The startup timeout runs from when a node's download last moved, so a
model of hundreds of gigabytes on a slow mirror is not failed while it
arrives. ctl model serve --env sets engine variables (NCCL_DEBUG=INFO)
on every node, and a layout given by hand keeps expert parallelism for a
model of experts. docs/large-models.md: how a model is split, what each
way needs of a card, and a checklist for bringing one up on a fabric.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A mixture-of-experts model on more than one card is served with vLLM's
--enable-expert-parallel; config.json's experts are read by any of their
names, and a model that also reads pictures is sized by its text_config.
Tensor parallelism past the KV heads is allowed when it is a multiple of
them (Qwen3-235B, four KV heads, runs on eight cards), which the plan
had refused. Several nodes are offered in any number, not only powers of
two. Each way says what it needs of a card, part by part, and of all its
cards; the serve form checks a context longer than planned still fits.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Found on a fresh HA install with a judge machine: the leader saw the
policy's judge missing while init was still fetching it, served one
itself, and its entry replaced init's, so a second judge ran off the
judge machine. The keeper now waits ten minutes after it starts, and a
serve job it starts gives way to a model added meanwhile (AddNew).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A judge a policy names and the catalog lacked was served with 3 GB, no
cores and llama.cpp's prompt cache; on a busy node its judgments missed
their budget and fail: closed refused the requests. It now gets init's
2 GB and cores, and every judge modelserve serves runs with
--cache-ram 0.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
CI run 394, on a commit that changed only BACKLOG.md, failed with a data
race in TestAVisionModelIsServedWithItsProjectorBesideIt. The serve job's
goroutine wrote j.Request without the lock (since efffdb6, which sized
embedding slots), while Get's snapshot read the job under it. Every other
write to a running job holds the lock. It did not reproduce in 30 runs
here; the race detector's trace names this write and that read.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Apertus 70B with a 65536-token pool, and then 8B with 131072, on a 24 GB
card loaded their weights and failed to make their context on every start,
the 8B taking the card from the 70B each time only to fail. ctl model serve
now shrinks a pool the card cannot hold beside the weights (KV cache from
the GGUF's shape, 90% of the card, a gigabyte for the engine), saying so,
and refuses weights that do not fit at all; the portal's serve form refuses
such a pool and says the largest that fits.
The mirror page removed the 70B's weights while the catalog still had the
model, whose replicas went on loading the node's copy with nothing said.
Removing weights a catalog model loads is refused, naming the model.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
ctl model serve served one as a chat model: no --embedding, four slots, a
512-token batch, embeddings unset, and a chat smoke test it could not
pass. A GGUF that pools (<arch>.pooling_type), or --embeddings, gets
--embedding, 32 slots of 1024 tokens unless sized otherwise, a batch that
holds a whole passage, the embeddings capability, and one vector asked for
as its smoke test (canaries and updates too). An import of a hundred
thousand chunks had gone at the pace of four slots with the card a third
busy.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Llama Guard 3 1B could not tell ordinary coding requests from harmful ones:
on the PoC it scored "continue" 0.51, "continue the code above" 0.89 and
"how do I kill all child processes in bash" 0.92 unsafe, against 0.97 to
1.00 for harmful requests, at Q4_K_M and Q8_0 alike, with its own template.
A site had "continue" refused. Qwen3Guard-Gen 0.6B gave the ordinary ones
0.15 at the most and the harmful ones 1.00, in 0.8 GB on CPUs.
- format qwen3guard: its prompt written out as its template writes it, for
requests and answers, and its Safe / Unsafe / Controversial verdict read;
controversial counts as the policy's judge.controversial says, safe by
default.
- judge.models: judges asked in turn, the first that answers deciding; a
failure is recorded against the judge that failed. guard judge use sets
the order, replacing categories none of the judges names.
- known judges (Qwen3Guard 0.6B/4B/8B, Llama Guard 1B/8B): guard judge
enable serves any of them, --gpu the 4B on the roomiest GPU class as a
standard model; the leader fetches and serves one a policy names that the
catalog lacks.
- server init installs qwen3guard-0.6b with the baseline's categories.
- a judge reads a long text by its start and end, which fit its context,
rather than overflowing it and refusing the request under fail: closed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A --context-length 32768 --ctx-pool 32768 model got llama.cpp's default four
slots of 8192, and refused what the gateway admitted; --parallel is now
what the pool holds of the context. max_output was 1024 for every model,
which cut an editor's agent off mid-file; it is now a quarter of the
context, from 1024 to 8192. --tools writes capabilities.tools, which the
Continue configuration now reads, as it reads max_output.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Conflicts: docs.go (manual, OpenAPI and suites all kept), modelserve Job
(trace and eval both kept). The eval routes are added to the OpenAPI route
table, update and promote documented with their optional body, and the
document regenerated.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Conflicts: config imports and docgen sources (both kept); agent.pb.go
regenerated with buf from the merged proto (MIG devices and trace context).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A suite is a set of cases in YAML or JSON lines, each scored by exact,
contains, regex, a number within a tolerance, a JSON Schema subset, or a
judge model with a rubric; weighed into one score against a threshold.
internal/bench/eval runs one on the control plane through Router.Replay, as
the gateway's own requests, and keeps every run and its cases in the store
(migration 0028: eval_runs, eval_cases).
- llmgw ctl eval MODEL --suite FILE|NAME, --history, --run, --suites; exits
non-zero below the score. Suites by name from control.eval_dir, or the
two built in from deploy/eval (arithmetic, json-format).
- model update and promote take --eval SUITE --min-score N, or a model's
standing gate (ctl model gate, eval: in the catalog): a promote asks the
canary first and leaves it below the score; an update tries the new
weights as a canary given no share, then promotes or removes it.
- The console's model menu: ask a suite with its progress, its runs, its
gate.
- The audit middleware no longer hands a handler a body cut off at 64 KiB.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A reconcile pass that does something is a trace, with placement (where it
went, or every candidate turned down), scale-up/down, drain, stop, eviction,
preemption, reaping and restarts beneath it. A replica's start is a span open
from placement until it serves; the start command carries its context in a
new ControlMessage.trace field, and the agent continues the trace through
the weights fetch, engine launch, rendezvous and readiness wait. Stops,
drains, adoption and engine installs are traced on the node, model jobs are
a trace with a span per step linked to the replica's start, and agent.yaml
takes the same tracing block as server.yaml. Off, every span is the no-op one
and commands carry nothing.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
notify.webhooks sends signed JSON or Slack-compatible messages, queued,
retried, de-duplicated hourly; ctl notify test checks each webhook.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The console starts a serve or update job over the admin API and stays,
instead of closing to follow the download. Jobs lists them with their
progress and steps; the sidebar counts those running. ctl model serve and
update take --detach; ctl jobs lists jobs and follows one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Ranking what was free first made every file come out on the CPU on a cluster
whose one card was in use. The plan now ranks by hardware alone, free now
only breaking a tie; a model served onto cards that are busy is added and
waits for them, the job ending as waiting rather than failed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The control plane reads the model's shape from one GGUF header or its
config.json and lists, for every file, the best way it runs here: one card,
several of a node's (by layer for llama.cpp, tensor-parallel for vLLM),
several nodes' over one fabric (vLLM, pipeline-parallel between them), or a
CPU node; what each card holds and whether it is free now. ctl model plan
prints it; model serve takes its file and way unless told, gives a GGUF on
GPUs a GPU profile, and with --fallbacks adds the next best ways. The
console shows it per file and asks for the way and the fallback.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
model serve takes --projector, fetched beside the weights and loaded by
llama.cpp with --mmproj, and --format safetensors with --gpu-class and --tp,
which fetches the weights with their config and tokenizer and runs them on
vLLM. The console offers both, F switching between GGUF and original
weights. The mirror view is one row per version with its files on Enter.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The mirror keeps where it fetched a file beside it, and reports that and the
checksum after it restarts. A model whose files the mirror cannot trace is
found on the registry: a repository holding a file of that name that has the
commit the mirror's directory is named after, with every file at the same
size, the most downloaded first. ctl model origin and the console set one by
hand, checked the same way. Commits compare shortened.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A model served from the registry records its origin: repository, commit and
each file's checksum; one served before is worked out from the mirror's
record of where it fetched the file. Every mirror.update_check (6h) the
control plane asks whether those files changed there. ctl model updates
lists it, ctl model update fetches beside the old files, checks, swaps and
replaces replicas one at a time, ctl model rollback goes back. The console
shows it on Models and offers the same.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A profile's cpu_cores is reserved out of its node's CPUs by placement, as
memory is, and becomes the engine's thread count. Agents report their CPU
count in registration and heartbeats. New CPU models must declare memory;
model views show reserved against measured memory with warnings, and
ctl model resources sets both for a stored model.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Adding a model took four steps and a YAML file — plan it, fetch it into the
mirror, write its catalog entry with a memory figure somebody estimated,
scale it — and whether it would load was found out at the end.
`llmgw ctl model serve OWNER/REPO` does it as one job in the control plane,
where the registry, the mirror, the catalog and the fleet already are:
- the registry: the GGUF file set matching --quant (Q4_K_M by default) or
--file, leaving out vision projectors and importance matrices, and naming
what the repository does have when nothing matches; gated repositories are
refused;
- the mirror: fetched at the repository's revision, with progress, and each
file's sha256 compared with what the registry published — the mirror is
the one place a corrupted download is caught once rather than loaded on
every node;
- the check: whether the nodes' llama.cpp loads the architecture, and the
memory it needs from its own header;
- the catalog, then a wait for a replica that stops early, with the reason,
when the model is parked or fits on no node;
- a smoke test: one short request to the replica, with its engine key, and
the answer and tokens per second.
ctl starts the job and prints each step as it changes; a download of hours
does not depend on the terminal. --on-demand adds the model without starting
a replica. Jobs are kept in memory until the control plane restarts.
The catalog name a repository suggests moved into the registry package, so
the mirror page and this agree on it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
