Skip to content

Commit 8328630

Browse files
andrewheardandrewheard
authored
[Vertex AI] Make GenerationConfig properties internal (#13904)
1 parent c249369 commit 8328630

File tree

3 files changed

+110
-89
lines changed

3 files changed

+110
-89
lines changed

‎FirebaseVertexAI/CHANGELOG.md

+3
Original file line numberDiff line numberDiff line change
@@ -55,6 +55,9 @@
5555
- [changed] **Breaking Change**: The minimum deployment target for the SDK is
5656
now macOS 12.0; all other platform minimums remain the same at iOS 15.0,
5757
macCatalyst 15.0, tvOS 15.0, and watchOS 8.0. (#13903)
58+
- [changed] **Breaking Change**: All of the public properties of
59+
`GenerationConfig` are now `internal`; they all remain configurable in the
60+
initializer. (#13904)
5861
- [changed] The default request timeout is now 180 seconds instead of the
5962
platform-default value of 60 seconds for a `URLRequest`; this timeout may
6063
still be customized in `RequestOptions`. (#13722)

‎FirebaseVertexAI/Sources/GenerateContentResponse.swift

+2-2
Original file line numberDiff line numberDiff line change
@@ -172,8 +172,8 @@ public struct FinishReason: DecodableProtoEnum, Hashable, Sendable {
172172

173173
/// The token generation was stopped because the response was flagged for safety reasons.
174174
///
175-
/// > NOTE: When streaming, the ``CandidateResponse/content`` will be empty if content filters
176-
/// > blocked the output.
175+
/// > NOTE: When streaming, the ``Candidate/content`` will be empty if content filters blocked the
176+
/// > output.
177177
public static let safety = FinishReason(kind: .safety)
178178

179179
/// The token generation was stopped because the response was flagged for unauthorized citations.

‎FirebaseVertexAI/Sources/GenerationConfig.swift

+105-87
Original file line numberDiff line numberDiff line change
@@ -18,109 +18,127 @@ import Foundation
1818
/// requests to the backend model.
1919
@available(iOS 15.0, macOS 12.0, macCatalyst 15.0, tvOS 15.0, watchOS 8.0, *)
2020
public struct GenerationConfig {
21-
/// A parameter controlling the degree of randomness in token selection. A
22-
/// temperature of zero is deterministic, always choosing the
23-
/// highest-probability response. Typical values are between 0 and 1
24-
/// inclusive. Defaults to 0 if unspecified.
25-
public let temperature: Float?
21+
/// Controls the degree of randomness in token selection.
22+
let temperature: Float?
2623

27-
/// The `topP` parameter changes how the model selects tokens for output.
28-
/// Tokens are selected from the most to least probable until the sum of
29-
/// their probabilities equals the `topP` value. For example, if tokens A, B,
30-
/// and C have probabilities of 0.3, 0.2, and 0.1 respectively and the topP
31-
/// value is 0.5, then the model will select either A or B as the next token
32-
/// by using the `temperature` and exclude C as a candidate.
33-
/// Defaults to 0.95 if unset.
34-
public let topP: Float?
24+
/// Controls diversity of generated text.
25+
let topP: Float?
3526

36-
/// The `topK` parameter changes how the model selects tokens for output. A
37-
/// `topK` of 1 means the selected token is the most probable among all the
38-
/// tokens in the model's vocabulary, while a `topK` of 3 means that the next
39-
/// token is selected from among the 3 most probable using the `temperature`.
40-
/// For each token selection step, the `topK` tokens with the highest
41-
/// probabilities are sampled. Tokens are then further filtered based on
42-
/// `topP` with the final token selected using `temperature` sampling.
43-
/// Defaults to 40 if unspecified.
44-
public let topK: Int?
27+
/// Limits the number of highest probability words considered.
28+
let topK: Int?
4529

46-
/// The maximum number of generated response messages to return. This value
47-
/// must be between [1, 8], inclusive. If unset, this will default to 1.
48-
///
49-
/// - Note: Only unique candidates are returned. Higher temperatures are more
50-
/// likely to produce unique candidates. Setting `temperature` to 0 will
51-
/// always produce exactly one candidate regardless of the
52-
/// `candidateCount`.
53-
public let candidateCount: Int?
30+
/// The number of response variations to return.
31+
let candidateCount: Int?
5432

55-
/// Specifies the maximum number of tokens that can be generated in the
56-
/// response. The number of tokens per word varies depending on the
57-
/// language outputted. The maximum value is capped at 1024. Defaults to 0
58-
/// (unbounded).
59-
public let maxOutputTokens: Int?
33+
/// Maximum number of tokens that can be generated in the response.
34+
let maxOutputTokens: Int?
6035

6136
/// Controls the likelihood of repeating the same words or phrases already generated in the text.
62-
///
63-
/// Higher values increase the penalty of repetition, resulting in more diverse output. The
64-
/// maximum value for `presencePenalty` is up to, but not including, `2.0`; the minimum value is
65-
/// `-2.0`.
66-
///
67-
/// > Note: While both `presencePenalty` and ``frequencyPenalty`` discourage repetition,
68-
/// > `presencePenalty` applies the same penalty regardless of how many times the word/phrase has
69-
/// > already appeared, whereas `frequencyPenalty` increases the penalty for *each* repetition of
70-
/// > a word/phrase.
71-
///
72-
/// > Important: Supported by `gemini-1.5-pro-002` and` gemini-1.5-flash-002` only.
73-
public let presencePenalty: Float?
37+
let presencePenalty: Float?
7438

7539
/// Controls the likelihood of repeating words, with the penalty increasing for each repetition.
76-
///
77-
/// Higher values increase the penalty of repetition, resulting in more diverse output. The
78-
/// maximum value for `frequencyPenalty` is up to, but not including, `2.0`; the minimum value is
79-
/// `-2.0`.
80-
///
81-
/// > Note: While both `frequencyPenalty` and ``presencePenalty`` discourage repetition,
82-
/// > `frequencyPenalty` increases the penalty for *each* repetition of a word/phrase, whereas
83-
/// > `presencePenalty` applies the same penalty regardless of how many times the word/phrase has
84-
/// > already appeared.
85-
///
86-
/// > Important: Supported by `gemini-1.5-pro-002` and` gemini-1.5-flash-002` only.
87-
public let frequencyPenalty: Float?
40+
let frequencyPenalty: Float?
8841

89-
/// A set of up to 5 `String`s that will stop output generation. If
90-
/// specified, the API will stop at the first appearance of a stop sequence.
91-
/// The stop sequence will not be included as part of the response.
92-
public let stopSequences: [String]?
42+
/// A set of up to 5 `String`s that will stop output generation.
43+
let stopSequences: [String]?
9344

9445
/// Output response MIME type of the generated candidate text.
95-
///
96-
/// Supported MIME types:
97-
/// - `text/plain`: Text output; the default behavior if unspecified.
98-
/// - `application/json`: JSON response in the candidates.
99-
public let responseMIMEType: String?
46+
let responseMIMEType: String?
10047

10148
/// Output schema of the generated candidate text.
102-
/// If set, a compatible ``responseMIMEType`` must also be set.
103-
///
104-
/// Compatible MIME types:
105-
/// - `application/json`: Schema for JSON response.
106-
///
107-
/// Refer to the [Control generated
108-
/// output](https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/control-generated-output)
109-
/// guide for more details.
110-
public let responseSchema: Schema?
49+
let responseSchema: Schema?
11150

11251
/// Creates a new `GenerationConfig` value.
11352
///
114-
/// - Parameter temperature: See ``temperature``
115-
/// - Parameter topP: See ``topP``
116-
/// - Parameter topK: See ``topK``
117-
/// - Parameter candidateCount: See ``candidateCount``
118-
/// - Parameter maxOutputTokens: See ``maxOutputTokens``
119-
/// - Parameter presencePenalty: See ``presencePenalty``
120-
/// - Parameter frequencyPenalty: See ``frequencyPenalty``
121-
/// - Parameter stopSequences: See ``stopSequences``
122-
/// - Parameter responseMIMEType: See ``responseMIMEType``
123-
/// - Parameter responseSchema: See ``responseSchema``
53+
/// See the
54+
/// [Configure model parameters](https://firebase.google.com/docs/vertex-ai/model-parameters)
55+
/// guide and the
56+
/// [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
57+
/// for more details.
58+
///
59+
/// - Parameters:
60+
/// - temperature:Controls the randomness of the language model's output. Higher values (for
61+
/// example, 1.0) make the text more random and creative, while lower values (for example,
62+
/// 0.1) make it more focused and deterministic.
63+
///
64+
/// > Note: A temperature of 0 means that the highest probability tokens are always selected.
65+
/// > In this case, responses for a given prompt are mostly deterministic, but a small amount
66+
/// > of variation is still possible.
67+
///
68+
/// > Important: The range of supported temperature values depends on the model; see the
69+
/// > [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
70+
/// > for more details.
71+
/// - topP: Controls diversity of generated text. Higher values (e.g., 0.9) produce more diverse
72+
/// text, while lower values (e.g., 0.5) make the output more focused.
73+
///
74+
/// The supported range is 0.0 to 1.0.
75+
///
76+
/// > Important: The default `topP` value depends on the model; see the
77+
/// [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
78+
/// for more details.
79+
/// - topK: Limits the number of highest probability words the model considers when generating
80+
/// text. For example, a topK of 40 means only the 40 most likely words are considered for the
81+
/// next token. A higher value increases diversity, while a lower value makes the output more
82+
/// deterministic.
83+
///
84+
/// The supported range is 1 to 40.
85+
///
86+
/// > Important: Support for `topK` and the default value depends on the model; see the
87+
/// [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
88+
/// for more details.
89+
/// - candidateCount: The number of response variations to return; defaults to 1 if not set.
90+
/// Support for multiple candidates depends on the model; see the
91+
/// [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
92+
/// for more details.
93+
/// - maxOutputTokens: Maximum number of tokens that can be generated in the response.
94+
/// See the configure model parameters [documentation](https://firebase.google.com/docs/vertex-ai/model-parameters?platform=ios#max-output-tokens)
95+
/// for more details.
96+
/// - presencePenalty: Controls the likelihood of repeating the same words or phrases already
97+
/// generated in the text. Higher values increase the penalty of repetition, resulting in more
98+
/// diverse output.
99+
///
100+
/// > Note: While both `presencePenalty` and `frequencyPenalty` discourage repetition,
101+
/// > `presencePenalty` applies the same penalty regardless of how many times the word/phrase
102+
/// > has already appeared, whereas `frequencyPenalty` increases the penalty for *each*
103+
/// > repetition of a word/phrase.
104+
///
105+
/// > Important: The range of supported `presencePenalty` values depends on the model; see the
106+
/// > [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
107+
/// > for more details
108+
/// - frequencyPenalty: Controls the likelihood of repeating words or phrases, with the penalty
109+
/// increasing for each repetition. Higher values increase the penalty of repetition,
110+
/// resulting in more diverse output.
111+
///
112+
/// > Note: While both `frequencyPenalty` and `presencePenalty` discourage repetition,
113+
/// > `frequencyPenalty` increases the penalty for *each* repetition of a word/phrase, whereas
114+
/// > `presencePenalty` applies the same penalty regardless of how many times the word/phrase
115+
/// > has already appeared.
116+
///
117+
/// > Important: The range of supported `frequencyPenalty` values depends on the model; see
118+
/// > the
119+
/// > [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
120+
/// > for more details
121+
/// - stopSequences: A set of up to 5 `String`s that will stop output generation. If specified,
122+
/// the API will stop at the first appearance of a stop sequence. The stop sequence will not
123+
/// be included as part of the response. See the
124+
/// [Cloud documentation](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#generationconfig)
125+
/// for more details.
126+
/// - responseMIMEType: Output response MIME type of the generated candidate text.
127+
///
128+
/// Supported MIME types:
129+
/// - `text/plain`: Text output; the default behavior if unspecified.
130+
/// - `application/json`: JSON response in the candidates.
131+
/// - `text/x.enum`: For classification tasks, output an enum value as defined in the
132+
/// `responseSchema`.
133+
/// - responseSchema: Output schema of the generated candidate text. If set, a compatible
134+
/// `responseMIMEType` must also be set.
135+
///
136+
/// Compatible MIME types:
137+
/// - `application/json`: Schema for JSON response.
138+
///
139+
/// Refer to the
140+
/// [Control generated output](https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/control-generated-output)
141+
/// guide for more details.
124142
public init(temperature: Float? = nil, topP: Float? = nil, topK: Int? = nil,
125143
candidateCount: Int? = nil, maxOutputTokens: Int? = nil,
126144
presencePenalty: Float? = nil, frequencyPenalty: Float? = nil,

0 commit comments

Comments
 (0)