mirror of
https://github.com/mudler/LocalAI.git
synced 2026-09-12 22:33:54 -04:00
* feat(modelartifacts): support bounded parallel Hugging Face file downloads Closes #11114. Snapshot materialization fetched every file through the sequential executor in DownloadFilesWithContext, so a repository split into many shards spent most of its wall clock in per-file request latency rather than moving bytes. Add DownloadFilesWithConcurrency, an errgroup with SetLimit, and keep DownloadFilesWithContext as a wrapper that passes a limit of 1. That leaves the two non-artifact callers (core/gallery and the model config loader) on exactly the path they had: tasks still run in slice order, and the first failure still returns before any later task starts. Only whole files run in parallel. A single file is never split, so the .partial resume machinery and the per-file SHA check in downloadTaskWithRetry are untouched. Two details the parallel path forced: - completedBytes becomes an atomic.Int64. Several AfterDownload hooks add to it while other files' progress callbacks read it; without this the race detector reports three races on the new specs. - The caller's status callback is serialized. The sequential path gave it an implicit guarantee of never being entered twice at once, and it belongs to the caller, so the executor keeps that promise rather than pushing locking onto every caller. AfterDownload is deliberately not serialized -- it does the verify-and-promote work that parallelism exists to overlap. Manifest order needed no work: each hook already writes its own manifest.Files slot by snapshot index, so entries stay in snapshot order whatever the completion order. A spec now pins that. The default is 1, unchanged behaviour. A shared models volume is often the bottleneck rather than the link, so raising it is a deployment decision; --artifact-download-concurrency and LOCALAI_ARTIFACT_DOWNLOAD_CONCURRENCY expose it on both `run` and `models install`. Not done here, per the issue: no chunk-level parallelism within a single file, and no throughput measurements across concurrency 1/2/4/8 -- that needs a representative sharded repo and a real link. Assisted-by: Claude:claude-opus-5 go-test gofmt Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com> * feat(modelartifacts): expose download concurrency in settings Follow-up to review feedback on #11162: - The CLI flag and docs no longer describe the limit as Hugging Face specific. It applies to any artifact source, as @mudler pointed out. - artifact_download_concurrency is now a persisted runtime setting and is editable from the WebUI, so it can be changed without a restart. The manager's limit becomes an atomic.Int64 behind SetDownloadConcurrency, because a live runtime setting can be updated while a materialization is already in flight. Injected materializers stay compatible through an optional setter interface, so a manager that does not implement it is simply left alone. Verified before taking this on: go build, go vet and go test -race all pass for pkg/modelartifacts, pkg/downloader and core/config. The React UI builds with vite, artifact_download_concurrency is present in the built Settings chunk, and eslint reports the same 8 pre-existing warnings on Settings.jsx as it does without the change. Implementation contributed by localai-org-maint-bot on the review thread; reviewed, verified and signed off by me. Assisted-by: Codex:gpt-5 Assisted-by: Claude:claude-opus-5 go-test vite eslint Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com> --------- Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com>
107 lines
4.8 KiB
Go
107 lines
4.8 KiB
Go
package config
|
|
|
|
import (
|
|
"encoding/json"
|
|
|
|
"github.com/mudler/LocalAI/pkg/vrambudget"
|
|
"github.com/mudler/LocalAI/pkg/xsysinfo"
|
|
)
|
|
|
|
// DefaultGalleriesJSON / DefaultBackendGalleriesJSON are the gallery lists
|
|
// an option-less `local-ai run` gets. cmd/local-ai feeds them to kong as
|
|
// the "${galleries}" / "${backends}" vars, and DefaultRuntimeBaseline uses
|
|
// them to tell "kong default" apart from "user-configured" at settings-load
|
|
// time - they must stay the single source for both.
|
|
//
|
|
// The primary is served by index-server (github.com/localai-org/index-server),
|
|
// a caching mirror of the files below. The GitHub URI stays as a mirror so an
|
|
// install still resolves its gallery unchanged whenever the primary is
|
|
// unreachable - see the fallback chain in core/gallery/gallery_mirrors.go.
|
|
const DefaultGalleriesJSON = `[{"name":"localai", "url":"https://index.localai.io/models", "mirrors":["github:mudler/LocalAI/gallery/index.yaml@master"]}]`
|
|
const DefaultBackendGalleriesJSON = `[{"name":"localai", "url":"https://index.localai.io/backends", "mirrors":["github:mudler/LocalAI/backend/index.yaml@master"]}]`
|
|
|
|
func mustGalleries(jsonList string) []Gallery {
|
|
var g []Gallery
|
|
if err := json.Unmarshal([]byte(jsonList), &g); err != nil {
|
|
// Both inputs are compile-time constants above; failing loudly at
|
|
// init beats silently mis-detecting every gallery as env-set.
|
|
panic("invalid built-in gallery JSON: " + err.Error())
|
|
}
|
|
return g
|
|
}
|
|
|
|
// DefaultRuntimeBaseline is the ApplicationConfig an option-less `local-ai
|
|
// run` produces: NewApplicationConfig plus the flag defaults kong injects
|
|
// even when the user passes nothing (see the default:"..." tags in
|
|
// core/cli/run.go). ApplyRuntimeSettingsAtStartup compares the live config
|
|
// against it to decide whether env/CLI claimed a field (env > file).
|
|
func DefaultRuntimeBaseline() *ApplicationConfig {
|
|
o := NewApplicationConfig()
|
|
// WithDebug(log level) is always applied by run.go; the default level
|
|
// is not debug, so an option-less run boots with Debug=false even
|
|
// though the NewApplicationConfig literal says true.
|
|
o.Debug = false
|
|
o.Galleries = mustGalleries(DefaultGalleriesJSON)
|
|
o.BackendGalleries = mustGalleries(DefaultBackendGalleriesJSON)
|
|
o.AutoloadGalleries = true
|
|
o.AutoloadBackendGalleries = true
|
|
// core/cli/run.go injects WithMemoryReclaimer(enabled, threshold)
|
|
// unconditionally, so the kong threshold default (0.95) reaches the
|
|
// config even when the reclaimer flag is off - this overlay must match
|
|
// that contract or a UI-persisted threshold looks env-set and is
|
|
// skipped at boot.
|
|
o.MemoryReclaimerThreshold = 0.95
|
|
return o
|
|
}
|
|
|
|
// ApplyRuntimeSettingsAtStartup merges persisted settings into o with
|
|
// env-over-file precedence: a field is taken from the file only when the
|
|
// live value still equals the option-less baseline (env/CLI did not touch
|
|
// it) or when the file is that field's only source (fileAuthoritative).
|
|
// Used at boot (application.New) and by the runtime_settings.json file
|
|
// watcher, so a manual file edit behaves exactly like a boot-time load.
|
|
//
|
|
// Known limitation (accepted): an env var explicitly set to its default
|
|
// value is indistinguishable from "not set", so the file wins there; and a
|
|
// field previously changed via the API looks env-set to the watcher, so a
|
|
// manual file edit of that field lands on the next restart.
|
|
func (o *ApplicationConfig) ApplyRuntimeSettingsAtStartup(settings *RuntimeSettings) {
|
|
if settings == nil {
|
|
return
|
|
}
|
|
baseline := DefaultRuntimeBaseline()
|
|
for _, f := range runtimeSettingsFields {
|
|
if f.snapshotOnly || !f.isSet(settings) {
|
|
continue
|
|
}
|
|
if !f.fileAuthoritative && f.envSet(o, baseline) {
|
|
continue
|
|
}
|
|
f.apply(o, settings)
|
|
}
|
|
// Startup invariant, mirroring ApplyRuntimeSettings and the gating in
|
|
// startWatchdog: enabled idle/busy checks or the memory reclaimer imply
|
|
// the watchdog master flag. Never forced off here - an explicit
|
|
// watchdog_enabled=false row already applied above when unclaimed.
|
|
if o.WatchDogIdle || o.WatchDogBusy || o.MemoryReclaimerEnabled {
|
|
o.WatchDog = true
|
|
}
|
|
// VRAM budget post-processing, mirroring ApplyRuntimeSettings: at boot
|
|
// only run.go's env path installs the process-wide cap, so a
|
|
// file-persisted budget applied by the loop above would set
|
|
// o.VRAMBudget without ever capping allocations. When env set the
|
|
// budget the loop skipped the file value and this re-installs the env
|
|
// value - idempotent. Fail-open on a malformed persisted value so it
|
|
// cannot wedge startup.
|
|
if settings.VRAMBudget != nil {
|
|
if b, err := vrambudget.Parse(o.VRAMBudget); err == nil {
|
|
xsysinfo.SetDefaultVRAMBudget(b)
|
|
}
|
|
}
|
|
if settings.ArtifactDownloadConcurrency != nil {
|
|
if configurable, ok := o.ModelArtifactMaterializer.(interface{ SetDownloadConcurrency(int) }); ok {
|
|
configurable.SetDownloadConcurrency(o.ArtifactDownloadConcurrency)
|
|
}
|
|
}
|
|
}
|