package google.ai.generativelanguage.v1beta3
Get desktop application:
View/edit binary Protocol Buffers messages
An API for using Generative Language Models (GLMs) in dialog applications. Also known as large language models (LLMs), this API provides models that are trained for multi-turn dialog.
Generates a response from the model given an input `MessagePrompt`.
message
discuss_service.proto:61Request to generate a message response from the model.
string model = 1
Required. The name of the model to use. Format: `name=models/{model}`.
optional prompt = 2
Required. The structured textual input given to the model as a prompt. Given a prompt, the model will return what it predicts is the next message in the discussion.
optional float temperature = 3
Optional. Controls the randomness of the output. Values can range over `[0.0,1.0]`, inclusive. A value closer to `1.0` will produce responses that are more varied, while a value closer to `0.0` will typically result in less surprising responses from the model.
optional int32 candidate_count = 4
Optional. The number of generated response messages to return. This value must be between `[1, 8]`, inclusive. If unset, this will default to `1`.
optional float top_p = 5
Optional. The maximum cumulative probability of tokens to consider when sampling. The model uses combined Top-k and nucleus sampling. Nucleus sampling considers the smallest set of tokens whose probability sum is at least `top_p`.
optional int32 top_k = 6
Optional. The maximum number of tokens to consider when sampling. The model uses combined Top-k and nucleus sampling. Top-k sampling considers the set of `top_k` most probable tokens.
message
discuss_service.proto:114The response from the model. This includes candidate messages and conversation history in the form of chronologically-ordered messages.
repeated candidates = 1
Candidate response messages from the model.
repeated messages = 2
The conversation history used by the model.
repeated filters = 3
A set of content filtering metadata for the prompt and response text. This indicates which `SafetyCategory`(s) blocked a candidate from this response, the lowest `HarmProbability` that triggered a block, and the HarmThreshold setting for that category.
Runs a model's tokenizer on a string and returns the token count.
message
discuss_service.proto:220Counts the number of tokens in the `prompt` sent to a model. Models may tokenize text differently, so each model may return a different `token_count`.
string model = 1
Required. The model's resource name. This serves as an ID for the Model to use. This name should match a model name returned by the `ListModels` method. Format: `models/{model}`
optional prompt = 2
Required. The prompt, whose token count is to be returned.
message
discuss_service.proto:241A response from `CountMessageTokens`. It returns the model's `token_count` for the `prompt`.
int32 token_count = 1
The number of tokens that the `model` tokenizes the `prompt` into. Always non-negative.
Provides methods for getting metadata information about Generative Models.
rpc GetModel (, )
model_service.proto:39Gets information about a specific Model.
message
model_service.proto:112Request for getting information about a specific Model.
string name = 1
Required. The resource name of the model. This name should match a model name returned by the `ListModels` method. Format: `models/{model}`
rpc ListModels (, )
model_service.proto:47Lists models available through the API.
message
model_service.proto:127Request for listing all Models.
int32 page_size = 2
The maximum number of `Models` to return (per page). The service may return fewer models. If unspecified, at most 50 models will be returned per page. This method returns at most 1000 models per page, even if you pass a larger page_size.
string page_token = 3
A page token, received from a previous `ListModels` call. Provide the `page_token` returned by one request as an argument to the next request to retrieve the next page. When paginating, all other parameters provided to `ListModels` must match the call that provided the page token.
message
model_service.proto:147Response from `ListModel` containing a paginated list of Models.
repeated models = 1
The returned Models.
string next_page_token = 2
A token, which can be sent as `page_token` to retrieve the next page. If this field is omitted, there are no more pages.
rpc GetTunedModel (, )
model_service.proto:55Gets information about a specific TunedModel.
message
model_service.proto:158Request for getting information about a specific Model.
string name = 1
Required. The resource name of the model. Format: `tunedModels/my-model-id`
Lists tuned models owned by the user.
message
model_service.proto:171Request for listing TunedModels.
int32 page_size = 1
Optional. The maximum number of `TunedModels` to return (per page). The service may return fewer tuned models. If unspecified, at most 10 tuned models will be returned. This method returns at most 1000 models per page, even if you pass a larger page_size.
string page_token = 2
Optional. A page token, received from a previous `ListTunedModels` call. Provide the `page_token` returned by one request as an argument to the next request to retrieve the next page. When paginating, all other parameters provided to `ListTunedModels` must match the call that provided the page token.
message
model_service.proto:191Response from `ListTunedModels` containing a paginated list of Models.
repeated tuned_models = 1
The returned Models.
string next_page_token = 2
A token, which can be sent as `page_token` to retrieve the next page. If this field is omitted, there are no more pages.
Creates a tuned model. Intermediate tuning progress (if any) is accessed through the [google.longrunning.Operations] service. Status and results can be accessed through the Operations service. Example: GET /v1/tunedModels/az2mb0bpw6i/operations/000-111-222
message
model_service.proto:202Request to create a TunedModel.
optional string tuned_model_id = 1
Optional. The unique id for the tuned model if specified. This value should be up to 40 characters, the first character must be a letter, the last could be a letter or a number. The id must match the regular expression: [a-z]([a-z0-9-]{0,38}[a-z0-9])?.
optional tuned_model = 2
Required. The tuned model to create.
Updates a tuned model.
message
model_service.proto:235Request to update a TunedModel.
optional tuned_model = 1
Required. The tuned model to update.
optional update_mask = 2
Required. The list of fields to update.
Deletes a tuned model.
message
model_service.proto:245Request to delete a TunedModel.
string name = 1
Required. The resource name of the model. Format: `tunedModels/my-model-id`
Provides methods for managing permissions to PaLM API resources.
Create a permission to a specific resource.
message
permission_service.proto:93Request to create a `Permission`.
string parent = 1
Required. The parent resource of the `Permission`. Format: tunedModels/{tuned_model}
optional permission = 2
Required. The permission to create.
Gets information about a specific Permission.
message
permission_service.proto:108Request for getting information about a specific `Permission`.
string name = 1
Required. The resource name of the permission. Format: `tunedModels/{tuned_model}permissions/{permission}`
Lists permissions for the specific resource.
message
permission_service.proto:121Request for listing permissions.
string parent = 1
Required. The parent resource of the permissions. Format: tunedModels/{tuned_model}
int32 page_size = 2
Optional. The maximum number of `Permission`s to return (per page). The service may return fewer permissions. If unspecified, at most 10 permissions will be returned. This method returns at most 1000 permissions per page, even if you pass larger page_size.
string page_token = 3
Optional. A page token, received from a previous `ListPermissions` call. Provide the `page_token` returned by one request as an argument to the next request to retrieve the next page. When paginating, all other parameters provided to `ListPermissions` must match the call that provided the page token.
message
permission_service.proto:149Response from `ListPermissions` containing a paginated list of permissions.
repeated permissions = 1
Returned permissions.
string next_page_token = 2
A token, which can be sent as `page_token` to retrieve the next page. If this field is omitted, there are no more pages.
Updates the permission.
message
permission_service.proto:160Request to update the `Permission`.
optional permission = 1
Required. The permission to update. The permission's `name` field is used to identify the permission to update.
optional update_mask = 2
Required. The list of fields to update. Accepted ones: - role (`Permission.role` field)
Deletes the permission.
message
permission_service.proto:173Request to delete the `Permission`.
string name = 1
Required. The resource name of the permission. Format: `tunedModels/{tuned_model}/permissions/{permission}`
Transfers ownership of the tuned model. This is the only way to change ownership of the tuned model. The current owner will be downgraded to writer role.
message
permission_service.proto:185Request to transfer the ownership of the tuned model.
string name = 1
Required. The resource name of the tuned model to transfer ownership . Format: `tunedModels/my-model-id`
string email_address = 2
Required. The email address of the user to whom the tuned model is being transferred to.
message
permission_service.proto:202Response from `TransferOwnership`.
(message has no fields)
service TextService
text_service.proto:35API for using Generative Language Models (GLMs) trained to generate text. Also known as Large Language Models (LLM)s, these generate text given an input prompt from the user.
rpc GenerateText (, )
text_service.proto:39Generates a response from the model given an input message.
message
text_service.proto:83Request to generate a text completion response from the model.
string model = 1
Required. The name of the `Model` or `TunedModel` to use for generating the completion. Examples: models/text-bison-001 tunedModels/sentence-translator-u3b7m
optional prompt = 2
Required. The free-form input text given to the model as a prompt. Given a prompt, the model will generate a TextCompletion response it predicts as the completion of the input text.
optional float temperature = 3
Optional. Controls the randomness of the output. Note: The default value varies by model, see the `Model.temperature` attribute of the `Model` returned the `getModel` function. Values can range from [0.0,1.0], inclusive. A value closer to 1.0 will produce responses that are more varied and creative, while a value closer to 0.0 will typically result in more straightforward responses from the model.
optional int32 candidate_count = 4
Optional. Number of generated responses to return. This value must be between [1, 8], inclusive. If unset, this will default to 1.
optional int32 max_output_tokens = 5
Optional. The maximum number of tokens to include in a candidate. If unset, this will default to output_token_limit specified in the `Model` specification.
optional float top_p = 6
Optional. The maximum cumulative probability of tokens to consider when sampling. The model uses combined Top-k and nucleus sampling. Tokens are sorted based on their assigned probabilities so that only the most likely tokens are considered. Top-k sampling directly limits the maximum number of tokens to consider, while Nucleus sampling limits number of tokens based on the cumulative probability. Note: The default value varies by model, see the `Model.top_p` attribute of the `Model` returned the `getModel` function.
optional int32 top_k = 7
Optional. The maximum number of tokens to consider when sampling. The model uses combined Top-k and nucleus sampling. Top-k sampling considers the set of `top_k` most probable tokens. Defaults to 40. Note: The default value varies by model, see the `Model.top_k` attribute of the `Model` returned the `getModel` function.
repeated safety_settings = 8
A list of unique `SafetySetting` instances for blocking unsafe content. that will be enforced on the `GenerateTextRequest.prompt` and `GenerateTextResponse.candidates`. There should not be more than one setting for each `SafetyCategory` type. The API will block any prompts and responses that fail to meet the thresholds set by these settings. This list overrides the default settings for each `SafetyCategory` specified in the safety_settings. If there is no `SafetySetting` for a given `SafetyCategory` provided in the list, the API will use the default safety setting for that category.
repeated string stop_sequences = 9
The set of character sequences (up to 5) that will stop output generation. If specified, the API will stop at the first appearance of a stop sequence. The stop sequence will not be included as part of the response.
message
text_service.proto:163The response from the model, including candidate completions.
repeated candidates = 1
Candidate responses from the model.
repeated filters = 3
A set of content filtering metadata for the prompt and response text. This indicates which `SafetyCategory`(s) blocked a candidate from this response, the lowest `HarmProbability` that triggered a block, and the HarmThreshold setting for that category. This indicates the smallest change to the `SafetySettings` that would be necessary to unblock at least 1 response. The blocking is configured by the `SafetySettings` in the request (or the default `SafetySettings` of the API).
repeated safety_feedback = 4
Returns any safety feedback related to content filtering.
rpc EmbedText (, )
text_service.proto:53Generates an embedding from the model given an input message.
message
text_service.proto:212Request to get a text embedding from the model.
string model = 1
Required. The model name to use with the format model=models/{model}.
string text = 2
Required. The free-form input text that the model will turn into an embedding.
message
text_service.proto:227The response to a EmbedTextRequest.
optional embedding = 1
Output only. The embedding generated from the input text.
rpc BatchEmbedText (, )
text_service.proto:63Generates multiple embeddings from the model given input text in a synchronous call.
message
text_service.proto:233Batch request to get a text embedding from the model.
string model = 1
Required. The name of the `Model` to use for generating the embedding. Examples: models/embedding-gecko-001
repeated string texts = 2
Required. The free-form input texts that the model will turn into an embedding. The current limit is 100 texts, over which an error will be thrown.
message
text_service.proto:251The response to a EmbedTextRequest.
repeated embeddings = 1
Output only. The embeddings generated from the input text.
rpc CountTextTokens (, )
text_service.proto:72Runs a model's tokenizer on a text and returns the token count.
message
text_service.proto:266Counts the number of tokens in the `prompt` sent to a model. Models may tokenize text differently, so each model may return a different `token_count`.
string model = 1
Required. The model's resource name. This serves as an ID for the Model to use. This name should match a model name returned by the `ListModels` method. Format: `models/{model}`
optional prompt = 2
Required. The free-form input text given to the model as a prompt.
message
text_service.proto:287A response from `CountTextTokens`. It returns the model's `token_count` for the `prompt`.
int32 token_count = 1
The number of tokens that the `model` tokenizes the `prompt` into. Always non-negative.
message
citation.proto:27A collection of source attributions for a piece of content.
Used in: ,
repeated citation_sources = 1
Citations to sources for a specific response.
message
citation.proto:33A citation to a source for a portion of a specific response.
Used in:
optional int32 start_index = 1
Optional. Start of segment of the response that is attributed to this source. Index indicates the start of the segment, measured in bytes.
optional int32 end_index = 2
Optional. End of the attributed segment, exclusive.
optional string uri = 3
Optional. URI that is attributed as a source for a portion of the text.
optional string license = 4
Optional. License for the GitHub project that is attributed as a source for segment. License info is required for code citations.
message
safety.proto:58Content filtering metadata associated with processing a single request. ContentFilter contains a reason and an optional supporting string. The reason may be unspecified.
Used in: ,
ContentFilter.BlockedReason reason = 1
The reason content was blocked during request processing.
optional string message = 2
A string that describes the filtering behavior in more detail.
A list of reasons why content may have been blocked.
Used in:
BLOCKED_REASON_UNSPECIFIED = 0
A blocked reason was not specified.
SAFETY = 1
Content was blocked by safety settings.
OTHER = 2
Content was blocked, but the reason is uncategorized.
message
model_service.proto:215Metadata about the state and progress of creating a tuned model returned from the long-running operation
string tuned_model = 5
Name of the tuned model associated with the tuning operation.
int32 total_steps = 1
The total number of tuning steps.
int32 completed_steps = 2
The number of steps completed.
float completed_percent = 3
The completed percentage for the tuning operation.
repeated snapshots = 4
Metrics collected during tuning.
message
tuned_model.proto:195Dataset for training or validation.
Used in:
oneof dataset
Inline data or a reference to the data.
examples = 1
Optional. Inline examples.
message
text_service.proto:257A list of floats representing the embedding.
Used in: ,
repeated float value = 1
The embedding values.
message
discuss_service.proto:208An input/output example used to instruct the Model. It demonstrates how the model should respond or format its response.
Used in:
optional input = 1
Required. An example of an input `Message` from the user.
optional output = 2
Required. An example of what the model should output given the input.
The category of a rating. These categories cover various kinds of harms that developers may wish to adjust.
Used in: ,
HARM_CATEGORY_UNSPECIFIED = 0
Category is unspecified.
HARM_CATEGORY_DEROGATORY = 1
Negative or harmful comments targeting identity and/or protected attribute.
HARM_CATEGORY_TOXICITY = 2
Content that is rude, disrepspectful, or profane.
HARM_CATEGORY_VIOLENCE = 3
Describes scenarios depictng violence against an individual or group, or general descriptions of gore.
HARM_CATEGORY_SEXUAL = 4
Contains references to sexual acts or other lewd content.
HARM_CATEGORY_MEDICAL = 5
Promotes unchecked medical advice.
HARM_CATEGORY_DANGEROUS = 6
Dangerous content that promotes, facilitates, or encourages harmful acts.
message
tuned_model.proto:178Hyperparameters controlling the tuning process.
Used in:
optional int32 epoch_count = 14
Immutable. The number of training epochs. An epoch is one pass through the training data. If not set, a default of 10 will be used.
optional int32 batch_size = 15
Immutable. The batch size hyperparameter for tuning. If not set, a default of 16 or 64 will be used based on the number of training examples.
optional float learning_rate = 16
Immutable. The learning rate hyperparameter for tuning. If not set, a default of 0.0002 or 0.002 will be calculated based on the number of training examples.
message
discuss_service.proto:137The base unit of structured text. A `Message` includes an `author` and the `content` of the `Message`. The `author` is used to tag messages when they are fed to the model as text.
Used in: , ,
string author = 1
Optional. The author of this Message. This serves as a key for tagging the content of this Message when it is fed to the model as text. The author can be any alphanumeric string.
string content = 2
Required. The text content of the structured `Message`.
optional citation_metadata = 3
Output only. Citation information for model-generated `content` in this `Message`. If this `Message` was generated as output from the model, this field may be populated with attribution information for any text included in the `content`. This field is used only on output.
message
discuss_service.proto:166All of the structured input text passed to the model as a prompt. A `MessagePrompt` contains a structured set of fields that provide context for the conversation, examples of user input/model output message pairs that prime the model to respond in different ways, and the conversation history or list of messages representing the alternating turns of the conversation between the user and the model.
Used in: ,
string context = 1
Optional. Text that should be provided to the model first to ground the response. If not empty, this `context` will be given to the model first before the `examples` and `messages`. When using a `context` be sure to provide it with every request to maintain continuity. This field can be a description of your prompt to the model to help provide context and guide the responses. Examples: "Translate the phrase from English to French." or "Given a statement, classify the sentiment as happy, sad or neutral." Anything included in this field will take precedence over message history if the total input size exceeds the model's `input_token_limit` and the input request is truncated.
repeated examples = 2
Optional. Examples of what the model should generate. This includes both user input and the response that the model should emulate. These `examples` are treated identically to conversation messages except that they take precedence over the history in `messages`: If the total input size exceeds the model's `input_token_limit` the input will be truncated. Items will be dropped from `messages` before `examples`.
repeated messages = 3
Required. A snapshot of the recent conversation history sorted chronologically. Turns alternate between two authors. If the total input size exceeds the model's `input_token_limit` the input will be truncated: The oldest items will be dropped from `messages`.
message
model.proto:28Information about a Generative Language Model.
Used as response type in: ModelService.GetModel
Used as field type in:
string name = 1
Required. The resource name of the `Model`. Format: `models/{model}` with a `{model}` naming convention of: * "{base_model_id}-{version}" Examples: * `models/chat-bison-001`
string base_model_id = 2
Required. The name of the base model, pass this to the generation request. Examples: * `chat-bison`
string version = 3
Required. The version number of the model. This represents the major version
string display_name = 4
The human-readable name of the model. E.g. "Chat Bison". The name can be up to 128 characters long and can consist of any UTF-8 characters.
string description = 5
A short description of the model.
int32 input_token_limit = 6
Maximum number of input tokens allowed for this model.
int32 output_token_limit = 7
Maximum number of output tokens available for this model.
repeated string supported_generation_methods = 8
The model's supported generation methods. The method names are defined as Pascal case strings, such as `generateMessage` which correspond to API methods.
optional float temperature = 9
Controls the randomness of the output. Values can range over `[0.0,1.0]`, inclusive. A value closer to `1.0` will produce responses that are more varied, while a value closer to `0.0` will typically result in less surprising responses from the model. This value specifies default to be used by the backend while making the call to the model.
optional float top_p = 10
For Nucleus sampling. Nucleus sampling considers the smallest set of tokens whose probability sum is at least `top_p`. This value specifies default to be used by the backend while making the call to the model.
optional int32 top_k = 11
For Top-k sampling. Top-k sampling considers the set of `top_k` most probable tokens. This value specifies default to be used by the backend while making the call to the model.
message
permission.proto:40Permission resource grants user, group or the rest of the world access to the PaLM API resource (e.g. a tuned model, file). A role is a collection of permitted operations that allows users to perform specific actions on PaLM API resources. To make them available to users, groups, or service accounts, you assign roles. When you assign a role, you grant permissions that the role contains. There are three concentric roles. Each role is a superset of the previous role's permitted operations: - reader can use the resource (e.g. tuned model) for inference - writer has reader's permissions and additionally can edit and share - owner has writer's permissions and additionally can delete
Used as response type in: PermissionService.CreatePermission, PermissionService.GetPermission, PermissionService.UpdatePermission
Used as field type in: , ,
string name = 1
Output only. The permission name. A unique name will be generated on create. Example: tunedModels/{tuned_model}permssions/{permission} Output only.
optional Permission.GranteeType grantee_type = 2
Required. Immutable. The type of the grantee.
optional string email_address = 3
Optional. Immutable. The email address of the user of group which this permission refers. Field is not set when permission's grantee type is EVERYONE.
optional Permission.Role role = 4
Required. The role granted by this permission.
Defines types of the grantee of this permission.
Used in:
GRANTEE_TYPE_UNSPECIFIED = 0
The default value. This value is unused.
USER = 1
Represents a user. When set, you must provide email_address for the user.
GROUP = 2
Represents a group. When set, you must provide email_address for the group.
EVERYONE = 3
Represents access to everyone. No extra information is required.
Defines the role granted by this permission.
Used in:
ROLE_UNSPECIFIED = 0
The default value. This value is unused.
OWNER = 1
Owner can use, update, share and delete the resource.
WRITER = 2
Writer can use, update and share the resource.
READER = 3
Reader can use the resource.
message
safety.proto:85Safety feedback for an entire request. This field is populated if content in the input and/or response is blocked due to safety settings. SafetyFeedback may not exist for every HarmCategory. Each SafetyFeedback will return the safety settings used by the request as well as the lowest HarmProbability that should be allowed in order to return a result.
Used in:
optional rating = 1
Safety rating evaluated from content.
optional setting = 2
Safety settings applied to the request.
message
safety.proto:100Safety rating for a piece of content. The safety rating contains the category of harm and the harm probability level in that category for a piece of content. Content is classified for safety across a number of harm categories and the probability of the harm classification is included here.
Used in: ,
HarmCategory category = 3
Required. The category for this rating.
SafetyRating.HarmProbability probability = 4
Required. The probability of harm for this content.
The probability that a piece of content is harmful. The classification system gives the probability of the content being unsafe. This does not indicate the severity of harm for a piece of content.
Used in:
HARM_PROBABILITY_UNSPECIFIED = 0
Probability is unspecified.
NEGLIGIBLE = 1
Content has a negligible chance of being unsafe.
LOW = 2
Content has a low chance of being unsafe.
MEDIUM = 3
Content has a medium chance of being unsafe.
HIGH = 4
Content has a high chance of being unsafe.
message
safety.proto:133Safety setting, affecting the safety-blocking behavior. Passing a safety setting for a category changes the allowed proability that content is blocked.
Used in: ,
HarmCategory category = 3
Required. The category for this setting.
SafetySetting.HarmBlockThreshold threshold = 4
Required. Controls the probability threshold at which harm is blocked.
Block at and beyond a specified harm probability.
Used in:
HARM_BLOCK_THRESHOLD_UNSPECIFIED = 0
Threshold is unspecified.
BLOCK_LOW_AND_ABOVE = 1
Content with NEGLIGIBLE will be allowed.
BLOCK_MEDIUM_AND_ABOVE = 2
Content with NEGLIGIBLE and LOW will be allowed.
BLOCK_ONLY_HIGH = 3
Content with NEGLIGIBLE, LOW, and MEDIUM will be allowed.
BLOCK_NONE = 4
All content will be allowed.
message
text_service.proto:193Output text returned from a model.
Used in:
string output = 1
Output only. The generated text returned from the model.
repeated safety_ratings = 2
Ratings for the safety of a response. There is at most one rating per category.
optional citation_metadata = 3
Output only. Citation information for model-generated `output` in this `TextCompletion`. This field may be populated with attribution information for any text included in the `output`.
message
text_service.proto:187Text given to the model as a prompt. The Model will use this TextPrompt to Generate a text completion.
Used in: ,
string text = 1
Required. The prompt text.
message
tuned_model.proto:29A fine-tuned model created using ModelService.CreateTunedModel.
Used as response type in: ModelService.GetTunedModel, ModelService.UpdateTunedModel
Used as field type in: , ,
oneof source_model
The model used as the starting point for tuning.
tuned_model_source = 3
Optional. TunedModel to use as the starting point for training the new model.
string base_model = 4
Immutable. The name of the `Model` to tune. Example: `models/text-bison-001`
string name = 1
Output only. The tuned model name. A unique name will be generated on create. Example: `tunedModels/az2mb0bpw6i` If display_name is set on create, the id portion of the name will be set by concatenating the words of the display_name with hyphens and adding a random portion for uniqueness. Example: display_name = "Sentence Translator" name = "tunedModels/sentence-translator-u3b7m"
string display_name = 5
Optional. The name to display for this model in user interfaces. The display name must be up to 40 characters including spaces.
string description = 6
Optional. A short description of this model.
optional float temperature = 11
Optional. Controls the randomness of the output. Values can range over `[0.0,1.0]`, inclusive. A value closer to `1.0` will produce responses that are more varied, while a value closer to `0.0` will typically result in less surprising responses from the model. This value specifies default to be the one used by the base model while creating the model.
optional float top_p = 12
Optional. For Nucleus sampling. Nucleus sampling considers the smallest set of tokens whose probability sum is at least `top_p`. This value specifies default to be the one used by the base model while creating the model.
optional int32 top_k = 13
Optional. For Top-k sampling. Top-k sampling considers the set of `top_k` most probable tokens. This value specifies default to be used by the backend while making the call to the model. This value specifies default to be the one used by the base model while creating the model.
TunedModel.State state = 7
Output only. The state of the tuned model.
optional create_time = 8
Output only. The timestamp when this model was created.
optional update_time = 9
Output only. The timestamp when this model was updated.
optional tuning_task = 10
Required. The tuning task that creates the tuned model.
The state of the tuned model.
Used in:
STATE_UNSPECIFIED = 0
The default value. This value is unused.
CREATING = 1
The model is being created.
ACTIVE = 2
The model is ready to be used.
FAILED = 3
The model failed to be created.
message
tuned_model.proto:130Tuned model as a source for training a new model.
Used in:
string tuned_model = 1
Immutable. The name of the `TunedModel` to use as the starting point for training the new model. Example: `tunedModels/my-tuned-model`
string base_model = 2
Output only. The name of the base `Model` this `TunedModel` was tuned from. Example: `models/text-bison-001`
message
tuned_model.proto:211A single example for tuning.
Used in:
oneof model_input
The input to the model for this example.
string text_input = 1
Optional. Text model input.
string output = 3
Required. The expected model output.
message
tuned_model.proto:204A set of tuning examples. Can be training or validatation data.
Used in:
repeated examples = 1
Required. The examples. Example input can be for text or discuss, but all examples in a set must be of the same type.
message
tuned_model.proto:223Record for a single tuning step.
Used in: ,
int32 step = 1
Output only. The tuning step.
int32 epoch = 2
Output only. The epoch this step was part of.
float mean_loss = 3
Output only. The mean loss of the training examples for this step.
optional compute_time = 4
Output only. The timestamp when this metric was computed.
message
tuned_model.proto:152Tuning tasks that create tuned models.
Used in:
optional start_time = 1
Output only. The timestamp when tuning this model started.
optional complete_time = 2
Output only. The timestamp when tuning this model completed.
repeated snapshots = 3
Output only. Metrics collected during tuning.
optional training_data = 4
Required. Input only. Immutable. The model training data.
optional hyperparameters = 5
Immutable. Hyperparameters controlling the tuning process. If not provided, default values will be used.