diff --git a/presets/models/falcon/model.go b/presets/models/falcon/model.go
index 00dfd0d77..7501dce23 100644
--- a/presets/models/falcon/model.go
+++ b/presets/models/falcon/model.go
@@ -37,10 +37,10 @@ var (
 	PresetFalcon40BInstructModel = PresetFalcon40BModel + "-instruct"
 
 	PresetFalconTagMap = map[string]string{
-		"Falcon7B":          "0.0.2",
-		"Falcon7BInstruct":  "0.0.2",
-		"Falcon40B":         "0.0.2",
-		"Falcon40BInstruct": "0.0.2",
+		"Falcon7B":          "0.0.3",
+		"Falcon7BInstruct":  "0.0.3",
+		"Falcon40B":         "0.0.3",
+		"Falcon40BInstruct": "0.0.3",
 	}
 
 	baseCommandPresetFalcon = "accelerate launch"
diff --git a/presets/models/mistral/model.go b/presets/models/mistral/model.go
index 3f1d79d15..7089eafb6 100644
--- a/presets/models/mistral/model.go
+++ b/presets/models/mistral/model.go
@@ -23,12 +23,12 @@ func init() {
 }
 
 var (
-	PresetMistral7BModel          = "mistral-7b"
-	PresetMistral7BInstructModel  = PresetMistral7BModel + "-instruct"
+	PresetMistral7BModel         = "mistral-7b"
+	PresetMistral7BInstructModel = PresetMistral7BModel + "-instruct"
 
 	PresetMistralTagMap = map[string]string{
-		"Mistral7B":          "0.0.2",
-		"Mistral7BInstruct":  "0.0.2",
+		"Mistral7B":         "0.0.3",
+		"Mistral7BInstruct": "0.0.3",
 	}
 
 	baseCommandPresetMistral = "accelerate launch"
@@ -46,7 +46,7 @@ func (*mistral7b) GetInferenceParameters() *model.PresetInferenceParam {
 	return &model.PresetInferenceParam{
 		ModelFamilyName:           "Mistral",
 		ImageAccessMode:           string(kaitov1alpha1.ModelImageAccessModePublic),
-		DiskStorageRequirement:    "50Gi",
+		DiskStorageRequirement:    "100Gi",
 		GPUCountRequirement:       "1",
 		TotalGPUMemoryRequirement: "14Gi",
 		PerGPUMemoryRequirement:   "0Gi", // We run Mistral using native vertical model parallel, no per GPU memory requirement.
@@ -70,7 +70,7 @@ func (*mistral7bInst) GetInferenceParameters() *model.PresetInferenceParam {
 	return &model.PresetInferenceParam{
 		ModelFamilyName:           "Mistral",
 		ImageAccessMode:           string(kaitov1alpha1.ModelImageAccessModePublic),
-		DiskStorageRequirement:    "50Gi",
+		DiskStorageRequirement:    "100Gi",
 		GPUCountRequirement:       "1",
 		TotalGPUMemoryRequirement: "16Gi",
 		PerGPUMemoryRequirement:   "0Gi", // We run mistral using native vertical model parallel, no per GPU memory requirement.
diff --git a/presets/models/phi/model.go b/presets/models/phi/model.go
index e819256b6..2e54dce38 100644
--- a/presets/models/phi/model.go
+++ b/presets/models/phi/model.go
@@ -19,10 +19,10 @@ func init() {
 }
 
 var (
-	PresetPhi2Model          = "phi-2"
+	PresetPhi2Model = "phi-2"
 
 	PresetPhiTagMap = map[string]string{
-		"Phi2":          "0.0.1",
+		"Phi2": "0.0.2",
 	}
 
 	baseCommandPresetPhi = "accelerate launch"
@@ -40,7 +40,7 @@ func (*phi2) GetInferenceParameters() *model.PresetInferenceParam {
 	return &model.PresetInferenceParam{
 		ModelFamilyName:           "Phi",
 		ImageAccessMode:           string(kaitov1alpha1.ModelImageAccessModePublic),
-		DiskStorageRequirement:    "30Gi",
+		DiskStorageRequirement:    "50Gi",
 		GPUCountRequirement:       "1",
 		TotalGPUMemoryRequirement: "12Gi",
 		PerGPUMemoryRequirement:   "0Gi", // We run Phi using native vertical model parallel, no per GPU memory requirement.