A3B vision MoE on :8007 under its TRUE name only — no alias.
qwen3.5-9b-fp8 is killed at vLLM AND the litellm gateway (404/400); a model is
never served under a prior model's name. Consumer (comfy-dev/arbo) notified +
migrated; arbo vkeys flipped to all-proxy-models; shared all-agents-local key
repointed to qwen3.6-35b-a3b.
GPU-1 rebalance for the heavier FP8 weights (~34 GB): granite 0.35->0.24 /
131K->64K, embed/rerank 0.05->0.03 (reclaimed util-reservation waste). Verified:
vision correct, 20-concurrent/endpoint load test = no OOM (~7.5 GB headroom).
Drop the llama-swap qwen3.5-9b GPU-0 pin (GPU 0 freed for the creative-writing
hot-swap card). NVFP4 was the lighter fit (~21 GB) but its vLLM ModelOpt-MoE
loader is broken (KeyError w2_input_scale / lm_head.input_scale, vllm #44081);
revisit when fixed.
Operator's call: keep the arbo stack in eshpfi and version its deploy machinery
alongside the compose (was host-only on irv-ml1 = recoverability foot-gun).
- arbo-webhook.py: :9009 HMAC listener (secret externalized to host file, not git)
- arbo-deploy.sh: internal-route fetch + catalog-only targeted restart
Document both in the README Q5 section + the internal-gitea-route gotcha.
Fleet/colo hosts must reach gitea over the internal route (ana-docker
container git-SSH at 10.250.50.70:222), not the public gitea.phasefinal.com:22
which fail2bans the host's egress IP and silently wedges webhook auto-deploys.
Bit irv-ml1's arbo deploy 2026-06-13.
storetank archive fully resolved (919 G -> 0): ~739 G killed (superseded/niche),
177 G migrated into arbo, rest dupes. Rewrite the archive doc as a decommission
record; refresh the arbo catalog to its post-migration 502 G state (+ SDXL/Pony
stack + 9 gen-agnostic utility categories).
Swept the orphaned llava_llama3 (HunyuanVideo text encoder, 23.5 G) after the
Hunyuan kill left it unreferenced. Update the curation table + remaining total.
Capture the 2026-06-13 archive curation pass (919->238 G, 681 G reclaimed:
Hunyuan + WAN2.1 + FLUX.1 + umt5 orphan, all superseded by arbo's current-gen
stack) and a detailed catalog of the remaining 238 G (SDXL/Pony stack, SD3.5/
Chroma, gen-agnostic utilities, shared encoders) for comfy-dev's migration
decisions into the active arbo set.
Operator decision 2026-06-13 (relayed by comfy-dev, confirmed in-session):
turn off the prod arbo engine's bearer auth and rely on the WireGuard
perimeter. Reverses ADR-0001's open-auth-hole-closed posture (comfy-dev owns
the ADR update on the vh/arbo side).
The app's protected-gate no-ops only when ENGINE_TOKEN is ABSENT — an empty
string still gates (verified: ENGINE_TOKEN="" -> /workflows 401). So both
inject paths are removed: the compose environment line is commented out and
the .env line deleted on the host. Result: tokenless GET /workflows 200 (was
401), matching the dev engine. Original token preserved in the host's
.env.pre-auth-off.bak for re-enable.
playbooks/arbo-disable-engine-token.yaml captures the reversible procedure.
The arbo-prompt-enhance vkey was extended to reach qwen3.5-9b-fp8 for the
hero auto-judge step (v0.11.3+), not granite-only. Confirmed via /v1/models
for the key. Docs-only; no version bump.
ana-ml2 was upgraded 2026-06 from dual RTX 6000 Ada (46GB, cc 8.9) to
dual RTX PRO 6000 Blackwell Max-Q (96GB, cc 12.0 / sm_120). Update the
stale hardware facts across the workspace:
- CLAUDE.md servers table row
- servers/ana-ml2/README.md hardware spec (+ refreshed system-details.txt)
- stacks/vllm compose + .env.example FP8/KV comments (Ada cc 8.9 -> Blackwell cc 12.0)
- stacks/llama-swap config VRAM-budget comment (48GB -> 96GB, GPU-0 pin)
Also corrects the adjacent stale 'Phi-4-mini' comment in the granite
service block (the service has been Granite 4.1 8B since 34a43a0).
Doc/comment-only; no runtime change.
Benched granite prefix caching at ~6.5x faster TTFT (45ms cached vs 292ms
uncached) on a shared ~4.5k-token summarizer template. granite already had it
on by vLLM-v1 default; pinned explicit so a version flip can't silently disable
it. qwen (nightly) defaulted it OFF -> flipped on (free for the text-chat path,
marginal for vision where each image is a distinct prefix). Soft/evictable KV,
zero memory change (GPU1 still ~3.7GB free), all 5 services healthy.
ComfyUI's ~325G model tree moved off the near-full worktank NVMe (97%->26%,
342G free) to /storetank/arbo (roomy SATA SSD on irv-ml1), overlay-mounted
back at /basedir/models so ComfyUI behaviour is unchanged. rsync byte-verified
(src==dst), one comfyui restart, worktank original removed. Inventory of the
set in docs/arbo-comfyui-model-catalog.md for the retain decision. The older
919G /storetank/image-models/comfy archive is untouched (separate reclaim).
Qwen3.5-9B vision-language served FP8 on ana-ml2 GPU1 (co-located with
granite + the embed/rerank/reward trio; GPU0 kept free for hot-loading
large models), :8007, fronted by LiteLLM as qwen3.5-9b-fp8.
Pinned to vllm/vllm-openai nightly@sha256:49211ab2 — :latest (v0.19.1)
quantizes the VL vision tower under fp8 and garbles vision; the nightly
correctly excludes it (LM stays FP8, vision tower BF16). util 0.40
(~38GB) on the shared card (vLLM needs free>=util*total here). Vision
verified end-to-end through the gateway.
SGLang 0.5.13 confirmed to support our formats on Blackwell sm_120
(compressed-tensors NVFP4 W4A4, fp8, modelopt_fp4, petit_nvfp4, fp4_e2m1 KV),
so the bench can be a real NVFP4 head-to-head. Parameterized compose (model/
quant/GPU via .env) + a common streaming load generator (bench.py: agg tok/s,
TTFT p50/p99, TPOT) so both engines are driven identically on an exclusive GPU.
Bench-oriented; promote to a real stack only if SGLang wins. Launch deferred
until the NVFP4 eval frees a GPU.
Ada->Blackwell swap doubled card VRAM, so the Ada-era fractions (0.07/0.07/
0.18) reserved ~2x the bytes for the same models. Empirically re-floored via
0.01-step climb until each service was stable under 20x parallel inference:
embed/rerank 0.05 (load-floor for the 0.6B models), reward 0.10 (the real
over-provision). Frees ~11 GB on GPU 1. Live .env on ana-ml2 already applied.
ana-ml2's Ada->Blackwell swap (2x96GB) frees GPU 0 entirely. Pin llama-swap
to GPU 0 via NVIDIA_VISIBLE_DEVICES so on-demand large-model hot-loads land
there, off GPU 1 where the always-on vLLM services (granite + embed/rerank/
reward) live. Closes the long-standing 'pin llama-swap to GPU 0' item.
Extends the catalog/graphs git-pull-mount pattern to the SPA frontend so
frontend changes reach prod via git pull + restart, no image rebuild.
Delivers the v0.11.2 auth-on catalog-load fix without a rebuild; baked
image frontend stays the fallback.
New stack mirroring the canonical convention for the Arbo (catalog) engine,
co-located beside comfyui on irv-ml1 per ADR-0001 D1/D3:
- engine<->ComfyUI over traefik-net container DNS (http://comfyui:8188),
SSH dependency eliminated; file ops bind the shared basedir input/output
- named local-disk volumes for the gallery SQLite (arbo_db, restic-backed)
and hero images (arbo_heroes); catalog as a ro git-checkout mount (D2)
- ENGINE_TOKEN + GRANITE_KEY via on-host .env; GRANITE via the LiteLLM gateway
- Q5 catalog-pull: manual day-1, ytvc-style webhook follow-on
Image build, /healthz, catalog in-container path, and non-root UID are
comfy-dev's to confirm (CONFIRM items in README).
glm-5.1 now disables GLM thinking by default via extra_body (LiteLLM strips
top-level thinking under drop_params but forwards extra_body verbatim to z.ai).
New glm-5.1-reasoning alias = same upstream with thinking enabled, so reasoning
is opt-in. Operator call 2026-06-11; primary driver is the pi coding harness.
Verified live: glm-5.1 reasoning_tokens=0, glm-5.1-reasoning reasoning_tokens>0.
Adds graphify-out/GRAPH_REPORT.md (knowledge-graph map of the repo) and a
.gitignore block that tracks only the lightweight map while ignoring the
regenerable graph.json/cache/html. Part of the fleet-wide Graphify rollout.
Local post-commit auto-rebuild hook retained.
- migration-plan.md: workstation-only migration of nh3-dev's dev env to
the M2 Ultra Mac Pro Rack (on-subnet, sidecars stay on Linux).
- nh3-dev README: drop volva from live sidecars + add retired note.
volva.service/heid.service removed by infra-ops at heid's request
(re-architected daemon->CC-orchestrator; volva.service was crash-looping).
LLM observability for the fleet — pretty trace UI over the gateway: prompts,
completions, reasoning, latency, token counts. The pretty layer LiteLLM's
spend_logs lacked.
- stacks/langfuse: v3 self-host stack (web/worker/postgres/clickhouse/redis/
minio) on ana-docker, adapted from upstream. UI on :3001 (gitea owns :3000).
Project + API keys auto-provisioned via LANGFUSE_INIT_*. HOSTNAME=0.0.0.0 on
langfuse-web so it's reachable via the published port while also on tnet.
- litellm: enabled success_callback/failure_callback: ["langfuse"] (the
passthrough env was already wired); keys + host go in the litellm .env.
Verified: stack healthy, project keys authenticate, and a real gateway call
landed a litellm-acompletion trace in Langfuse within ~6s. Secrets live only in
the server .env (never committed).
Granite 4.1 8B beat phi4-mini on precision in brokkr's R15 P03 eval, so it's
the new production summarizer/dreamer for nevermore.
- vllm-phi4 -> vllm-granite: official IBM FP8 (ibm-granite/granite-4.1-8b-fp8,
compressed-tensors), GPU 1, 50K ctx, FP8-KV, CUDA graphs. Same :8004 slot.
- GPU 1 retune: the embed/rerank/reward trio was over-provisioned (embed ran a
5.89x KV pool, reward 3.90x). Trimmed utils 0.20/0.20/0.30 -> 0.07/0.07/0.18,
freeing ~10 GB so granite runs with CUDA graphs (not --enforce-eager) and
keeps ~10 GB free as a hedge for future Granite text-LoRAs (--enable-lora).
- LiteLLM: phi4-mini model_list entry -> granite-4.1-8b (hosted_vllm @ :8004);
explicit entry shadows the '*' wildcard's llama-swap route.
- nevermore repointed (LLAMA_SWAP_MODEL=granite-4.1-8b via the gateway) live.
Verified end-to-end: vLLM :8004 generates, gateway routes (gateway-granite-ok),
KV 86,768 tokens/1.69x at 50K, 0 restarts, GPU 1 10.3 GB free.
ComfyUI's default cudaMallocAsync allocator phantom-OOMs ('allocation
would exceed allowed memory', 0 bytes allocated) when the A6000 is
shared with the pinned TTS services (~22 GB used). --disable-cuda-malloc
switches to PyTorch's native allocator. --fp8_e4m3fn-text-enc loads the
FLUX.2 Qwen3-8B text encoder as fp8 (~8.7 GB) instead of upcasting the
fp8 file to fp16 (~16 GB), matching the box's Ampere-fp8 posture.
Applied via COMFY_CMDLINE_EXTRA in the canonical compose; verified the
allocator flipped to 'native' and both flags are on the live cmdline.
IBM Granite 4.1 dense instruct (3B + 8B), unsloth Q4_K_M GGUF via -hf
syntax, 64K ctx with q8 KV. Auto-exposed through the LiteLLM gateway
wildcard (ana-docker:4000) and direct on llama-swap (:9292).
Replaces the retired irv-ml1 Ollama granite4.1 — Ollama is now banned
fleet-wide; serving consolidates onto the sanctioned llama-swap/vLLM
substrate.
The asset-engine catalog source of truth. Removes the chatterbox
exaggeration/cfg_weight sliders (proven Turbo no-ops) — reconciling the
canonical with the fix previously applied only to asset-engine's vendored
copy — and adds the csm-expressiva whisper TTS entry (irv-ml1:8198).
Reflects the shared install: browsers in /opt/ms-playwright (root-owned,
world-readable) + PLAYWRIGHT_BROWSERS_PATH wired globally via
/etc/environment + profile.d, so any user/project resolves the shared
binary without a per-home playwright install. Includes the add-a-browser
command for future bumps.
Adds the NH3 dev box (10.100.10.50) as a managed host: README, first
system-details snapshot, ssh-target. Adds it to the fleet bootstrap's
Tier 1 and the CLAUDE.md servers table. infra-ops identity bootstrapped
there (operator-run) — NOPASSWD sudo + key, so root-level infra work on
the box (it runs the egress proxy, ttyd seat, mead-hall, volva, and dev
checkouts) no longer needs a per-task password.
First use: installed Playwright headless-Chromium system deps + binary
for bloom_music's OSMD browser-geometry test harness; headless launch +
real SVG geometry verified.
Adds explicit gateway entries for the four z.ai GLM models (glm-5.1,
glm-5-turbo, glm-4.7, glm-4.5-air) routed to api.z.ai with Z_AI_API_KEY,
plus the compose env passthrough + .env.example doc. Explicit entries
win over the llama-swap wildcard (distinct IDs, no collision). Extends
the gateway's unified logging to cloud inference, not just local
vLLM/llama-swap.
Cost note: paid API — only gateway-keyed callers reach these, but calls
spend z.ai credits (documented in config + compose comments).
Adds a `model_name: "*"` entry routing any unmatched model to llama-swap
(ana-ml2:9292) so its whole swappable LLM zoo logs through the gateway
without per-model registration — add/swap models in llama-swap freely,
litellm logs them all. Exact entries (phi4-mini/qwen3-embedding/
qwen3-reranker → vLLM) still win; the wildcard only catches the rest.
litellm does no inference; llama-swap keeps loading + serving. Enables
routing worldtree-personal's generative chat through the gateway for
full req/resp logging while preserving llama-swap's on-demand swapping.
LiteLLM proxy fronting the vLLM services on ana-ml2 so every request +
response is captured and inspectable in a browser Logs UI — the
visibility vLLM itself lacks (Dozzle shows only connection metadata).
- compose: litellm (proxy + /ui Logs) + litellm-db (Postgres store)
- conf/config.yaml: routes phi4-mini (chat, :8004), qwen3-embedding
(:8001), qwen3-reranker (:8002); store_prompts_in_spend_logs persists
full prompt/completion text. reward classifier (:8003) stays direct
(no first-class LiteLLM route).
- Langfuse-ready: lean first cut intentionally skips Langfuse's heavy v3
stack; graduating is one env-var + callback step, no re-architecture.
- roadmap: mark the vLLM-observability item's first cut as shipped.
Lean first cut of docs/roadmap.md "Observability for the vLLM stack".
Operator chose option (ii): keep the OFFICIAL Phi-4 format globally rather than
impose Ollama's leaner scaffold on every phi4 consumer. Removes the
--chat-template override + the conf/phi4-chat-template.jinja file (90e08f0).
vLLM now uses the tokenizer's built-in template (system <|end|> present);
verified 7-token render via tokenize/detokenize. brokkr re-baselines its R15
canonical on the official scaffold so baseline == production.
vLLM's official Phi-4 tokenizer template emits <|end|> after the system turn;
Ollama's does not. That single boundary token regressed brokkr's R15 P02
admission eval (type macro-F1 -33pp) vs the Ollama-measured canonical, while
valid_format held at 1.0. Operator chose to make vLLM match Ollama's leaner
scaffold globally (baseline == production). Adds conf/phi4-chat-template.jinja
(drops the system <|end|>) + mounts it + --chat-template on vllm-phi4. Applied
prompt verified via tokenize/detokenize; brokkr re-smokes probe_vllm.yaml.
Default gcc is gcc-11 (CUDA/torch R14 pin) but the only gfortran is 12,
whose libgfortran-12-dev puts the unversioned libgfortran.so only in the
gcc-12 dir. R links shared objects with gcc-11 + -lgfortran, so the whole
Fortran dependency chain (quadprog/tseries/RcppArmadillo -> soundgen) failed
'cannot find -lgfortran'. Adds an idempotent step installing libgfortran-11-dev
(guarded on 'gcc -print-file-name=libgfortran.so' not resolving). Additive;
default toolchain untouched.
No-sudo phase: uv, R16 world venv + pyworld (setuptools<81 pin for
pyworld's pkg_resources import). Sudo phase (run as infra-ops): R +
soundgen system deps, and pin the Ollama service to the A6000 BY UUID
(CUDA FASTEST_FIRST ordering != nvidia-smi PCI order on this box, so an
index pin is ambiguous). R15 models share the A6000's free headroom —
yt-voice-clipper + R14 keep running, card not freed.
Adds a host-agnostic elway play + fleet driver that stand up an
`infra-ops` system user (dedicated ed25519 key, NOPASSWD sudo with
log_output audit, docker group) so the infra-ops agent completes
DevOps work end-to-end instead of handing sudo steps back to the
operator. Scoped to PFI-owned Linux boxes; tiered (compute/app/
sensitive-infra) with SureFire/corviduo/esh/Synology explicitly
excluded. Validated live on irv-ml1.
worldtree-dev sent infra-ops a manual deploy request for v0.30.6 to
personal:8081; operator confirmed all three corviduo-dev Worldtree
deployments go through the Worldtree team's CI/CD, never a manual
infra-ops deploy. The old 'personal — manually managed' line was wrong
and misleading. docker-as-root pattern is emergency-ops (key restore)
only.
Auto-deploy on push to main: gitea webhook → HMAC-validated listener on irv-ml1:9008
→ git fetch/reset + docker compose up -d --build. Documents the gitea-server
ALLOWED_HOST_LIST anti-SSRF relaxation (scoped to the WG net), the irv-ml1 components
(deploy key, git-clone deploy dir preserving the proxy override/secrets, listener +
user service), and verify/debug steps.
YouTube (and a growing set of services) hard-flag datacenter IPs, bot-gating
even public content regardless of cookies/PO-tokens. Origin case: yt-voice-clipper
on irv-ml1 (Irvine colo) — every yt-dlp fetch returned LOGIN_REQUIRED. Confirmed
pure IP reputation: the same public video fetches cleanly (no cookies) once routed
through nh3-dev's residential egress (70.230.226.88).
- scripts/setup-nh3-egress-proxy.sh: idempotent dante (SOCKS5) install + config.
Internal-only ACL (10.100.0.0/16), bound to the WG interface, systemd-managed.
- docs/runbooks/nh3-egress-proxy.md: purpose, usage, security model, caveats.
Reusable fleet egress, not yt-voice-clipper-specific.
chatterbox-fast is authored software with a test suite, not a config-mirror stack —
so it moves to its own MIT-licensed, versioned, CI'd repo (gitea vh/chatterbox-fast,
v0.1.0) following the sister-repo pattern. Replace stacks/chatterbox-fast/ with a
pointer README; the moved code (scheduler/app/bench/tests/Dockerfile/compose) now
lives in the new repo. The deployed :8197 service is unaffected (still runs the
legacy devnen-based image; self-contained-image migration is an optional follow-up).
The fleet catalog entry stays in docs/asset-engine/services.yaml.
- chatterbox-fast experimental -> ready: browser audition verified end-to-end
(operator confirmed progressive playback "excellent" 2026-06-02).
- vibevoice ready -> down: no container running on irv-ml1 (connection refused);
catalog status was stale.
- voxtral: NOT a stale typo — its stack genuinely claimed :8197, the port now held
by the live chatterbox-fast. voxtral is down, so moved IT to :8201 (catalog
endpoint + source_url, stacks/voxtral/.env.example + README, host .env) rather
than disturb the live service. No live clash existed (voxtral down) but it was a
latent deploy-time collision I introduced by placing chatterbox-fast on 8197.
No catalog_version bump (status changes + endpoint correction, additive). Validates
against the schema.
Land chatterbox-fast in the asset-engine catalog as an additive service, per
asset-engine-dev's shipped streaming-audition path (asset-engine v0.1.17-19):
- streamable:true -> UI routes Generate to an ephemeral progressive-<audio>
audition (no Job/Asset); re-run on `chatterbox` to keep output.
- New service-level `streamable` bool added to services.schema.json (additive,
default false; mirrors the Pydantic model asset-engine-dev regenerates).
- Fields: text, voice (select via /voices), temperature/top_p/top_k/
repetition_penalty/seed, format (pcm default; UI forces wav). exaggeration/
cfg_weight omitted — Turbo ignores them.
- status experimental until the first real browser audition verifies progressive
playback (the one thing asset-engine-dev couldn't machine-verify).
- reproducibility + audit entries added. No catalog_version bump (additive).
Validates against the updated schema.
TTSRequest gains `seed` (0=random); seeded once per request under the lock via
torch.manual_seed + cuda.manual_seed_all. One-shot output is then byte-reproducible
for a fixed seed+params (verified: seed=42 -> identical sha256 across runs).
Streaming stays non-reproducible by design — adaptive-chunk boundaries depend on
live-measured RTF. Needed for the asset-engine catalog reproducibility contract
(parity with the chatterbox sibling, which exposes seed).
stream=false + format=wav emitted the streaming 0xFFFFFFFF-length header, so a
buffered consumer reading a complete wav got bogus RIFF/data sizes. One-shot knows
the full length, so emit correct sizes; streaming keeps the open-ended header
(length genuinely unknown up front). Verified remote: one-shot wav data size ==
bytes-44, python wave.open() reads 2.20s cleanly; streaming still 0xFFFFFFFF.
Deployed on irv-ml1 beside live chatterbox (:8196): healthy on :8197, TTFB ~0.5s,
no starvation. Measured VRAM 5.34 GB (fp32) settles the placement: the 3090's
~3.8 GB free does NOT fit, A6000 (device 1) is the only viable card.
Revert the priming feature from d707439. Live A/B caught an audible artifact: the
context-priming discard-cut left part of the throwaway prefix in the output, so a
clause ("...without a trace of sarcasm,") was spoken an extra time.
Root cause is structural: generate() returns one finished waveform with no marker
for where the prefix ends, and the model renders the same prefix with different
timing when followed by content than when generated solo — so the duration-estimate
+ energy-minimum cut is a guess and can leave a sliver (or a whole clause) of prefix
in. A reliable cut would need token-level access (the abandoned native-streaming
arc) or a per-chunk ASR/alignment pass (heavy, still imperfect, eats the latency
budget). Fails the agreed bar: "keep only if it closes the gap without a seam."
Kept from d707439: the .gitignore (build artifacts). NOT re-applied: the bundled
margin_first fix — wiring it would shrink chunk 1 (more joins = worse coherence),
against the operator's priority, and margin=0.8 there is already starvation-safe.
Coherence loss at joins stays an accepted limitation; cold streaming was judged
"really good". Phase 1 + Phase 2 parity/perf untouched. Next: Phase 3 deploy.
Prime early joins by prepending the prior sentence as backward prosodic context,
generating context+content together, then discarding the context audio. The cut
snaps to the inter-sentence pause (energy-minimum search around the context's
solo duration) with a 5ms fade-in to kill any seam click (app: _cut_at_pause /
_fade_in / Engine.generate_primed). Opt-in via request `prime` (default off).
Scheduler: priming is AFFORDABILITY-GATED so it can never starve. A primed chunk
costs ~(2·context + content)/rtf (a 2nd context-solo pass); a chunk is only primed
when buffer ≥ prime_buffer_factor (1.5) × that cost, else it falls back to a cold
generate. Consequences proven in the GPU-free sim (17 tests):
- fires on early joins for any GPU at/above rtf_prior (3.4 = 3090; A6000 ~3.8-4.0)
- self-skips (degrades to cold) on a slower-than-fleet GPU rather than starving
- never primes chunk 0 (latency-critical)
Also fixed a latent Phase-1 bug: margin_first was applied at chunk 0 (budget always
0 there) so it never did anything — now applied at chunk 1 (the first transition).
Live A/B on irv-ml1 (A6000, GLaDOS): TTFB unaffected (445 vs 467ms), no starvation;
priming fired on chunk 2 (gen 1.6s for the doubled pass). On typical text exactly
ONE early join safely primes — priming chunk 2 flattens the buffer so later/larger
chunks no longer clear the safety gate. Samples: ~/chatterbox-ab/_p2_{cold,primed}.wav.
- /voices endpoint lists predefined voice stems (excludes `_`-prefixed bench/A-B
scratch wavs); shared _predefined_wavs() also feeds default-voice discovery.
- Perf levers: TF32 matmul/cudnn + flash/mem-efficient SDPA, default ON, env-gated
(CBF_TF32 / CBF_SDPA_FLASH). Startup logs model dtype.
Measured on irv-ml1 (turbo, A6000): the model loads FLOAT32 (not the fp16 older
notes assumed). TF32+SDPA do NOT move TTFA (489->514ms, noise) — first-sentence
latency is bound by the sequential AR token decode at batch-1, not matmul
throughput. bf16 (the lever that would help) is DEFERRED: from_pretrained() has no
dtype arg and turbo's fp32 conditioning path + dtype-sensitive vocoder make a
clean cast nontrivial; not worth the quality risk at ~0.5s TTFA. torch.compile
also deferred (batch-1 regression). Findings recorded in README.
Voice management parity (predefined dir + per-request clone refs) was already in
the Phase-1 resolve path; /voices completes the surface.
Build the streaming TTS server MVP per docs/design/chatterbox-fast-plan.md §4.
- scheduler.py: adaptive buffer-ratchet chunker (the meat) — GPU-free pure
logic. First sentence emitted alone for low TTFA, then chunks ratchet ~3x by
packing whole sentences to margin x buffered-audio; drives off measured RTF +
sec/char (EMA). relieve_leader() clause-splits a too-big mid-stream sentence
to avoid starvation (joins land on commas); a long comma-less sentence is the
one honored-but-flagged limitation.
- test_scheduler.py: GPU-free simulation, 13 tests — asserts no-starvation
(incl. overestimated RTF) and the ratchet.
- app.py: FastAPI model holder + POST /tts StreamingResponse (raw PCM s16le
default, wav optional, stream/oneshot) + GET /health.
- bench.py: client — ground-truth TTFB + real 1x-consumer starvation check.
Live test on irv-ml1 (turbo, A6000, GLaDOS voice): streaming TTFB 499ms vs
oneshot 5230ms (~10x), stayed ahead of a 1x player (no starvation), ratchet
1.64->4.08->8.60->8.60s audio, measured RTF self-corrected 3.38->4.01.
Kill the superseded docs/design/chatterbox-fast.md — its §5 windowed-token
streaming was the abandoned native-frame-streaming arc; the adaptive-chunk plan
supersedes it. Repoint persistent-memory + README at the canonical plan.
Self-contained build plan for the chatterbox-fast streaming engine: the
adaptive buffer-ratchet chunking design, validated turbo API + facts, the
GPU-1 dev/test container pattern, 4 build phases, the base-fork A/B, and
watch-outs (incl. native-turbo-streaming is abandoned). Intended for a
fresh-context session to execute at full strength.
asset-engine shipped the per-field enable-toggle (v0.1.14/.15) — the
durable fix for the "form submits untouched fields" family. A field
marked togglable:true renders with an OFF-by-default switch: while off
the control is disabled (excluded from submission) AND the server skips
injecting its default, so it is genuinely not sent until the user opts
in.
Per operator direction, opt fish-s2's `references` (inline-base64
Custom-clone) field in — it already satisfies the togglable-requires-
optional validator (optional:true, no default). The advanced clone
field now renders dormant and can never silently override the Voice
dropdown again.
This is a SCHEMA change (new CatalogField property), so:
- services.schema.json: add `togglable` (boolean, default false),
mirroring the asset_engine Pydantic model that generates this schema.
- catalog_version 1 -> 2 (header: bump on schema changes).
- CATALOG-CONTRACT.md: consumer pin note -> catalog_version=2.
Scoped to `references` only. The chatterbox/dia2 clone fields are the
same family but NOT toggled: dia2 deliberately defaults voice_mode=clone
+ a clone ref as its stable out-of-box voice, and toggling that field
would change dia2's default-voice behavior (the earlier 404 fix).
Validated: jsonschema accepts togglable; additionalProperties:false
guard still rejects unknown props.
reference_id=<name> resolves against the DIRECTORY references/<name>/
(audio + same-basename .lab), not a flat references/<name>.wav. Voices
were staged flat with the per-name dirs left empty, so every
reference_id resolved to nothing and Fish fell back to its default
speaker — every dropdown voice produced byte-identical audio (proven:
Abigail == Imogen == no-ref, same text+seed). This was the real "no
accent" root cause, independent of the asset-engine "undefined" select
bug.
Server fix (applied to irv-ml1): populated references/<name>/<name>.wav
+ <name>.lab for all 32 voices; re-test confirms Imogen/Eleanor/
Beatrice/Abigail/no-ref now all distinct.
Durable hardening + record correction:
- playbook: normalize-layout step (flat <name>.wav -> nested dir, cp -u
idempotent, when-gated on count mismatch) + an A/B verify gate that
hard-fails the deploy if two reference_ids yield identical output.
- services.yaml: correct the reference_id resolution doc (dir + .lab,
not flat wav).
- README + persistent-memory: correct the "reference_id-by-name is THE
working path, verified" claim — it was a no-op until this fix; the
prior ECAPA 0.79 result came through the inline base64 path.
Pitch-shifted deepening (rubberband, formant on/off) sounded bad at every
depth (tuba / over-gravelly), so abandoned. Removed Imogen_Contralto from the
dropdown + deleted the staged variants (fish + chatterbox). Plain unmodified
Imogen remains. version 5->6.
vh