Files
llmgw/internal/cli/model.go
T
Ilker ManapandClaude Opus 5 300e298053 Work out whether a model fits the cluster, and how to split it
llmgw ctl model plan Qwen/Qwen2.5-Coder-32B-Instruct

The arithmetic this replaces is the last thing between having GPUs and
serving a model on them. Choosing tp, gpu_class and a context that will
actually fit means knowing the layer count, the KV head count, what the
weights weigh at a chosen quantisation and how much of a card the runtime
keeps — and getting it wrong is not an error message, it is an
out-of-memory twenty minutes into a load. I computed those by hand for
every model on this cluster.

Both halves are read rather than retyped: the architecture from the model's
own config.json and parameter count, the hardware from the running fleet.
Attention heads in place of KV heads is the classic slip and is worth a
factor of five on a model with grouped-query attention, so it is not a
number anyone should be entering.

Every candidate is listed, not one answer. With cards to spare you want the
full-precision fit; with one card free you want to know a quantised one
exists. Choosing between those is not the planner's job.

The tests are calibrated against files actually downloaded here — llama.cpp
reported 14.62 GiB and 8.37 GiB for the 14B at q8_0 and q4_K_M, and the
model reproduces both — and they caught two errors in my own arithmetic.
The overhead was charged per rank from the whole model's weights rather
than each rank's share, and it reserved a tenth of a card that
usableFraction had already reserved. Between them, Llama-3.3-70B at bf16
across two 80 GB cards came out as not fitting, which is a configuration
people run every day. Slack reserved in two places, each unaware of the
other, reads as caution and behaves as a bug.

Pools are grouped by class and fabric together, because a replica may never
span two fabrics: eight free cards split across two of them are not eight
cards for a model that needs eight, and merging them would plan something
that starts, passes its health checks and hangs on its first all-reduce.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-17 07:46:53 +02:00

111 lines
3.8 KiB
Go

package cli
import (
"encoding/json"
"fmt"
"os/user"
"github.com/knadh/koanf/parsers/yaml"
"github.com/knadh/koanf/providers/file"
"github.com/knadh/koanf/v2"
"github.com/spf13/cobra"
"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
)
// newModelCmd adds and removes models without a restart.
//
// Adding a model used to mean editing a file on the control plane and
// restarting it: a settling window, adoption churn, and on one occasion two
// minutes of outage. For a fixed production set that is a reasonable trade; for
// a platform whose promise is that people can try things, it is the difference
// between trying one and not bothering.
func newModelCmd(client func() *ctlClient) *cobra.Command {
c := &cobra.Command{
Use: "model",
Short: "Add or remove models while the cluster is running",
}
c.AddCommand(newModelAddCmd(client), newModelRemoveCmd(client), newModelPlanCmd(client))
return c
}
func newModelAddCmd(client func() *ctlClient) *cobra.Command {
var path string
c := &cobra.Command{
Use: "add -f <file.yaml>",
Short: "Add a model from a definition file",
Long: "The definition is the same YAML a model takes in server.yaml, so an\n" +
"entry can be moved between the two by copying it.\n\n" +
"The catalog is rebuilt and validated before anything is written, so a\n" +
"definition that collides with an existing name, or names a fallback that\n" +
"does not exist, is refused rather than stored and then ignored on every\n" +
"start. Once accepted it takes effect on the reconciler's next pass — no\n" +
"restart — and survives one.",
Args: cobra.NoArgs,
RunE: func(cmd *cobra.Command, _ []string) error {
if path == "" {
return fmt.Errorf("--file is required: the model definition, as YAML")
}
// Read by koanf, the same parser and the same tags that read
// server.yaml — so an entry can be moved between the two by copying
// it, and a field that works in one cannot silently mean nothing in
// the other.
k := koanf.New(".")
if err := k.Load(file.Provider(path), yaml.Parser()); err != nil {
return fmt.Errorf("%s: %w", path, err)
}
var m catalog.Model
if err := k.UnmarshalWithConf("", &m, koanf.UnmarshalConf{Tag: "koanf"}); err != nil {
return fmt.Errorf("%s: %w", path, err)
}
if m.Name == "" {
return fmt.Errorf("%s: the definition has no name", path)
}
body, err := json.Marshal(m)
if err != nil {
return err
}
var res map[string]string
if err := client().post(cmd.Context(), "/api/v1/models", body, &res); err != nil {
return err
}
fmt.Fprintf(cmd.OutOrStdout(),
"%s added. It will be placed on the next reconcile pass; watch it with:\n"+
" llmgw ctl models\n", m.Name)
return nil
},
}
c.Flags().StringVarP(&path, "file", "f", "", "YAML file holding the model definition")
return c
}
func newModelRemoveCmd(client func() *ctlClient) *cobra.Command {
return &cobra.Command{
Use: "remove <name>",
Short: "Remove a model that was added at runtime",
Long: "Only models added at runtime. One declared in server.yaml would come\n" +
"back on the next restart, so removing it here would leave the cluster\n" +
"disagreeing with its own configuration until then.\n\n" +
"Its replicas are stopped on the next pass, after the endpoint has been\n" +
"withdrawn, so traffic stops before the engine does.",
Args: cobra.ExactArgs(1),
RunE: func(cmd *cobra.Command, args []string) error {
var res map[string]string
if err := client().delete(cmd.Context(), "/api/v1/models/"+args[0], &res); err != nil {
return err
}
fmt.Fprintf(cmd.OutOrStdout(), "%s removed.\n", args[0])
return nil
},
}
}
// actor names who made a change, for the audit trail on a stored model.
func actor() string {
if u, err := user.Current(); err == nil && u.Username != "" {
return u.Username
}
return "unknown"
}