Files
LocalAI/core/backend/rerank.go
mudler's LocalAI [bot] 1cd7d63c7b fix(distributed): reject wrong-model requests on the remaining modalities (#10990)
#10970 gave the four PredictOptions RPCs a model-identity check so a
backend reached through a stale distributed route rejects the request
instead of answering from whatever model it holds (#10952). Every other
modality shares that exposure: the route is cached by host:port, a worker
can recycle a stopped backend's port for another model's backend, and a
liveness-only probe cannot tell a stale row from a valid one.

Extends the same mechanism to the 21 remaining request messages that reach
a backend through the router, using the pattern #10970 established rather
than a parallel one:

- proto: ModelIdentity on each modality request message.
- controller: populated from ModelConfig.Model at the call site that also
  builds ModelOptions, so load-time and request-time values are equal by
  construction.
- backends: one generic guard in pkg/grpc/server.go (27 Go backends), the
  method set in backend/python/common (36 Python backends), llama-cpp
  (AudioTranscription/Stream, Rerank, Score) and privacy-filter
  (TokenClassify).
- reconcile already drops the stale row on IsModelMismatch; no change.

TTSRequest and SoundGenerationRequest get a SEPARATE ModelIdentity field
rather than reusing their existing `model`: FileStagingClient rewrites
`model` to a worker-local path, so comparing it would reject valid
requests in exactly the configuration this guards.

AudioEncode/AudioDecode are deliberately left unguarded: the opus codec
backend is loaded from a literal rather than a ModelConfig, so no value
carries the equality guarantee the comparison depends on. The four
bidirectional stream RPCs are out of scope; they bypass reconcile.

Empty means skip on both sides, so an old controller, an old backend, and
the bare request structs in tests/e2e-backends all keep working.


Assisted-by: Claude Code:claude-opus-4-8 [Read] [Edit] [Bash]

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
2026-07-20 21:58:19 +02:00

111 lines
3.5 KiB
Go

package backend
import (
"context"
"fmt"
"time"
"github.com/mudler/LocalAI/core/config"
"github.com/mudler/LocalAI/core/trace"
"github.com/mudler/LocalAI/pkg/grpc/proto"
model "github.com/mudler/LocalAI/pkg/model"
)
// RerankResult is the per-document score returned to consumers,
// narrowed from proto.RerankResult so callers don't need to depend on
// the proto package.
type RerankResult struct {
Index int
RelevanceScore float32
}
// Reranker scores a list of candidate documents against a query.
// Returns one RerankResult per input document (no top-N truncation -
// callers that need it can sort and slice).
type Reranker interface {
Rerank(ctx context.Context, query string, documents []string) ([]RerankResult, error)
}
// NewReranker binds (loader, modelConfig, appConfig) into a Reranker.
func NewReranker(loader *model.ModelLoader, modelConfig config.ModelConfig, appConfig *config.ApplicationConfig) Reranker {
return &modelReranker{loader: loader, modelConfig: modelConfig, appConfig: appConfig}
}
type modelReranker struct {
loader *model.ModelLoader
modelConfig config.ModelConfig
appConfig *config.ApplicationConfig
}
func (r *modelReranker) Rerank(ctx context.Context, query string, documents []string) ([]RerankResult, error) {
req := &proto.RerankRequest{
Query: query,
Documents: documents,
// TopN=0: backend returns scores for every document. Truncating
// here would silently zero out labels the reranker considered
// unlikely, which the router classifier needs.
}
res, err := Rerank(ctx, req, r.loader, r.appConfig, r.modelConfig)
if err != nil {
return nil, err
}
out := make([]RerankResult, 0, len(res.GetResults()))
for _, dr := range res.GetResults() {
out = append(out, RerankResult{Index: int(dr.GetIndex()), RelevanceScore: dr.GetRelevanceScore()})
}
return out, nil
}
func Rerank(ctx context.Context, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, modelConfig config.ModelConfig) (*proto.RerankResult, error) {
// model.WithContext(ctx) overrides the app-context default set in
// ModelOptions so distributed routing decisions reach the request's
// X-LocalAI-Node holder via distributedhdr.Stamp.
opts := ModelOptions(modelConfig, appConfig, model.WithContext(ctx))
rerankModel, err := loader.Load(opts...)
if err != nil {
recordModelLoadFailure(appConfig, modelConfig.Name, modelConfig.Backend, err, nil)
return nil, err
}
if rerankModel == nil {
return nil, fmt.Errorf("could not load rerank model")
}
var startTime time.Time
if appConfig.EnableTracing {
trace.InitBackendTracingIfEnabled(appConfig.TracingMaxItems, appConfig.TracingMaxBodyBytes)
startTime = time.Now()
}
// Stamped here, not at the HTTP handler: this is the function that also
// builds ModelOptions from the same config, so the two values are equal by
// construction (#10952).
request.ModelIdentity = modelConfig.Model
res, err := rerankModel.Rerank(ctx, request)
if appConfig.EnableTracing {
errStr := ""
if err != nil {
errStr = err.Error()
}
trace.RecordBackendTrace(trace.BackendTrace{
Timestamp: startTime,
Duration: time.Since(startTime),
Type: trace.BackendTraceRerank,
ModelName: modelConfig.Name,
Backend: modelConfig.Backend,
Summary: trace.TruncateString(request.Query, 200),
Error: errStr,
Data: map[string]any{
"query": request.Query,
"documents_count": len(request.Documents),
"top_n": request.TopN,
},
})
}
return res, err
}