Files
LocalAI/core/application/router_factories.go
Richard Palethorpe 085fc53bbc fix(router): production-ready request router + auto-size batch for embedding/rerank (#10104)
* fix(router): score classifier production-readiness

Conversation trimming runs through the classifier model's chat template
and trims by exact token count, sized to the model's n_batch which is
now scaled to context so long probes can't crash the backend. Missing
chat_message templates are a hard error at router build time. Router-
facing factories (Embedder/Scorer/Reranker/TokenCounter) re-resolve
ModelConfig per call so a model installed post-startup doesn't bind a
stub Backend="" config and silently fall into the loader's auto-
iterate path.

New 'vector_store' backend trace recorded inside localVectorStore on
every Search/Insert — including the backend-load-failure path that
previously vanished into an xlog.Warn — with outcome tagging
(hit/miss/empty_store/backend_load_error/find_error/insert_error/ok).
Companion cleanup drops misleading similarity:0 and input_tokens_count:0
from non-hit and text-mode traces.

Gallery local-store-development aliases to 'local-store' so the master
image satisfies pkg/model.LocalStoreBackend lookups from the embedding
cache.

Misc: llama-cpp TokenizeString reads the correct 'prompt' JSON key
(the original bug); ModelTokenize nil-guard; non-fatal mitm proxy
startup; PII 'route_local' renamed to 'allow' with docs/UI in sync;
model-editor footer no longer eats the edit area on small screens;
several config-editor template/dropdown/section fixes.

Tests: e2e router specs (casual/code-hint + long-conversation trim),
vector_store trace specs, lazy-factory specs, gallery dev-alias
resolution, Playwright trace badge + scroll regression.

Assisted-by: Claude:claude-opus-4-7 [Claude Code]
Signed-off-by: Richard Palethorpe <io@richiejp.com>

* feat(backend): auto-size batch to context for embedding and rerank models

Embedding and rerank models pool over the whole input in a single physical batch (n_ubatch). With batch left at the 512 default, the backend rejects longer inputs with "input is too large to process", silently capping a large-context embedder (e.g. 8k/32k) at 512 tokens. Size n_batch to the context for these single-pass usecases, mirroring the existing FLAG_SCORE behaviour; an explicit batch: still wins.

Extracts EffectiveContextSize/EffectiveBatchSize from grpcModelOpts so the effective decode window has one home for other callers to reuse.

Adds an e2e-aio regression test that embeds a >512-token input. The AIO embedding model is switched to nomic-embed-text-v1.5 (2048 context) because the previous granite model was capped at 512 tokens and could not exercise the larger batch.

Assisted-by: claude-code:claude-opus-4-8 [Claude Code]
Signed-off-by: Richard Palethorpe <io@richiejp.com>

* fix(gallery): raise arch-router scoring output cap via parallel:64

Scoring decodes the whole prompt+candidate in a single llama_decode and
reads one logit row per candidate token. The vendored llama.cpp server
caps causal output rows at n_parallel, so the default of 1 aborts with
GGML_ASSERT(n_outputs_max <= cparams.n_outputs_max) on multi-token route
labels. Set options: [parallel:64] on both arch-router quant entries to
lift the cap; kv_unified (the grpc-server default) keeps the full context
per sequence, so this does not split the KV cache.

Assisted-by: claude-code:claude-opus-4-8 [Claude Code]
Signed-off-by: Richard Palethorpe <io@richiejp.com>

---------

Signed-off-by: Richard Palethorpe <io@richiejp.com>
2026-06-12 16:21:15 +02:00

121 lines
4.1 KiB
Go

package application
import (
"context"
"fmt"
"github.com/mudler/LocalAI/core/backend"
"github.com/mudler/LocalAI/core/config"
)
// adapterConfig resolves a model name to its runtime ModelConfig, or nil when
// unknown. LoadModelConfigFileByNameDefaultOptions never returns nil — for an
// unknown name it returns a defaults-filled stub with an empty Name (the YAML
// `name:` field is required by Validate), which is how we tell the two apart.
func (a *Application) adapterConfig(modelName string) *config.ModelConfig {
cfg, err := a.backendLoader.LoadModelConfigFileByNameDefaultOptions(modelName, a.applicationConfig)
if err != nil || cfg == nil || cfg.Name == "" {
return nil
}
return cfg
}
// ModelConfigLookup is the lookup the router middleware's classifier validator
// uses to confirm classifier_model declares FLAG_SCORE before binding it.
func (a *Application) ModelConfigLookup() func(modelName string) *config.ModelConfig {
return a.adapterConfig
}
// The router-facing factories below (Scorer, Embedder, Reranker, TokenCounter)
// bind a model NAME at construction and re-resolve the CONFIG on every call.
// Capturing the config at construction would bake in whatever state
// adapterConfig saw first — including a stub returned before the YAML reached
// bcl.configs (e.g. /import-model or gallery install racing startup). The
// classifier registry caches factories by router-config fingerprint, so a
// once-stale capture stays stale until the router config is edited.
func (a *Application) Scorer(modelName string) backend.Scorer {
if a.adapterConfig(modelName) == nil {
return nil
}
return &lazyScorer{app: a, modelName: modelName}
}
type lazyScorer struct {
app *Application
modelName string
}
func (l *lazyScorer) Score(ctx context.Context, prompt string, candidates []string) ([]backend.CandidateScore, error) {
cfg := l.app.adapterConfig(l.modelName)
if cfg == nil {
return nil, fmt.Errorf("scorer: model %q no longer available", l.modelName)
}
return backend.NewScorer(l.app.modelLoader, *cfg, l.app.applicationConfig).Score(ctx, prompt, candidates)
}
// TokenCounter returns a func so the middleware's literal field type accepts
// it as a method value without importing core/http/middleware from here.
func (a *Application) TokenCounter(modelName string) func(string) (int, error) {
if a.adapterConfig(modelName) == nil {
return nil
}
return func(text string) (int, error) {
cfg := a.adapterConfig(modelName)
if cfg == nil {
return 0, fmt.Errorf("token counter: model %q no longer available", modelName)
}
resp, err := backend.ModelTokenize(text, a.modelLoader, *cfg, a.applicationConfig)
if err != nil {
return 0, err
}
return len(resp.Tokens), nil
}
}
func (a *Application) Reranker(modelName string) backend.Reranker {
if a.adapterConfig(modelName) == nil {
return nil
}
return &lazyReranker{app: a, modelName: modelName}
}
type lazyReranker struct {
app *Application
modelName string
}
func (l *lazyReranker) Rerank(ctx context.Context, query string, documents []string) ([]backend.RerankResult, error) {
cfg := l.app.adapterConfig(l.modelName)
if cfg == nil {
return nil, fmt.Errorf("reranker: model %q no longer available", l.modelName)
}
return backend.NewReranker(l.app.modelLoader, *cfg, l.app.applicationConfig).Rerank(ctx, query, documents)
}
func (a *Application) Embedder(modelName string) backend.Embedder {
if a.adapterConfig(modelName) == nil {
return nil
}
return &lazyEmbedder{app: a, modelName: modelName}
}
type lazyEmbedder struct {
app *Application
modelName string
}
func (l *lazyEmbedder) Embed(ctx context.Context, text string) ([]float32, error) {
cfg := l.app.adapterConfig(l.modelName)
if cfg == nil {
return nil, fmt.Errorf("embedder: model %q no longer available", l.modelName)
}
return backend.NewEmbedder(l.app.modelLoader, *cfg, l.app.applicationConfig).Embed(ctx, text)
}
// VectorStore takes a store name, not a model name — no adapterConfig, no
// staleness to avoid.
func (a *Application) VectorStore(storeName string) backend.VectorStore {
return backend.NewVectorStore(a.modelLoader, a.applicationConfig, storeName)
}