Add frequencePenalty and presencePenalty parameters support for gener… (

#264) * Add frequencePenalty and presencePenalty parameters support for generate content. * Change frequence to frequency * Update format * Update parameter description * Update the parameter doc.
google-gemini · Sep 25, 2024 · dda0b5c · dda0b5c
1 parent 85d1eb1
commit dda0b5c
Show file tree

Hide file tree

Showing 12 changed files with 97 additions and 2 deletions.
diff --git a/.changeset/cyan-pants-move.md b/.changeset/cyan-pants-move.md
@@ -0,0 +1,5 @@
+---
+"@google/generative-ai": minor
+---
+
+Add `frequencyPenalty` and `presencePenalty` parameters support for `generateContent()`
diff --git a/common/api-review/generative-ai.api.md b/common/api-review/generative-ai.api.md
@@ -6,8 +6,10 @@
 
 // @public
 export interface BaseParams {
+    frequencyPenalty?: number;
     // (undocumented)
     generationConfig?: GenerationConfig;
+    presencePenalty?: number;
     // (undocumented)
     safetySettings?: SafetySetting[];
 }
@@ -458,13 +460,17 @@ export class GenerativeModel {
     cachedContent: CachedContent;
     countTokens(request: CountTokensRequest | string | Array<string | Part>, requestOptions?: SingleRequestOptions): Promise<CountTokensResponse>;
     embedContent(request: EmbedContentRequest | string | Array<string | Part>, requestOptions?: SingleRequestOptions): Promise<EmbedContentResponse>;
+    // (undocumented)
+    frequencyPenalty?: number;
     generateContent(request: GenerateContentRequest | string | Array<string | Part>, requestOptions?: SingleRequestOptions): Promise<GenerateContentResult>;
     generateContentStream(request: GenerateContentRequest | string | Array<string | Part>, requestOptions?: SingleRequestOptions): Promise<GenerateContentStreamResult>;
     // (undocumented)
     generationConfig: GenerationConfig;
     // (undocumented)
     model: string;
     // (undocumented)
+    presencePenalty?: number;
+    // (undocumented)
     safetySettings: SafetySetting[];
     startChat(startChatParams?: StartChatParams): ChatSession;
     // (undocumented)

diff --git a/docs/reference/main/generative-ai.baseparams.frequencypenalty.md b/docs/reference/main/generative-ai.baseparams.frequencypenalty.md
@@ -0,0 +1,13 @@
+<!-- Do not edit this file. It is automatically generated by API Documenter. -->
+
+[Home](./index.md) &gt; [@google/generative-ai](./generative-ai.md) &gt; [BaseParams](./generative-ai.baseparams.md) &gt; [frequencyPenalty](./generative-ai.baseparams.frequencypenalty.md)
+
+## BaseParams.frequencyPenalty property
+
+Frequency penalty applied to the next token's logprobs, multiplied by the number of times each token has been seen in the respponse so far.
+
+**Signature:**
+
+```typescript
+frequencyPenalty?: number;
+```
diff --git a/docs/reference/main/generative-ai.baseparams.md b/docs/reference/main/generative-ai.baseparams.md
@@ -16,6 +16,8 @@ export interface BaseParams
 
 |  Property | Modifiers | Type | Description |
 |  --- | --- | --- | --- |
+|  [frequencyPenalty?](./generative-ai.baseparams.frequencypenalty.md) |  | number | _(Optional)_ Frequency penalty applied to the next token's logprobs, multiplied by the number of times each token has been seen in the respponse so far. |
 |  [generationConfig?](./generative-ai.baseparams.generationconfig.md) |  | [GenerationConfig](./generative-ai.generationconfig.md) | _(Optional)_ |
+|  [presencePenalty?](./generative-ai.baseparams.presencepenalty.md) |  | number | _(Optional)_ Presence penalty applied to the next token's logprobs if the token has already been seen in the response. |
 |  [safetySettings?](./generative-ai.baseparams.safetysettings.md) |  | [SafetySetting](./generative-ai.safetysetting.md)<!-- -->\[\] | _(Optional)_ |
 
diff --git a/docs/reference/main/generative-ai.baseparams.presencepenalty.md b/docs/reference/main/generative-ai.baseparams.presencepenalty.md
@@ -0,0 +1,13 @@
+<!-- Do not edit this file. It is automatically generated by API Documenter. -->
+
+[Home](./index.md) &gt; [@google/generative-ai](./generative-ai.md) &gt; [BaseParams](./generative-ai.baseparams.md) &gt; [presencePenalty](./generative-ai.baseparams.presencepenalty.md)
+
+## BaseParams.presencePenalty property
+
+Presence penalty applied to the next token's logprobs if the token has already been seen in the response.
+
+**Signature:**
+
+```typescript
+presencePenalty?: number;
+```
diff --git a/docs/reference/main/generative-ai.generativemodel.frequencypenalty.md b/docs/reference/main/generative-ai.generativemodel.frequencypenalty.md
@@ -0,0 +1,11 @@
+<!-- Do not edit this file. It is automatically generated by API Documenter. -->
+
+[Home](./index.md) &gt; [@google/generative-ai](./generative-ai.md) &gt; [GenerativeModel](./generative-ai.generativemodel.md) &gt; [frequencyPenalty](./generative-ai.generativemodel.frequencypenalty.md)
+
+## GenerativeModel.frequencyPenalty property
+
+**Signature:**
+
+```typescript
+frequencyPenalty?: number;
+```
diff --git a/docs/reference/main/generative-ai.generativemodel.md b/docs/reference/main/generative-ai.generativemodel.md
@@ -24,8 +24,10 @@ export declare class GenerativeModel
 |  --- | --- | --- | --- |
 |  [apiKey](./generative-ai.generativemodel.apikey.md) |  | string |  |
 |  [cachedContent](./generative-ai.generativemodel.cachedcontent.md) |  | [CachedContent](./generative-ai.cachedcontent.md) |  |
+|  [frequencyPenalty?](./generative-ai.generativemodel.frequencypenalty.md) |  | number | _(Optional)_ |
 |  [generationConfig](./generative-ai.generativemodel.generationconfig.md) |  | [GenerationConfig](./generative-ai.generationconfig.md) |  |
 |  [model](./generative-ai.generativemodel.model.md) |  | string |  |
+|  [presencePenalty?](./generative-ai.generativemodel.presencepenalty.md) |  | number | _(Optional)_ |
 |  [safetySettings](./generative-ai.generativemodel.safetysettings.md) |  | [SafetySetting](./generative-ai.safetysetting.md)<!-- -->\[\] |  |
 |  [systemInstruction?](./generative-ai.generativemodel.systeminstruction.md) |  | [Content](./generative-ai.content.md) | _(Optional)_ |
 |  [toolConfig?](./generative-ai.generativemodel.toolconfig.md) |  | [ToolConfig](./generative-ai.toolconfig.md) | _(Optional)_ |

diff --git a/docs/reference/main/generative-ai.generativemodel.presencepenalty.md b/docs/reference/main/generative-ai.generativemodel.presencepenalty.md
@@ -0,0 +1,11 @@
+<!-- Do not edit this file. It is automatically generated by API Documenter. -->
+
+[Home](./index.md) &gt; [@google/generative-ai](./generative-ai.md) &gt; [GenerativeModel](./generative-ai.generativemodel.md) &gt; [presencePenalty](./generative-ai.generativemodel.presencepenalty.md)
+
+## GenerativeModel.presencePenalty property
+
+**Signature:**
+
+```typescript
+presencePenalty?: number;
+```
diff --git a/src/methods/generate-content.test.ts b/src/methods/generate-content.test.ts
@@ -42,6 +42,8 @@ const fakeRequestParams: GenerateContentRequest = {
       threshold: HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
     },
   ],
+  presencePenalty: 0.5,
+  frequencyPenalty: 0.1,
 };
 
 describe("generateContent()", () => {

diff --git a/src/models/generative-model.test.ts b/src/models/generative-model.test.ts
@@ -71,6 +71,8 @@ describe("GenerativeModel", () => {
             threshold: HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
           },
         ],
+        presencePenalty: 0.6,
+        frequencyPenalty: 0.5,
         tools: [{ functionDeclarations: [{ name: "myfunc" }] }],
         toolConfig: {
           functionCallingConfig: { mode: FunctionCallingMode.NONE },
@@ -92,6 +94,8 @@ describe("GenerativeModel", () => {
       genModel.generationConfig?.responseSchema.properties.testField.type,
     ).to.equal(SchemaType.STRING);
     expect(genModel.safetySettings?.length).to.equal(1);
+    expect(genModel.presencePenalty).to.equal(0.6);
+    expect(genModel.frequencyPenalty).to.equal(0.5);
     expect(genModel.tools?.length).to.equal(1);
     expect(genModel.toolConfig?.functionCallingConfig.mode).to.equal(
       FunctionCallingMode.NONE,
@@ -116,7 +120,9 @@ describe("GenerativeModel", () => {
           value.includes("be friendly") &&
           value.includes("temperature") &&
           value.includes("testField") &&
-          value.includes(HarmBlockThreshold.BLOCK_LOW_AND_ABOVE)
+          value.includes(HarmBlockThreshold.BLOCK_LOW_AND_ABOVE) &&
+          value.includes("presencePenalty") &&
+          value.includes("frequencyPenalty")
         );
       }),
       match((value) => {
@@ -210,6 +216,8 @@ describe("GenerativeModel", () => {
           threshold: HarmBlockThreshold.BLOCK_NONE,
         },
       ],
+      presencePenalty: 0.6,
+      frequencyPenalty: 0.5,
       contents: [{ role: "user", parts: [{ text: "hello" }] }],
       tools: [{ functionDeclarations: [{ name: "otherfunc" }] }],
       toolConfig: { functionCallingConfig: { mode: FunctionCallingMode.AUTO } },
@@ -228,7 +236,9 @@ describe("GenerativeModel", () => {
           value.includes("topK") &&
           value.includes("newTestField") &&
           !value.includes("testField") &&
-          value.includes(HarmCategory.HARM_CATEGORY_HARASSMENT)
+          value.includes(HarmCategory.HARM_CATEGORY_HARASSMENT) &&
+          value.includes("presencePenalty") &&
+          value.includes("frequencyPenalty")
         );
       }),
       {},

diff --git a/src/models/generative-model.ts b/src/models/generative-model.ts
@@ -63,6 +63,8 @@ export class GenerativeModel {
   toolConfig?: ToolConfig;
   systemInstruction?: Content;
   cachedContent: CachedContent;
+  presencePenalty?: number;
+  frequencyPenalty?: number;
 
   constructor(
     public apiKey: string,
@@ -84,6 +86,8 @@ export class GenerativeModel {
       modelParams.systemInstruction,
     );
     this.cachedContent = modelParams.cachedContent;
+    this.presencePenalty = modelParams.presencePenalty;
+    this.frequencyPenalty = modelParams.frequencyPenalty;
   }
 
   /**
@@ -113,6 +117,8 @@ export class GenerativeModel {
         toolConfig: this.toolConfig,
         systemInstruction: this.systemInstruction,
         cachedContent: this.cachedContent?.name,
+        presencePenalty: this.presencePenalty,
+        frequencyPenalty: this.frequencyPenalty,
         ...formattedParams,
       },
       generativeModelRequestOptions,
@@ -148,6 +154,8 @@ export class GenerativeModel {
         toolConfig: this.toolConfig,
         systemInstruction: this.systemInstruction,
         cachedContent: this.cachedContent?.name,
+        presencePenalty: this.presencePenalty,
+        frequencyPenalty: this.frequencyPenalty,
         ...formattedParams,
       },
       generativeModelRequestOptions,
@@ -194,6 +202,8 @@ export class GenerativeModel {
       toolConfig: this.toolConfig,
       systemInstruction: this.systemInstruction,
       cachedContent: this.cachedContent,
+      presencePenalty: this.presencePenalty,
+      frequencyPenalty: this.frequencyPenalty,
     });
     const generativeModelRequestOptions: SingleRequestOptions = {
       ...this._requestOptions,

diff --git a/types/requests.ts b/types/requests.ts
@@ -31,6 +31,16 @@ import {
 export interface BaseParams {
   safetySettings?: SafetySetting[];
   generationConfig?: GenerationConfig;
+  /**
+   * Presence penalty applied to the next token's logprobs if the token has
+   * already been seen in the response.
+   */
+  presencePenalty?: number
+  /**
+   * Frequency penalty applied to the next token's logprobs, multiplied by the
+   * number of times each token has been seen in the respponse so far.
+   */
+  frequencyPenalty?: number
 }
 
 /**