inference: Fix ignoring parse errors for unknown models

We ignore parse errors for models that gguf-parser-go can't parse yet,
for now. This regressed in the pre-pull memeory estimation PR.

Signed-off-by: Piotr Stankiewicz <piotr.stankiewicz@docker.com>

pkg/inference/backend.go

@@ -2,7 +2,6 @@ package inference
 
 import (
 	"context"
-	"errors"
 	"net/http"
 )
 
@@ -18,9 +17,13 @@ const (
 	BackendModeEmbedding
 )
 
-var (
-	ErrGGUFParse = errors.New("failed to parse GGUF file")
-)
+type ErrGGUFParse struct {
+	Err error
+}
+
+func (e *ErrGGUFParse) Error() string {
+	return "failed to parse GGUF: " + e.Err.Error()
+}
 
 // String implements Stringer.String for BackendMode.
 func (m BackendMode) String() string {

pkg/inference/backends/llamacpp/llamacpp.go

@@ -235,12 +235,12 @@ func (l *llamaCpp) GetRequiredMemoryForModel(ctx context.Context, model string,
 	if inStore {
 		mdlGguf, mdlConfig, err = l.parseLocalModel(model)
 		if err != nil {
-			return nil, fmt.Errorf("parsing local gguf: %w", err)
+			return nil, &inference.ErrGGUFParse{Err: err}
 		}
 	} else {
 		mdlGguf, mdlConfig, err = l.parseRemoteModel(ctx, model)
 		if err != nil {
-			return nil, fmt.Errorf("parsing remote model: %w", err)
+			return nil, &inference.ErrGGUFParse{Err: err}
 		}
 	}
 

pkg/inference/scheduling/loader.go

@@ -421,11 +421,12 @@ func (l *loader) load(ctx context.Context, backendName, modelID, modelRef string
 		runnerConfig = &rc
 	}
 	memory, err := backend.GetRequiredMemoryForModel(ctx, modelID, runnerConfig)
-	if errors.Is(err, inference.ErrGGUFParse) {
+	var parseErr *inference.ErrGGUFParse
+	if errors.As(err, &parseErr) {
 		// TODO(p1-0tr): For now override memory checks in case model can't be parsed
 		// e.g. model is too new for gguf-parser-go to know. We should provide a cleaner
 		// way to bypass these checks.
-		l.log.Warnf("Could not parse model(%s), memory checks will be ignored for it.", modelID)
+		l.log.Warnf("Could not parse model(%s), memory checks will be ignored for it. Error: %s", modelID, parseErr)
 		memory = &inference.RequiredMemory{
 			RAM:  0,
 			VRAM: 0,