Merge remote-tracking branch 'upstream/main'

Author: dchourasia

tests/kfto/core/config_mistral_7b_v03.json

@@ -0,0 +1,19 @@
+{
+    "model_name_or_path": "mistralai/Mistral-7B-v0.3",
+    "training_data_path": "/mnt/scratch/dataset/alpaca_data.json",
+    "output_dir": "/mnt/output/model",
+    "save_model_dir": "/mnt/output/model",
+    "num_train_epochs": 1.0,
+    "per_device_train_batch_size": 1,
+    "per_device_eval_batch_size": 4,
+    "gradient_accumulation_steps": 1,
+    "save_strategy": "no",
+    "learning_rate": 1e-5,
+    "weight_decay": 0.0,
+    "lr_scheduler_type": "cosine",
+    "include_tokens_per_second": true,
+    "response_template": "\n### Response:",
+    "dataset_text_field": "output",
+    "use_flash_attn": false
+}
+

tests/kfto/core/config_mixtral_8x7b_instruct_v01.json renamed to tests/kfto/core/config_mixtral_8x7b_instruct_v01_lora.json

@@ -14,6 +14,7 @@
     "include_tokens_per_second": true,
     "response_template": "\n### Response:",
     "dataset_text_field": "output",
-    "use_flash_attn": false
+    "use_flash_attn": false,
+    "peft_method": "lora"
 }
 

tests/kfto/core/kfto_kueue_sft_GPU_test.go

@@ -67,12 +67,16 @@ func TestMultiGpuPytorchjobMetaLlama370bInstructLoRa(t *testing.T) {
 	runMultiGpuPytorchjob(t, "config_meta_llama3_70b_instruct_lora.json")
 }
 
+func TestMultiGpuPytorchjobMistral7bv03(t *testing.T) {
+	runMultiGpuPytorchjob(t, "config_mistral_7b_v03.json")
+}
+
 func TestMultiGpuPytorchjobMixtral8x7bv01(t *testing.T) {
 	runMultiGpuPytorchjob(t, "config_mixtral_8x7b_v01.json")
 }
 
-func TestMultiGpuPytorchjobMixtral8x7bInstructv01(t *testing.T) {
-	runMultiGpuPytorchjob(t, "config_mixtral_8x7b_instruct_v01.json")
+func TestMultiGpuPytorchjobMixtral8x7bInstructv01LoRa(t *testing.T) {
+	runMultiGpuPytorchjob(t, "config_mixtral_8x7b_instruct_v01_lora.json")
 }
 
 func TestMultiGpuPytorchjobMerlinite7b(t *testing.T) {

tests/kfto/upgrade/kfto_kueue_sft_upgrade_training_test.go

@@ -154,11 +154,12 @@ func createPyTorchJob(test Test, namespace, localQueueName string, config corev1
 									ImagePullPolicy: corev1.PullIfNotPresent,
 									VolumeMounts: []corev1.VolumeMount{
 										{
-											Name:      "model-volume",
-											MountPath: "/tmp/model",
+											Name:      "tmp-volume",
+											MountPath: "/tmp",
 										},
 									},
-									Command: []string{"cp", "-r", "/models/bloom-560m", "/tmp/model"},
+									Command: []string{"/bin/sh", "-c"},
+									Args:    []string{"mkdir /tmp/model; cp -r /models/bloom-560m /tmp/model"},
 								},
 							},
 							Containers: []corev1.Container{
@@ -171,21 +172,29 @@ func createPyTorchJob(test Test, namespace, localQueueName string, config corev1
 											Name:  "SFT_TRAINER_CONFIG_JSON_PATH",
 											Value: "/etc/config/config.json",
 										},
+										{
+											Name:  "HF_HOME",
+											Value: "/tmp/huggingface",
+										},
 									},
 									VolumeMounts: []corev1.VolumeMount{
 										{
 											Name:      "config-volume",
 											MountPath: "/etc/config",
 										},
 										{
-											Name:      "model-volume",
-											MountPath: "/tmp/model",
+											Name:      "tmp-volume",
+											MountPath: "/tmp",
 										},
 									},
 									Resources: corev1.ResourceRequirements{
 										Requests: corev1.ResourceList{
 											corev1.ResourceCPU:    resource.MustParse("2"),
-											corev1.ResourceMemory: resource.MustParse("5Gi"),
+											corev1.ResourceMemory: resource.MustParse("7Gi"),
+										},
+										Limits: corev1.ResourceList{
+											corev1.ResourceCPU:    resource.MustParse("2"),
+											corev1.ResourceMemory: resource.MustParse("7Gi"),
 										},
 									},
 								},
@@ -212,7 +221,7 @@ func createPyTorchJob(test Test, namespace, localQueueName string, config corev1
 									},
 								},
 								{
-									Name: "model-volume",
+									Name: "tmp-volume",
 									VolumeSource: corev1.VolumeSource{
 										EmptyDir: &corev1.EmptyDirVolumeSource{},
 									},