Class: Aws::SageMakerTrainingSessionRuntime::Client

Inherits:
Seahorse::Client::Base show all
Includes:
ClientStubs
Defined in:
gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb

Overview

An API client for SageMakerTrainingSessionRuntime. To construct a client, you need to configure a :region and :credentials.

client = Aws::SageMakerTrainingSessionRuntime::Client.new(
  region: region_name,
  credentials: credentials,
  # ...
)

For details on configuring region and credentials see the developer guide.

See #initialize for a full list of supported configuration options.

Instance Attribute Summary

Attributes inherited from Seahorse::Client::Base

#config, #handlers

API Operations collapse

Instance Method Summary collapse

Methods included from ClientStubs

#api_requests, #stub_data, #stub_responses

Methods inherited from Seahorse::Client::Base

add_plugin, api, clear_plugins, define, new, #operation_names, plugins, remove_plugin, set_api, set_plugins

Methods included from Seahorse::Client::HandlerBuilder

#handle, #handle_request, #handle_response

Constructor Details

#initialize(options) ⇒ Client

Returns a new instance of Client.

Parameters:

  • options (Hash)

Options Hash (options):

  • :plugins (Array<Seahorse::Client::Plugin>) — default: []] —

    A list of plugins to apply to the client. Each plugin is either a class name or an instance of a plugin class.

  • :credentials (required, Aws::CredentialProvider) —

    Your AWS credentials used for authentication. This can be any class that includes and implements Aws::CredentialProvider, or instance of any one of the following classes:

    • Aws::Credentials - Used for configuring static, non-refreshing credentials.

    • Aws::SharedCredentials - Used for loading static credentials from a shared file, such as ~/.aws/config.

    • Aws::AssumeRoleCredentials - Used when you need to assume a role.

    • Aws::AssumeRoleWebIdentityCredentials - Used when you need to assume a role after providing credentials via the web.

    • Aws::SSOCredentials - Used for loading credentials from AWS SSO using an access token generated from aws login.

    • Aws::ProcessCredentials - Used for loading credentials from a process that outputs to stdout.

    • Aws::InstanceProfileCredentials - Used for loading credentials from an EC2 IMDS on an EC2 instance.

    • Aws::ECSCredentials - Used for loading credentials from instances running in ECS.

    • Aws::CognitoIdentityCredentials - Used for loading credentials from the Cognito Identity service.

    When :credentials are not configured directly, the following locations will be searched for credentials:

    • Aws.config[:credentials]

    • The :access_key_id, :secret_access_key, :session_token, and :account_id options.

    • ENV['AWS_ACCESS_KEY_ID'], ENV['AWS_SECRET_ACCESS_KEY'], ENV['AWS_SESSION_TOKEN'], and ENV['AWS_ACCOUNT_ID'].

    • ~/.aws/credentials

    • ~/.aws/config

    • EC2/ECS IMDS instance profile - When used by default, the timeouts are very aggressive. Construct and pass an instance of Aws::InstanceProfileCredentials or Aws::ECSCredentials to enable retries and extended timeouts. Instance profile credential fetching can be disabled by setting ENV['AWS_EC2_METADATA_DISABLED'] to true.

  • :region (required, String) —

    The AWS region to connect to. The configured :region is used to determine the service :endpoint. When not passed, a default :region is searched for in the following locations:

    • Aws.config[:region]
    • ENV['AWS_REGION']
    • ENV['AMAZON_REGION']
    • ENV['AWS_DEFAULT_REGION']
    • ~/.aws/credentials
    • ~/.aws/config
  • :access_key_id (String)
  • :account_id (String)
  • :active_endpoint_cache (Boolean) — default: false —

    When set to true, a thread polling for endpoints will be running in the background every 60 secs (default). Defaults to false.

  • :adaptive_retry_wait_to_fill (Boolean) — default: true —

    Used only in adaptive retry mode. When true, the request will sleep until there is sufficent client side capacity to retry the request. When false, the request will raise a RetryCapacityNotAvailableError and will not retry instead of sleeping.

  • :auth_scheme_preference (Array<String>) —

    A list of preferred authentication schemes to use when making a request. Supported values are: sigv4, sigv4a, httpBearerAuth, and noAuth. When set using ENV['AWS_AUTH_SCHEME_PREFERENCE'] or in shared config as auth_scheme_preference, the value should be a comma-separated list.

  • :client_side_monitoring (Boolean) — default: false —

    When true, client-side metrics will be collected for all API requests from this client.

  • :client_side_monitoring_client_id (String) — default: "" —

    Allows you to provide an identifier for this client which will be attached to all generated client side metrics. Defaults to an empty string.

  • :client_side_monitoring_host (String) — default: "127.0.0.1" —

    Allows you to specify the DNS hostname or IPv4 or IPv6 address that the client side monitoring agent is running on, where client metrics will be published via UDP.

  • :client_side_monitoring_port (Integer) — default: 31000 —

    Required for publishing client metrics. The port that the client side monitoring agent is running on, where client metrics will be published via UDP.

  • :client_side_monitoring_publisher (Aws::ClientSideMonitoring::Publisher) — default: Aws::ClientSideMonitoring::Publisher —

    Allows you to provide a custom client-side monitoring publisher class. By default, will use the Client Side Monitoring Agent Publisher.

  • :convert_params (Boolean) — default: true —

    When true, an attempt is made to coerce request parameters into the required types.

  • :correct_clock_skew (Boolean) — default: true —

    Used only in standard and adaptive retry modes. Specifies whether to apply a clock skew correction and retry requests with skewed client clocks. Takes effect only when disable_clock_skew_correction is not set.

  • :defaults_mode (String) — default: "legacy" —

    See DefaultsModeConfiguration for a list of the accepted modes and the configuration defaults that are included.

  • :disable_clock_skew_correction (Boolean) —

    Used only in standard and adaptive retry modes. When true, the SDK does not apply a clock skew correction, does not update its stored skew from responses, and does not retry clock skew errors. Clock skew correction is enabled by default.

  • :disable_host_prefix_injection (Boolean) — default: false —

    When true, the SDK will not prepend the modeled host prefix to the endpoint.

  • :disable_request_compression (Boolean) — default: false —

    When set to 'true' the request body will not be compressed for supported operations.

  • :endpoint (String, URI::HTTPS, URI::HTTP) —

    Normally you should not configure the :endpoint option directly. This is normally constructed from the :region option. Configuring :endpoint is normally reserved for connecting to test or custom endpoints. The endpoint should be a URI formatted like:

    'http://example.com'
'https://example.com'
'http://example.com:123'
  • :endpoint_cache_max_entries (Integer) — default: 1000 —

    Used for the maximum size limit of the LRU cache storing endpoints data for endpoint discovery enabled operations. Defaults to 1000.

  • :endpoint_cache_max_threads (Integer) — default: 10 —

    Used for the maximum threads in use for polling endpoints to be cached, defaults to 10.

  • :endpoint_cache_poll_interval (Integer) — default: 60 —

    When :endpoint_discovery and :active_endpoint_cache is enabled, Use this option to config the time interval in seconds for making requests fetching endpoints information. Defaults to 60 sec.

  • :endpoint_discovery (Boolean) — default: false —

    When set to true, endpoint discovery will be enabled for operations when available.

  • :ignore_configured_endpoint_urls (Boolean) —

    Setting to true disables use of endpoint URLs provided via environment variables and the shared configuration file.

  • :log_formatter (Aws::Log::Formatter) — default: Aws::Log::Formatter.default —

    The log formatter.

  • :log_level (Symbol) — default: :info —

    The log level to send messages to the :logger at.

  • :logger (Logger) —

    The Logger instance to send log messages to. If this option is not set, logging will be disabled.

  • :max_attempts (Integer) — default: 3 —

    An integer representing the maximum number attempts that will be made for a single request, including the initial attempt. For example, setting this value to 5 will result in a request being retried up to 4 times. Used in standard and adaptive retry modes. Defaults to 3, except for DynamoDB and DynamoDB Streams, which default to 4.

  • :profile (String) — default: "default" —

    Used when loading credentials from the shared credentials file at HOME/.aws/credentials. When not specified, 'default' is used.

  • :request_checksum_calculation (String) — default: "when_supported" —

    Determines when a checksum will be calculated for request payloads. Values are:

    • when_supported - (default) When set, a checksum will be calculated for all request payloads of operations modeled with the httpChecksum trait where requestChecksumRequired is true and/or a requestAlgorithmMember is modeled.
    • when_required - When set, a checksum will only be calculated for request payloads of operations modeled with the httpChecksum trait where requestChecksumRequired is true or where a requestAlgorithmMember is modeled and supplied.
  • :request_min_compression_size_bytes (Integer) — default: 10240 —

    The minimum size in bytes that triggers compression for request bodies. The value must be non-negative integer value between 0 and 10485780 bytes inclusive.

  • :response_checksum_validation (String) — default: "when_supported" —

    Determines when checksum validation will be performed on response payloads. Values are:

    • when_supported - (default) When set, checksum validation is performed on all response payloads of operations modeled with the httpChecksum trait where responseAlgorithms is modeled, except when no modeled checksum algorithms are supported.
    • when_required - When set, checksum validation is not performed on response payloads of operations unless the checksum algorithm is supported and the requestValidationModeMember member is set to ENABLED.
  • :retry_backoff (Proc) —

    A proc or lambda used for backoff. Defaults to 2**retries * retry_base_delay. This option is only used in the legacy retry mode.

  • :retry_base_delay (Float) — default: 0.3 —

    The base delay in seconds used by the default backoff function. This option is only used in the legacy retry mode.

  • :retry_jitter (Symbol) — default: :none —

    A delay randomiser function used by the default backoff function. Some predefined functions can be referenced by name - :none, :equal, :full, otherwise a Proc that takes and returns a number. This option is only used in the legacy retry mode.

    @see https://www.awsarchitectureblog.com/2015/03/backoff.html

  • :retry_limit (Integer) — default: 3 —

    The maximum number of times to retry failed requests. Only ~ 500 level server errors and certain ~ 400 level client errors are retried. Generally, these are throttling errors, data checksum errors, networking errors, timeout errors, auth errors, endpoint discovery, and errors from expired credentials. This option is only used in the legacy retry mode.

  • :retry_max_delay (Integer) — default: 0 —

    The maximum number of seconds to delay between retries (0 for no limit) used by the default backoff function. This option is only used in the legacy retry mode.

  • :retry_mode (String) — default: "standard" —

    Specifies which retry algorithm to use. Values are:

    • legacy - The pre-existing retry behavior.

    • standard - A standardized set of retry rules across the AWS SDKs. This includes support for retry quotas, which limit the number of unsuccessful retries a client can make. This is the default value if no retry mode is provided.

    • adaptive - A retry mode that includes all the functionality of standard mode along with automatic client side throttling.

  • :sdk_ua_app_id (String) —

    A unique and opaque application ID that is appended to the User-Agent header as app/sdk_ua_app_id. It should have a maximum length of 50. This variable is sourced from environment variable AWS_SDK_UA_APP_ID or the shared config profile attribute sdk_ua_app_id.

  • :secret_access_key (String)
  • :session_token (String)
  • :sigv4a_signing_region_set (Array) —

    A list of regions that should be signed with SigV4a signing. When not passed, a default :sigv4a_signing_region_set is searched for in the following locations:

    • Aws.config[:sigv4a_signing_region_set]
    • ENV['AWS_SIGV4A_SIGNING_REGION_SET']
    • ~/.aws/config
  • :stub_responses (Boolean) — default: false —

    Causes the client to return stubbed responses. By default fake responses are generated and returned. You can specify the response data to return or errors to raise by calling ClientStubs#stub_responses. See ClientStubs for more information.

    Please note When response stubbing is enabled, no HTTP requests are made, and retries are disabled.

  • :telemetry_provider (Aws::Telemetry::TelemetryProviderBase) — default: Aws::Telemetry::NoOpTelemetryProvider —

    Allows you to provide a telemetry provider, which is used to emit telemetry data. By default, uses NoOpTelemetryProvider which will not record or emit any telemetry data. The SDK supports the following telemetry providers:

    • OpenTelemetry (OTel) - To use the OTel provider, install and require the opentelemetry-sdk gem and then, pass in an instance of a Aws::Telemetry::OTelProvider for telemetry provider.
  • :token_provider (Aws::TokenProvider) —

    Your Bearer token used for authentication. This can be any class that includes and implements Aws::TokenProvider, or instance of any one of the following classes:

    • Aws::StaticTokenProvider - Used for configuring static, non-refreshing tokens.

    • Aws::SSOTokenProvider - Used for loading tokens from AWS SSO using an access token generated from aws login.

    When :token_provider is not configured directly, the Aws::TokenProviderChain will be used to search for tokens configured for your profile in shared configuration files.

  • :use_dualstack_endpoint (Boolean) —

    When set to true, dualstack enabled endpoints (with .aws TLD) will be used if available.

  • :use_fips_endpoint (Boolean) —

    When set to true, fips compatible endpoints will be used if available. When a fips region is used, the region is normalized and this config is set to true.

  • :validate_params (Boolean) — default: true —

    When true, request parameters are validated before sending the request.

  • :endpoint_provider (Aws::SageMakerTrainingSessionRuntime::EndpointProvider) —

    The endpoint provider used to resolve endpoints. Any object that responds to #resolve_endpoint(parameters) where parameters is a Struct similar to Aws::SageMakerTrainingSessionRuntime::EndpointParameters.

  • :http_continue_timeout (Float) — default: 1 —

    The number of seconds to wait for a 100-continue response before sending the request body. This option has no effect unless the request has "Expect" header set to "100-continue". Defaults to nil which disables this behaviour. This value can safely be set per request on the session.

  • :http_idle_timeout (Float) — default: 5 —

    The number of seconds a connection is allowed to sit idle before it is considered stale. Stale connections are closed and removed from the pool before making a request.

  • :http_open_timeout (Float) — default: 15 —

    The default number of seconds to wait for response data. This value can safely be set per-request on the session.

  • :http_proxy (URI::HTTP, String) —

    A proxy to send requests through. Formatted like 'http://proxy.com:123'.

  • :http_read_timeout (Float) — default: 60 —

    The default number of seconds to wait for response data. This value can safely be set per-request on the session.

  • :http_wire_trace (Boolean) — default: false —

    When true, HTTP debug output will be sent to the :logger.

  • :on_chunk_received (Proc) —

    When a Proc object is provided, it will be used as callback when each chunk of the response body is received. It provides three arguments: the chunk, the number of bytes received, and the total number of bytes in the response (or nil if the server did not send a content-length).

  • :on_chunk_sent (Proc) —

    When a Proc object is provided, it will be used as callback when each chunk of the request body is sent. It provides three arguments: the chunk, the number of bytes read from the body, and the total number of bytes in the body.

  • :raise_response_errors (Boolean) — default: true —

    When true, response errors are raised.

  • :ssl_ca_bundle (String) —

    Full path to the SSL certificate authority bundle file that should be used when verifying peer certificates. If you do not pass :ssl_ca_bundle or :ssl_ca_directory the the system default will be used if available.

  • :ssl_ca_directory (String) —

    Full path of the directory that contains the unbundled SSL certificate authority files for verifying peer certificates. If you do not pass :ssl_ca_bundle or :ssl_ca_directory the the system default will be used if available.

  • :ssl_ca_store (String) —

    Sets the X509::Store to verify peer certificate.

  • :ssl_cert (OpenSSL::X509::Certificate) —

    Sets a client certificate when creating http connections.

  • :ssl_key (OpenSSL::PKey) —

    Sets a client key when creating http connections.

  • :ssl_timeout (Float) —

    Sets the SSL timeout in seconds

  • :ssl_verify_peer (Boolean) — default: true —

    When true, SSL peer certificates are verified when establishing a connection.



479
480
481
 
# File 'gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb', line 479

def initialize(*args)
  super
end

Instance Method Details

#execute_sampling_operation(params = {}) ⇒ Types::ExecuteSamplingOperationOutput

Submits an asynchronous sampling operation to generate text or compute log-probabilities from the model associated with the specified training session. The session must be in Running or Blocked state. Use GetSamplingOperationResult to poll for the result.

Examples:

Example: Generate samples from a prompt


# Generates completions for a tokenized prompt using the session's current weights. Generation behavior is controlled by
# `samplingParams`, and the operation is dispatched asynchronously.

resp = client.execute_sampling_operation({
  operation_config: {
    sample: {
      num_samples: 4, 
      prompt: {
        chunks: [
          {
            type: "encoded_text", 
            tokens: [
              791, 
              6864, 
              315, 
              9822, 
              374, 
            ], 
          }, 
        ], 
      }, 
      sampling_params: {
        max_tokens: 1024, 
        temperature: 0.7, 
        top_p: 0.9, 
      }, 
    }, 
  }, 
  training_session_name: "my-training-session", 
})

resp.to_h outputs the following:
{
  operation_id: "b2c3d4e5-6789-01bc-def0-234567890abc", 
}

Request syntax with placeholder values


resp = client.execute_sampling_operation({
  training_session_name: "TrainingSessionName", # required
  operation_config: { # required
    sample: {
      prompt: { # required
        chunks: [ # required
          {
            tokens: [1], # required
            type: "encoded_text", # required, accepts encoded_text
          },
        ],
      },
      num_samples: 1, # required
      sampling_params: { # required
        max_tokens: 1, # required
        temperature: 1.0,
        top_k: 1,
        top_p: 1.0,
        seed: 1,
        stop: ["String"],
      },
      include_prompt_logprobs: false,
      topk_prompt_logprobs: 1,
      model_package_arn: "String",
    },
    compute_logprobs: {
      prompt: { # required
        chunks: [ # required
          {
            tokens: [1], # required
            type: "encoded_text", # required, accepts encoded_text
          },
        ],
      },
      model_package_arn: "String",
    },
    load_weights_for_sampling: {
      model_package_arn: "String", # required
    },
  },
})

Response structure


resp.operation_id #=> String

Parameters:

  • params (Hash) (defaults to: {}) —

    ({})

Options Hash (params):

  • :training_session_name (required, String) —

    The name of the training session to sample from.

  • :operation_config (required, Types::SampleOperationConfig) —

    The configuration for the sample operation. Specify either a sample configuration to generate text or a computeLogprobs configuration to compute token log-probabilities.

Returns:

See Also:



593
594
595
596
 
# File 'gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb', line 593

def execute_sampling_operation(params = {}, options = {})
  req = build_request(:execute_sampling_operation, params)
  req.send_request(options)
end

#execute_training_operation(params = {}) ⇒ Types::ExecuteTrainingOperationOutput

Submits an asynchronous training operation to the specified training session. The service executes operations in sequence number order. The session must be in Running or Blocked state. Use GetTrainingOperationResult to poll for the result. Supported operation types: forward, forwardBackward, optimizerStep, saveWeights, saveState, and loadState.

Examples:

Example: Run a forward and backward pass


# Computes gradients for a batch of tokenized prompts and accumulates them into the session. Operations are applied in
# `sequenceNumber` order, so each call must supply the next number in the sequence.

resp = client.execute_training_operation({
  operation_config: {
    forward_backward: {
      datum_list: [
        {
          model_input: {
            chunks: [
              {
                type: "encoded_text", 
                tokens: [
                  791, 
                  6864, 
                  315, 
                  9822, 
                  374, 
                  12366, 
                ], 
              }, 
            ], 
          }, 
        }, 
      ], 
      loss_fn_name: "cross_entropy", 
    }, 
  }, 
  sequence_number: 1, 
  training_session_name: "my-training-session", 
})

resp.to_h outputs the following:
{
  operation_id: "a1b2c3d4-5678-90ab-cdef-1234567890ab", 
}

Request syntax with placeholder values


resp = client.execute_training_operation({
  training_session_name: "TrainingSessionName", # required
  sequence_number: 1, # required
  client_token: "ClientToken",
  operation_config: { # required
    forward: {
      loss_fn_name: "cross_entropy", # required, accepts cross_entropy, importance_sampling, ppo, cispo
      datum_list: [ # required
        {
          model_input: { # required
            chunks: [ # required
              {
                tokens: [1], # required
                type: "encoded_text", # required, accepts encoded_text
              },
            ],
          },
          loss_fn_inputs: {
            "String" => {
              data: { # required
              },
              dtype: "String", # required
              shape: [1],
            },
          },
        },
      ],
      loss_fn_config: {
      },
    },
    forward_backward: {
      loss_fn_name: "cross_entropy", # required, accepts cross_entropy, importance_sampling, ppo, cispo
      datum_list: [ # required
        {
          model_input: { # required
            chunks: [ # required
              {
                tokens: [1], # required
                type: "encoded_text", # required, accepts encoded_text
              },
            ],
          },
          loss_fn_inputs: {
            "String" => {
              data: { # required
              },
              dtype: "String", # required
              shape: [1],
            },
          },
        },
      ],
      loss_fn_config: {
      },
    },
    optimizer_step: {
      adam_params_config: { # required
        learning_rate: 1.0, # required
        beta1: 1.0,
        beta2: 1.0,
        eps: 1.0,
        weight_decay: 1.0,
        grad_clip_norm: 1.0,
        lora_plus_lr_ratio: 1.0,
      },
    },
    save_weights: {
      merge_weights: false,
      description: "SaveWeightsConfigDescriptionString",
    },
    save_state: {
      description: "SaveStateConfigDescriptionString",
    },
    load_state: {
      model_package_arn: "String", # required
      load_optimizer: false, # required
    },
  },
})

Response structure


resp.operation_id #=> String

Parameters:

  • params (Hash) (defaults to: {}) —

    ({})

Options Hash (params):

  • :training_session_name (required, String) —

    The name of the training session to submit the operation to.

  • :sequence_number (required, Integer) —

    The monotonically increasing sequence number that determines execution order. The training session processes operations strictly in sequence number order. If a gap exists, the session blocks until the missing operation arrives or times out. Must be at least 1.

  • :client_token (String) —

    A unique, case-sensitive identifier that you provide to ensure that the operation completes no more than one time. If you retry a request with the same client token and the same parameters, the service ignores the request but does not return an error.

  • :operation_config (required, Types::OperationConfig) —

    The configuration that specifies the type and parameters of the training operation. Exactly one member of the union must be set.

Returns:

See Also:



758
759
760
761
 
# File 'gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb', line 758

def execute_training_operation(params = {}, options = {})
  req = build_request(:execute_training_operation, params)
  req.send_request(options)
end

#get_sampling_operation_result(params = {}) ⇒ Types::GetSamplingOperationResultOutput

Retrieves the current status and result of a previously submitted sampling operation. Poll this operation until the status reaches a terminal state (Completed or Failed).

Examples:

Example: Get the result of a sampling operation


# Retrieves the status and generated sequences for a sampling operation submitted with ExecuteSamplingOperation.

resp = client.get_sampling_operation_result({
  operation_id: "a1b2c3d4-5678-90ab-cdef-1234567890ab", 
  training_session_name: "my-training-session", 
})

resp.to_h outputs the following:
{
  operation_result: {
    sample: {
      sequences: [
        {
          stop_reason: "stop", 
          token_usage: {
            decode: 7, 
            prefill: 5, 
          }, 
          tokens: [
            791, 
            6864, 
            315, 
            9822, 
            374, 
            12366, 
            13, 
          ], 
        }, 
      ], 
    }, 
  }, 
  status: "Completed", 
}

Request syntax with placeholder values


resp = client.get_sampling_operation_result({
  training_session_name: "TrainingSessionName", # required
  operation_id: "OperationId", # required
})

Response structure


resp.status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Skipped"
resp.operation_result.sample.sequences #=> Array
resp.operation_result.sample.sequences[0].stop_reason #=> String, one of "stop", "length"
resp.operation_result.sample.sequences[0].tokens #=> Array
resp.operation_result.sample.sequences[0].tokens[0] #=> Integer
resp.operation_result.sample.sequences[0].logprobs #=> Array
resp.operation_result.sample.sequences[0].logprobs[0] #=> Float
resp.operation_result.sample.sequences[0].token_usage.prefill #=> Integer
resp.operation_result.sample.sequences[0].token_usage.decode #=> Integer
resp.operation_result.sample.prompt_logprobs #=> Array
resp.operation_result.sample.prompt_logprobs[0] #=> Float
resp.operation_result.sample.topk_prompt_logprobs #=> Array
resp.operation_result.sample.topk_prompt_logprobs[0] #=> Array
resp.operation_result.sample.topk_prompt_logprobs[0][0].token_id #=> Integer
resp.operation_result.sample.topk_prompt_logprobs[0][0].logprob #=> Float
resp.operation_result.compute_logprobs.logprobs #=> Array
resp.operation_result.compute_logprobs.logprobs[0] #=> Float
resp.error_details #=> String

Parameters:

  • params (Hash) (defaults to: {}) —

    ({})

Options Hash (params):

  • :training_session_name (required, String) —

    The name of the training session that owns the sample operation.

  • :operation_id (required, String) —

    The unique identifier of the sample operation to retrieve, returned by ExecuteSamplingOperation.

Returns:

See Also:



849
850
851
852
 
# File 'gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb', line 849

def get_sampling_operation_result(params = {}, options = {})
  req = build_request(:get_sampling_operation_result, params)
  req.send_request(options)
end

#get_training_operation_result(params = {}) ⇒ Types::GetTrainingOperationResultOutput

Retrieves the current status and result of a previously submitted training operation. Poll this operation until the status reaches a terminal state (Completed, Failed, or Skipped).

Examples:

Example: Get the result of a training operation


# Retrieves the status and result of a training operation submitted with ExecuteTrainingOperation.

resp = client.get_training_operation_result({
  operation_id: "a1b2c3d4-5678-90ab-cdef-1234567890ab", 
  training_session_name: "my-training-session", 
})

resp.to_h outputs the following:
{
  operation_result: {
    forward_backward: {
      loss_fn_output_type: "logprobs", 
      metrics: {
        "loss" => 1.8342, 
      }, 
    }, 
  }, 
  status: "Completed", 
  token_usage: {
    training_tokens: 4096, 
  }, 
}

Request syntax with placeholder values


resp = client.get_training_operation_result({
  training_session_name: "TrainingSessionName", # required
  operation_id: "OperationId", # required
})

Response structure


resp.status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Skipped"
resp.operation_result.forward.loss_fn_output_type #=> String
resp.operation_result.forward.loss_fn_outputs #=> Array
resp.operation_result.forward.loss_fn_outputs[0] #=> Hash
resp.operation_result.forward.loss_fn_outputs[0]["String"].dtype #=> String
resp.operation_result.forward.loss_fn_outputs[0]["String"].shape #=> Array
resp.operation_result.forward.loss_fn_outputs[0]["String"].shape[0] #=> Integer
resp.operation_result.forward.metrics #=> Hash
resp.operation_result.forward.metrics["String"] #=> Float
resp.operation_result.forward.token_usage.training_tokens #=> Integer
resp.operation_result.forward_backward.loss_fn_output_type #=> String
resp.operation_result.forward_backward.loss_fn_outputs #=> Array
resp.operation_result.forward_backward.loss_fn_outputs[0] #=> Hash
resp.operation_result.forward_backward.loss_fn_outputs[0]["String"].dtype #=> String
resp.operation_result.forward_backward.loss_fn_outputs[0]["String"].shape #=> Array
resp.operation_result.forward_backward.loss_fn_outputs[0]["String"].shape[0] #=> Integer
resp.operation_result.forward_backward.metrics #=> Hash
resp.operation_result.forward_backward.metrics["String"] #=> Float
resp.operation_result.forward_backward.token_usage.training_tokens #=> Integer
resp.operation_result.optimizer_step.metrics #=> Hash
resp.operation_result.optimizer_step.metrics["String"] #=> Float
resp.operation_result.save_weights.model_package_arn #=> String
resp.operation_result.save_state.model_package_arn #=> String
resp.error_details #=> String
resp.token_usage.training_tokens #=> Integer

Parameters:

  • params (Hash) (defaults to: {}) —

    ({})

Options Hash (params):

  • :training_session_name (required, String) —

    The name of the training session that owns the operation.

  • :operation_id (required, String) —

    The unique identifier of the operation to retrieve, returned by ExecuteTrainingOperation.

Returns:

See Also:



937
938
939
940
 
# File 'gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb', line 937

def get_training_operation_result(params = {}, options = {})
  req = build_request(:get_training_operation_result, params)
  req.send_request(options)
end

#sample_training_session(params = {}) ⇒ Types::SampleTrainingSessionOutput

Sends an inference request to the model associated with the specified training session and returns the response synchronously. The service passes the request and response bodies to and from the model without modification.

Examples:

Example: Sample from a training session


# Sends a chat completion request to the session's current weights and receives the complete response in a single payload.

resp = client.sample_training_session({
  accept: "application/json", 
  body: "eyJtZXNzYWdlcyI6W3sicm9sZSI6InVzZXIiLCJjb250ZW50IjoiV2hhdCBpcyB0aGUgY2FwaXRhbCBvZiBGcmFuY2U/In1dLCJtYXhfdG9rZW5zIjoxMjh9", 
  content_type: "application/json", 
  training_session_name: "my-training-session", 
})

resp.to_h outputs the following:
{
  body: "eyJjaG9pY2VzIjpbeyJtZXNzYWdlIjp7InJvbGUiOiJhc3Npc3RhbnQiLCJjb250ZW50IjoiUGFyaXMifSwiZmluaXNoX3JlYXNvbiI6InN0b3AifV19", 
  content_type: "application/json", 
}

Request syntax with placeholder values


resp = client.sample_training_session({
  training_session_name: "TrainingSessionName", # required
  content_type: "String",
  accept: "String",
  model_package_arn: "String",
  body: "data",
})

Response structure


resp.content_type #=> String
resp.body #=> String

Parameters:

  • params (Hash) (defaults to: {}) —

    ({})

Options Hash (params):

  • :training_session_name (required, String) —

    The name of the training session whose model to invoke.

  • :content_type (String) —

    The MIME type of the input data in the request body.

  • :accept (String) —

    The desired MIME type of the inference body in the response.

  • :model_package_arn (String) —

    The ARN of the model package whose checkpoint to invoke.

  • :body (String, StringIO, File) —

    The raw inference request body.

Returns:

See Also:



1004
1005
1006
1007
 
# File 'gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb', line 1004

def sample_training_session(params = {}, options = {})
  req = build_request(:sample_training_session, params)
  req.send_request(options)
end

#sample_training_session_with_response_stream(params = {}) ⇒ Types::SampleTrainingSessionWithResponseStreamOutput

Streams inference responses from the model associated with the specified training session. Use this operation for real-time token-by-token generation. The service passes the request body to the model without modification.

Examples:

Example: Stream samples from a training session


# Sends a chat completion request to the session's current weights and receives the response as a stream of payload
# chunks.

resp = client.sample_training_session_with_response_stream({
  accept: "application/json", 
  body: "eyJtZXNzYWdlcyI6W3sicm9sZSI6InVzZXIiLCJjb250ZW50IjoiV2hhdCBpcyB0aGUgY2FwaXRhbCBvZiBGcmFuY2U/In1dLCJtYXhfdG9rZW5zIjoxMjgsInN0cmVhbSI6dHJ1ZX0=", 
  content_type: "application/json", 
  training_session_name: "my-training-session", 
})

Request syntax with placeholder values


resp = client.sample_training_session_with_response_stream({
  training_session_name: "TrainingSessionName", # required
  content_type: "String",
  accept: "String",
  model_package_arn: "String",
  body: "data",
})

Response structure


resp.content_type #=> String
resp.body #=> IO

Parameters:

  • params (Hash) (defaults to: {}) —

    ({})

Options Hash (params):

  • :training_session_name (required, String) —

    The name of the training session whose model to invoke.

  • :content_type (String) —

    The MIME type of the input data in the request body.

  • :accept (String) —

    The desired MIME type of the inference body in the response.

  • :model_package_arn (String) —

    The ARN of the model package whose checkpoint to invoke.

  • :body (String, StringIO, File) —

    The raw inference request body.

Returns:

See Also:



1066
1067
1068
1069
 
# File 'gems/aws-sdk-sagemakertrainingsessionruntime/lib/aws-sdk-sagemakertrainingsessionruntime/client.rb', line 1066

def sample_training_session_with_response_stream(params = {}, options = {}, &block)
  req = build_request(:sample_training_session_with_response_stream, params)
  req.send_request(options, &block)
end