service-catalog/manifests/helm/nemo/25.4.0/charts/nemo-customizer/values.yaml

# -- Default values for nemollm.
# -- This is a YAML-formatted file.
# -- Declare variables to be passed into your templates.

# -- String to override chart name on resulting objects when deployed.
nameOverride: ""
# -- String to fully override the chart and release name on resulting objects when deployed.
fullnameOverride: ""

# -- NeMo Customizer image that supports training and standalone mode.
# @default -- This object has the following default values for the NeMo Customizer microservice image.
image:
  # -- Registry for the NeMo Customizer image.
  registry: nvcr.io
  # -- Repository for the NeMo Customizer image.
  repository: nvidia/nemo-microservices/customizer
  # -- Image pull policy for the NeMo Customizer image.
  imagePullPolicy: IfNotPresent

# -- Customizer API only image configuration.
# @default -- This object has the following default values for the NeMo Customizer API only image.
apiImage:
  # -- Registry for the NeMo Customizer API image.
  registry: nvcr.io
  # -- Repository for the NeMo Customizer API image.
  repository: nvidia/nemo-microservices/customizer-api
  # -- Image pull policy for the NeMo Customizer API image.
  imagePullPolicy: IfNotPresent

# -- Image pull secrets configuration.
imagePullSecrets:
  - name: nvcrimagepullsecret

# -- Download models to PVC model cache configuration.
# @default -- This object has the following default values for the model downloader.
modelDownloader:
  # -- Security context for the model downloader.
  securityContext:
    fsGroup: 1000
    runAsNonRoot: true
    runAsUser: 1000
    runAsGroup: 1000
  # -- Time to live in seconds after the job finishes.
  ttlSecondsAfterFinished: 7200
  # -- Interval in seconds to poll for model download status.
  pollIntervalSeconds: 15

# -- Secret used for auto hydrating the model cache from NGC for enabled models.
ngcAPISecret: "ngc-api"
# -- Key in the NGC API secret containing the API key.
ngcAPISecretKey: "NGC_API_KEY"

# -- A map of environment variables to inject into the NeMo Customizer app container.
# Example:
#
# `{HOST_IP:
#   valueFrom:
#     fieldRef:
#       fieldPath: status.hostIP
# OTEL_EXPORTER_OTLP_ENDPOINT: "http://$(HOST_IP):4317"}`
env: {}


# -- Tools configuration for downloading and uploading entities to NeMo Data Store.
# @default -- This object has the following default values for the NeMo Data Store tools image.
nemoDataStoreTools:
  # -- Registry for the NeMo Data Store tools image.
  registry: nvcr.io
  # -- Repository for the NeMo Data Store tools image.
  repository: nvidia/nemo-microservices/nds-v2-huggingface-cli
  # -- Tag for the NeMo Data Store tools image.
  tag: ""
  # -- Image pull secret for the NeMo Data Store tools image.
  imagePullSecret: nvcrimagepullsecret

# -- Service configuration.
service:
  # -- Type of Kubernetes service to create.
  type: ClusterIP
  # -- External port for the service.
  port: 8000
  # -- Internal port for the service.
  internalPort: 9009

# -- Number of replicas to deploy.
replicaCount: 1

# -- Service account configuration.
serviceAccount:
  # -- Specifies whether a service account should be created.
  create: true
  # -- The name of the service account to use. If not set and create is true, a name is generated.
  name: ""
  # -- Annotations to add to the service account.
  annotations: {}
  # -- Automatically mount a ServiceAccount's API credentials.
  automountServiceAccountToken: true


# -- Configure the PVC for models mount, where we store the parent/base models.
modelsStorage:
  # -- Enable persistent volume for model storage.
  enabled: true
  # -- Storage class name for the models PVC. Empty string uses the default storage class.
  storageClassName: ""
  # -- Size of the persistent volume.
  size: 1Ti
  # -- Access modes for the persistent volume.
  accessModes:
    - ReadWriteMany

# -- Logging configuration.
logging:
  # -- Log level for the application.
  logLevel: INFO
  # -- Enable logging for health endpoints.
  logHealthEndpoints: false

# -- Configuration for the NeMo Customizer microservice.
# @default -- This object has default values for the following fields.
customizerConfig:
  # -- URL for the NeMo Entity Store microservice.
  entityStoreURL: "http://nemo-gateway.nemo-gateway.svc.cluster.local:8000"
  # -- URL for the NeMo Data Store microservice.
  nemoDataStoreURL: "http://nds-datastore-http.nds-v2.svc.cluster.local:3000"
  # -- URL for the MLflow tracking server.
  mlflowURL: "http://mlflow-tracking.mlflow-system.svc.cluster.local:80"

  # -- Weights and Biases (WandB) Python SDK intialization configuration for logging and monitoring training jobs in WandB.
  wandb:
    # -- The username or team name under which the runs will be logged.
    # -- If not specified, the run will default to a default entity set in the account settings.
    # -- To change the default entity, go to the account settings https://wandb.ai/settings
    # -- and update the "Default location to create new projects" under "Default team".
    # -- Reference: https://docs.wandb.ai/ref/python/init/
    entity: null
    # The name of the project under which this run will be logged
    project: "nvidia-nemo-customizer"

  # -- Network configuration for training jobs on Oracle Kubernetes Engine (OKE) on Oracle Cloud Infrastructure (OCI).
  trainingNetworking:
    - name: NCCL_IB_SL
      value: 0
    - name: NCCL_IB_TC
      value: 41
    - name: NCCL_IB_QPS_PER_CONNECTION
      value: 4
    - name: UCX_TLS
      value: TCP
    - name: UCX_NET_DEVICES
      value: eth0
    - name: HCOLL_ENABLE_MCAST_ALL
      value: 0
    - name: NCCL_IB_GID_INDEX
      value: 3

  # -- Training configuration for customization jobs.
  # @default -- This object has the following default values for the training configuration.
  training:
    # -- Queue name used by the underlying scheduler of NemoTrainingJob.
    # Maps to "resourceGroup" in NemoTrainingJob.
    queue: "default"

    # -- Directory path for training workspace.
    workspace_dir: "/pvc/workspace"

    # -- Training timeout in seconds. If job times out, it will be marked as failed and no checkpoints are saved.
    # training_timeout: 3600

    # -- Time to live in seconds after the training job pod completes.
    # ttl_seconds_after_finished: 3600

    # container_defaults lets you configure the training container similar to a K8s object.
    # Currently, it only supports a subset of the K8s Container Spec.
    # Note it is named _defaults because we plan to allow training specific sections in the future that can override this section.
    # -- Default container configuration for training jobs.
    container_defaults:
      # env holds a list of custom environment variables injected into the training container.
      # However, they cannot override env variables reserved by Customizer. The application validates this at start time.
      # An example use case is configuring OpenTelemetry in the training container.
      # By default, the training container inherits its OpenTelemetry environment values from the openTelemetry section.
      # However, by setting the OpenTelemetry env variables here, the user can override the behavior set in the openTelemetry section.
      # -- Environment variables for the training container.
      # Cannot override env variables reserved by NeMo Customizer.
      env:
        # - name: HOST_IP
        #   valueFrom:
        #     fieldRef:
        #       fieldPath: "status.hostIP"
        # - name: NAMESPACE
        #   valueFrom:
        #     fieldRef:
        #       fieldPath: "metadata.namespace"
        # - name: OTEL_RESOURCE_ATTRIBUTES
        #   value: "deployment.environment=$(NAMESPACE)"
        # - name: OTEL_EXPORTER_OTLP_ENDPOINT
        #   value: "http://$(HOST_IP):4317"
        # - name: OTEL_TRACES_EXPORTER
        #   value: otlp
        # - name: OTEL_METRICS_EXPORTER
        #   value: otlp
        # - name: OTEL_LOGS_EXPORTER
        #   value: none
      imagePullPolicy: IfNotPresent
    # PVC config for NemoTrainingJob, which automatically creates one for each job
    # This is implicitly mounted at /pvc to our training container
    pvc:
      # -- Storage class for the training job PVC.
      storageClass: "local-nfs"
      # -- Size of the training job PVC.
      size: 5Gi
      # -- Volume access mode for the training job PVC.
      volumeAccessMode: "ReadWriteMany"

  # -- Models configuration.
  # @default -- This object has default values for the supported models.
  models:
    # -- Llama 3.2 3B Instruct model configuration.
    # @default -- This object has the following default values for the Llama 3.2 3B Instruct model.
    meta/llama-3.2-3b-instruct:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI.
      model_uri: ngc://nvidia/nemo/llama-3_2-3b-instruct:2.0
      # -- Path where model files are stored.
      model_path: llama32_3b-instruct
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 1
          num_nodes: 1
          tensor_parallel_size: 1
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 3000000000
      # -- Model precision format.
      precision: bf16-mixed
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

    # -- Llama 3.2 1B model configuration.
    # @default -- This object has the following default values for the Llama 3.2 1B model.
    meta/llama-3.2-1b:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI for Llama 3.2 1B model.
      model_uri: ngc://nvidia/nemo/llama-3_2-1b:2.0
      # -- Path where model files are stored.
      model_path: llama32_1b
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 1
          num_nodes: 1
          tensor_parallel_size: 1
        - training_type: sft
          finetuning_type: all_weights
          num_gpus: 1
          num_nodes: 1
          tensor_parallel_size: 1
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 1000000000
      # -- Model precision format.
      precision: bf16-mixed
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

    # -- Llama 3.2 1B Instruct model configuration.
    # @default -- This object has the following default values for the Llama 3.2 1B Instruct model.
    meta/llama-3.2-1b-instruct:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI for Llama 3.2 1B Instruct model.
      model_uri: ngc://nvidia/nemo/llama-3_2-1b-instruct:2.0
      # -- Path where model files are stored.
      model_path: llama32_1b-instruct
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 1
          num_nodes: 1
          tensor_parallel_size: 1
        - training_type: sft
          finetuning_type: all_weights
          num_gpus: 1
          num_nodes: 1
          tensor_parallel_size: 1
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 1000000000
      # -- Model precision format.
      precision: bf16-mixed
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

    # -- Llama 3 70B Instruct model configuration.
    # @default -- This object has the following default values for the Llama 3 70B Instruct model.
    meta/llama3-70b-instruct:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI for Llama 3 70B Instruct model.
      model_uri: ngc://nvidia/nemo/llama-3-70b-instruct-nemo:2.0
      # -- Path where model files are stored.
      model_path: llama-3-70b-bf16
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 4
          num_nodes: 1
          tensor_parallel_size: 4
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 70000000000
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Model precision format.
      precision: bf16-mixed
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

    # -- Llama 3.1 8B Instruct model configuration.
    # @default -- This object has the following default values for the Llama 3.1 8B Instruct model.
    meta/llama-3.1-8b-instruct:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI for Llama 3.1 8B Instruct model.
      model_uri: ngc://nvidia/nemo/llama-3_1-8b-instruct-nemo:2.0
      # -- Path where model files are stored.
      model_path: llama-3_1-8b-instruct_0_0_1
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 1
        - training_type: sft
          finetuning_type: all_weights
          num_gpus: 8
          num_nodes: 1
          tensor_parallel_size: 4
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 8000000000
      # -- Model precision format.
      precision: bf16-mixed
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

    # -- Llama 3.1 70B Instruct model configuration.
    # @default -- This object has the following default values for the Llama 3.1 70B Instruct model.
    meta/llama-3.1-70b-instruct:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI for Llama 3.1 70B Instruct model.
      model_uri: ngc://nvidia/nemo/llama-3_1-70b-instruct-nemo:2.0
      # -- Path where model files are stored.
      model_path: llama-3_1-70b-instruct_0_0_1
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 4
          num_nodes: 1
          tensor_parallel_size: 4
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 70000000000
      # -- Model precision format.
      precision: bf16-mixed
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

    # -- Phi-4 model configuration.
    # @default -- This object has the following default values for the Phi-4.
    microsoft/phi-4:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI for Phi-4 model.
      model_uri: ngc://nvidia/nemo/phi-4:1.0
      # -- Path where model files are stored.
      model_path: phi-4
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 1
          num_nodes: 1
        - training_type: sft
          finetuning_type: all_weights
          num_gpus: 4
          num_nodes: 2
          tensor_parallel_size: 8
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 14659507200
      # -- Model precision format.
      precision: bf16
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

    # -- Llama 3.3 70B Instruct model configuration.
    # @default -- This object has the following default values for the Llama 3.3 70B Instruct model.
    meta/llama-3.3-70b-instruct:
      # -- Whether to enable the model.
      enabled: false
      # -- NGC model URI for Llama 3.3 70B Instruct model.
      model_uri: ngc://nvidia/nemo/llama-3_3-70b-instruct:2.0
      # -- Path where model files are stored.
      model_path: llama-3_3-70b-instruct_0_0_1
      # -- Training options for different fine-tuning methods.
      training_options:
        - training_type: sft
          finetuning_type: lora
          num_gpus: 4
          num_nodes: 1
          tensor_parallel_size: 4
      # -- Micro batch size for training.
      micro_batch_size: 1
      # -- Maximum sequence length for input tokens.
      max_seq_length: 4096
      # -- Number of model parameters.
      num_parameters: 70000000000
      # -- Model precision format.
      precision: bf16-mixed
      # -- Template for formatting prompts.
      prompt_template: "{prompt} {completion}"

  # -- OpenTelemetry settings.
  # @default -- This object has the following default values for the OpenTelemetry settings.
  openTelemetry:
    # -- Whether to enable OpenTelemetry.
    enabled: true
    # -- Sets the traces exporter type (otlp, console, none).
    tracesExporter: otlp
    # -- Sets the metrics exporter type (otlp, console, none).
    metricsExporter: otlp
    # -- Sets the logs exporter type (otlp, console, none).
    logsExporter: otlp
    # -- Endpoint to access a custom OTLP collector listening on port 4317. Example: "http://$(HOST_IP):4317".
    exporterOtlpEndpoint: ""

  # -- Tolerations on the customization job pods.
  tolerations: []

# -- PostgreSQL configuration for the NeMo Customizer microservice.
# @default -- This object has the following default values for the PostgreSQL configuration.
postgresql:
  # -- Whether to enable or disable the PostgreSQL helm chart.
  enabled: true
  auth:
    # -- Whether to assign a password to the "postgres" admin user. Otherwise, remote access will be blocked for this user.
    enablePostgresUser: true
    # -- Name for a custom user to create.
    username: nemo
    # -- Password for the custom user to create.
    password: nemo
    # -- Name for a custom database to create.
    database: finetuning
    # -- Name of existing secret to use for PostgreSQL credentials.
    existingSecret: ""
  # -- PostgreSQL architecture (`standalone` or `replication`).
  architecture: standalone

# -- External PostgreSQL configuration.
# @default -- This object has the following default values for the external PostgreSQL configuration.
externalDatabase:
  # -- External database host address.
  host: localhost
  # -- External database port number.
  port: 5432
  # -- Non-root username for the NeMo Customizer microservice.
  user: nemo
  # -- Password for the non-root username for the NeMo Customizer microservice.
  password: nemo
  # -- Name of the database to use.
  database: finetuning
  # -- Name of an existing secret resource containing the database credentials.
  existingSecret: ""
  # -- Name of an existing secret key containing the database credentials.
  existingSecretPasswordKey: ""
  uriSecret:
    name: ""
    key: ""

# -- WandB configuration.
# @default -- This object has the following default values for the WandB configuration.
wandb:
  # -- WandB secret value. Must contain exactly 32 alphanumeric characters. Creates a new Kubernetes secret named "wandb-secret" with key-value pair "encryption_key=<wandb.secretValue>". Ignored if wandb.existingSecret is set.
  secretValue: ec60d96b639764ccf9859bc10d4363d1
  # -- Name of an existing Kubernetes secret resource for the WandB encryption secret.
  existingSecret: ""
  # -- Name of the key in the existing WandB secret containing the secret value. The secret value must be exactly 32 alphanumeric characters: ^[a-zA-Z0-9]{32}$
  existingSecretKey: ""

# -- Ingress configuration.
# @default -- This object has the following default values for the Ingress configuration.
ingress:
  # -- Whether to enable the ingress resource.
  enabled: false
  # -- Ingress class name.
  className: ""
  # -- Additional annotations for the Ingress resource.
  annotations: {}
  # -- Hostname for the ingress resource.
  hostname: ""
  # -- (list) A list of maps, each containing the keys `host` and `paths` for the ingress resource. You must specify a list for configuring ingress for the microservice.
  # @default -- []
  hosts: {}
    # - host: ""
    #   paths:
    #     - path: /
    #       pathType: ImplementationSpecific
  # -- TLS configuration for the ingress resource.
  tls: []

# -- Open Telemetry Collector configuration.
# @default -- This object has the following default values for the Open Telemetry Collector configuration.
opentelemetry-collector:
  # -- Switch to enable or disable Open Telemetry Collector.
  enabled: true
  image:
    # -- Repository for Open Telemetry Collector image.
    repository: "otel/opentelemetry-collector-k8s"
    # -- Overrides the image tag whose default is the chart appVersion.
    tag: "0.102.1"
  # -- Deployment mode for Open Telemetry Collector. Valid values are "daemonset", "deployment", and "statefulset".
  mode: deployment
  # -- Base collector configuration for Open Telemetry Collector.
  config:
    receivers:
      otlp:
        protocols:
          grpc: {}
          http:
            cors:
              allowed_origins:
                - "*"
    exporters:
      debug:
        verbosity: detailed
    extensions:
      health_check: {}
      zpages:
        endpoint: 0.0.0.0:55679
    processors:
      batch: {}
    service:
      extensions: [zpages, health_check]
      pipelines:
        traces:
          receivers: [otlp]
          exporters: [debug]
          processors: [batch]
        metrics:
          receivers: [otlp]
          exporters: [debug]
          processors: [batch]
        logs:
          receivers: [otlp]
          exporters: [debug]
          processors: [batch]

# -- Enable or disable RunAI executor.
useRunAIExecutor: false