diff --git a/pkg/inference/backend.go b/pkg/inference/backend.go
index 944ec12..3d857f8 100644
--- a/pkg/inference/backend.go
+++ b/pkg/inference/backend.go
@@ -2,6 +2,7 @@ package inference
 
 import (
 	"context"
+	"errors"
 	"net/http"
 )
 
@@ -17,6 +18,10 @@ const (
 	BackendModeEmbedding
 )
 
+var (
+	ErrGGUFParse = errors.New("failed to parse GGUF file")
+)
+
 // String implements Stringer.String for BackendMode.
 func (m BackendMode) String() string {
 	switch m {
diff --git a/pkg/inference/backends/llamacpp/llamacpp.go b/pkg/inference/backends/llamacpp/llamacpp.go
index 09de11f..f53b26a 100644
--- a/pkg/inference/backends/llamacpp/llamacpp.go
+++ b/pkg/inference/backends/llamacpp/llamacpp.go
@@ -234,7 +234,8 @@ func (l *llamaCpp) GetRequiredMemoryForModel(model string, config *inference.Bac
 	}
 	mdlGguf, err := parser.ParseGGUFFile(mdlPath)
 	if err != nil {
-		return nil, fmt.Errorf("parsing gguf(%s): %w", mdlPath, err)
+		l.log.Warnf("Failed to parse gguf(%s): %s", mdlPath, err)
+		return nil, inference.ErrGGUFParse
 	}
 	mdlConfig, err := mdl.Config()
 	if err != nil {
diff --git a/pkg/inference/scheduling/loader.go b/pkg/inference/scheduling/loader.go
index ec7e1f5..ebbdd33 100644
--- a/pkg/inference/scheduling/loader.go
+++ b/pkg/inference/scheduling/loader.go
@@ -421,7 +421,16 @@ func (l *loader) load(ctx context.Context, backendName, modelID, modelRef string
 		runnerConfig = &rc
 	}
 	memory, err := backend.GetRequiredMemoryForModel(modelID, runnerConfig)
-	if err != nil {
+	if errors.Is(err, inference.ErrGGUFParse) {
+		// 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)
+		memory = &inference.RequiredMemory{
+			RAM:  0,
+			VRAM: 0,
+		}
+	} else if err != nil {
 		return nil, err
 	}
 	l.log.Infof("Loading %s, which will require %dMB RAM and %dMB VRAM", modelID, memory.RAM/1024/1024, memory.VRAM/1024/1024)