modelfusion 0.116.1 → 0.118.0

package/model-provider/huggingface/HuggingFaceTextGenerationModel.d.ts

@@ -43,7 +43,7 @@ export declare class HuggingFaceTextGenerationModel extends AbstractModel<Huggin
     callAPI(prompt: string, callOptions: FunctionCallOptions): Promise<HuggingFaceTextGenerationResponse>;
     get settingsForEvent(): Partial<HuggingFaceTextGenerationModelSettings>;
     doGenerateTexts(prompt: string, options: FunctionCallOptions): Promise<{
-        response: {
+        rawResponse: {
             generated_text: string;
         }[];
         textGenerationResults: {
@@ -52,7 +52,7 @@ export declare class HuggingFaceTextGenerationModel extends AbstractModel<Huggin
         }[];
     }>;
     restoreGeneratedTexts(rawResponse: unknown): {
-        response: {
+        rawResponse: {
             generated_text: string;
         }[];
         textGenerationResults: {
@@ -60,8 +60,8 @@ export declare class HuggingFaceTextGenerationModel extends AbstractModel<Huggin
             finishReason: "unknown";
         }[];
     };
-    processTextGenerationResponse(response: HuggingFaceTextGenerationResponse): {
-        response: {
+    processTextGenerationResponse(rawResponse: HuggingFaceTextGenerationResponse): {
+        rawResponse: {
             generated_text: string;
         }[];
         textGenerationResults: {

package/model-provider/huggingface/HuggingFaceTextGenerationModel.js

@@ -113,10 +113,10 @@ export class HuggingFaceTextGenerationModel extends AbstractModel {
             schema: zodSchema(huggingFaceTextGenerationResponseSchema),
         }));
     }
-    processTextGenerationResponse(response) {
+    processTextGenerationResponse(rawResponse) {
         return {
-            response,
-            textGenerationResults: response.map((response) => ({
+            rawResponse,
+            textGenerationResults: rawResponse.map((response) => ({
                 text: response.generated_text,
                 finishReason: "unknown",
             })),

package/model-provider/llamacpp/LlamaCppBakLLaVA1PromptTemplate.cjs

@@ -1,11 +1,25 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.chat = exports.instruction = void 0;
+exports.chat = exports.instruction = exports.text = void 0;
 const ContentPart_js_1 = require("../../model-function/generate-text/prompt-template/ContentPart.cjs");
 const InvalidPromptError_js_1 = require("../../model-function/generate-text/prompt-template/InvalidPromptError.cjs");
+const TextPromptTemplate_js_1 = require("../../model-function/generate-text/prompt-template/TextPromptTemplate.cjs");
 // default Vicuna 1 system message
 const DEFAULT_SYSTEM_MESSAGE = "A chat between a curious user and an artificial intelligence assistant. " +
     "The assistant gives helpful, detailed, and polite answers to the user's questions.";
+/**
+ * Text prompt.
+ */
+function text() {
+    const delegate = (0, TextPromptTemplate_js_1.text)();
+    return {
+        stopSequences: [],
+        format(prompt) {
+            return { text: delegate.format(prompt) };
+        },
+    };
+}
+exports.text = text;
 /**
  * BakLLaVA 1 uses a Vicuna 1 prompt. This mapping combines it with the LlamaCpp prompt structure.
  *

package/model-provider/llamacpp/LlamaCppBakLLaVA1PromptTemplate.d.ts

@@ -2,6 +2,10 @@ import { TextGenerationPromptTemplate } from "../../model-function/generate-text
 import { ChatPrompt } from "../../model-function/generate-text/prompt-template/ChatPrompt.js";
 import { InstructionPrompt } from "../../model-function/generate-text/prompt-template/InstructionPrompt.js";
 import { LlamaCppCompletionPrompt } from "./LlamaCppCompletionModel.js";
+/**
+ * Text prompt.
+ */
+export declare function text(): TextGenerationPromptTemplate<string, LlamaCppCompletionPrompt>;
 /**
  * BakLLaVA 1 uses a Vicuna 1 prompt. This mapping combines it with the LlamaCpp prompt structure.
  *

package/model-provider/llamacpp/LlamaCppBakLLaVA1PromptTemplate.js

@@ -1,8 +1,21 @@
 import { validateContentIsString } from "../../model-function/generate-text/prompt-template/ContentPart.js";
 import { InvalidPromptError } from "../../model-function/generate-text/prompt-template/InvalidPromptError.js";
+import { text as vicunaText } from "../../model-function/generate-text/prompt-template/TextPromptTemplate.js";
 // default Vicuna 1 system message
 const DEFAULT_SYSTEM_MESSAGE = "A chat between a curious user and an artificial intelligence assistant. " +
     "The assistant gives helpful, detailed, and polite answers to the user's questions.";
+/**
+ * Text prompt.
+ */
+export function text() {
+    const delegate = vicunaText();
+    return {
+        stopSequences: [],
+        format(prompt) {
+            return { text: delegate.format(prompt) };
+        },
+    };
+}
 /**
  * BakLLaVA 1 uses a Vicuna 1 prompt. This mapping combines it with the LlamaCpp prompt structure.
  *

package/model-provider/llamacpp/LlamaCppCompletionModel.cjs

@@ -8,12 +8,15 @@ const ZodSchema_js_1 = require("../../core/schema/ZodSchema.cjs");
 const parseJSON_js_1 = require("../../core/schema/parseJSON.cjs");
 const validateTypes_js_1 = require("../../core/schema/validateTypes.cjs");
 const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
+const StructureFromTextStreamingModel_js_1 = require("../../model-function/generate-structure/StructureFromTextStreamingModel.cjs");
 const PromptTemplateTextStreamingModel_js_1 = require("../../model-function/generate-text/PromptTemplateTextStreamingModel.cjs");
 const TextGenerationModel_js_1 = require("../../model-function/generate-text/TextGenerationModel.cjs");
 const AsyncQueue_js_1 = require("../../util/AsyncQueue.cjs");
 const parseEventSourceStream_js_1 = require("../../util/streaming/parseEventSourceStream.cjs");
 const LlamaCppApiConfiguration_js_1 = require("./LlamaCppApiConfiguration.cjs");
 const LlamaCppError_js_1 = require("./LlamaCppError.cjs");
+const LlamaCppGrammars_js_1 = require("./LlamaCppGrammars.cjs");
+const LlamaCppPrompt_js_1 = require("./LlamaCppPrompt.cjs");
 const LlamaCppTokenizer_js_1 = require("./LlamaCppTokenizer.cjs");
 class LlamaCppCompletionModel extends AbstractModel_js_1.AbstractModel {
     constructor(settings = {}) {
@@ -139,23 +142,23 @@ class LlamaCppCompletionModel extends AbstractModel_js_1.AbstractModel {
             schema: (0, ZodSchema_js_1.zodSchema)(llamaCppTextGenerationResponseSchema),
         }));
     }
-    processTextGenerationResponse(response) {
+    processTextGenerationResponse(rawResponse) {
         return {
-            response,
+            rawResponse,
             textGenerationResults: [
                 {
-                    text: response.content,
-                    finishReason: response.stopped_eos || response.stopped_word
+                    text: rawResponse.content,
+                    finishReason: rawResponse.stopped_eos || rawResponse.stopped_word
                         ? "stop"
-                        : response.stopped_limit
+                        : rawResponse.stopped_limit
                             ? "length"
                             : "unknown",
                 },
             ],
             usage: {
-                promptTokens: response.tokens_evaluated,
-                completionTokens: response.tokens_predicted,
-                totalTokens: response.tokens_evaluated + response.tokens_predicted,
+                promptTokens: rawResponse.tokens_evaluated,
+                completionTokens: rawResponse.tokens_predicted,
+                totalTokens: rawResponse.tokens_evaluated + rawResponse.tokens_predicted,
             },
         };
     }
@@ -167,27 +170,34 @@ class LlamaCppCompletionModel extends AbstractModel_js_1.AbstractModel {
     extractTextDelta(delta) {
         return delta.content;
     }
+    asStructureGenerationModel(promptTemplate) {
+        return "adaptModel" in promptTemplate
+            ? new StructureFromTextStreamingModel_js_1.StructureFromTextStreamingModel({
+                model: promptTemplate.adaptModel(this),
+                template: promptTemplate,
+            })
+            : new StructureFromTextStreamingModel_js_1.StructureFromTextStreamingModel({
+                model: this,
+                template: promptTemplate,
+            });
+    }
+    withJsonOutput() {
+        // don't override the grammar if it's already set (to support more restrictive grammars)
+        return this.settings.grammar == null
+            ? this.withSettings({ grammar: LlamaCppGrammars_js_1.json })
+            : this;
+    }
+    get promptTemplateProvider() {
+        return this.settings.promptTemplate ?? LlamaCppPrompt_js_1.Text;
+    }
     withTextPrompt() {
-        return this.withPromptTemplate({
-            format(prompt) {
-                return { text: prompt };
-            },
-            stopSequences: [],
-        });
+        return this.withPromptTemplate(this.promptTemplateProvider.text());
     }
-    /**
-     * Maps the prompt for a text version of the Llama.cpp prompt template (without image support).
-     */
-    withTextPromptTemplate(promptTemplate) {
-        return new PromptTemplateTextStreamingModel_js_1.PromptTemplateTextStreamingModel({
-            model: this.withTextPrompt().withSettings({
-                stopSequences: [
-                    ...(this.settings.stopSequences ?? []),
-                    ...promptTemplate.stopSequences,
-                ],
-            }),
-            promptTemplate,
-        });
+    withInstructionPrompt() {
+        return this.withPromptTemplate(this.promptTemplateProvider.instruction());
+    }
+    withChatPrompt() {
+        return this.withPromptTemplate(this.promptTemplateProvider.chat());
     }
     /**
      * Maps the prompt for the full Llama.cpp prompt template (incl. image support).

package/model-provider/llamacpp/LlamaCppCompletionModel.d.ts

@@ -4,9 +4,14 @@ import { ApiConfiguration } from "../../core/api/ApiConfiguration.js";
 import { ResponseHandler } from "../../core/api/postToApi.js";
 import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { Delta } from "../../model-function/Delta.js";
+import { FlexibleStructureFromTextPromptTemplate, StructureFromTextPromptTemplate } from "../../model-function/generate-structure/StructureFromTextPromptTemplate.js";
+import { StructureFromTextStreamingModel } from "../../model-function/generate-structure/StructureFromTextStreamingModel.js";
 import { PromptTemplateTextStreamingModel } from "../../model-function/generate-text/PromptTemplateTextStreamingModel.js";
 import { TextGenerationModelSettings, TextStreamingModel } from "../../model-function/generate-text/TextGenerationModel.js";
 import { TextGenerationPromptTemplate } from "../../model-function/generate-text/TextGenerationPromptTemplate.js";
+import { ChatPrompt } from "../../model-function/generate-text/prompt-template/ChatPrompt.js";
+import { InstructionPrompt } from "../../model-function/generate-text/prompt-template/InstructionPrompt.js";
+import { TextGenerationPromptTemplateProvider } from "../../model-function/generate-text/prompt-template/PromptTemplateProvider.js";
 import { LlamaCppTokenizer } from "./LlamaCppTokenizer.js";
 export interface LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE extends number | undefined> extends TextGenerationModelSettings {
     api?: ApiConfiguration;
@@ -121,6 +126,10 @@ export interface LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE extends num
      * If is -1 the task will be assigned to a Idle slot (default: -1)
      */
     slotId?: number;
+    /**
+     * Prompt template provider that is used when calling `.withTextPrompt()`, `withInstructionPrompt()` or `withChatPrompt()`.
+     */
+    promptTemplate?: TextGenerationPromptTemplateProvider<LlamaCppCompletionPrompt>;
 }
 export interface LlamaCppCompletionPrompt {
     /**
@@ -144,7 +153,7 @@ export declare class LlamaCppCompletionModel<CONTEXT_WINDOW_SIZE extends number
     get settingsForEvent(): Partial<LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>>;
     countPromptTokens(prompt: LlamaCppCompletionPrompt): Promise<number>;
     doGenerateTexts(prompt: LlamaCppCompletionPrompt, options: FunctionCallOptions): Promise<{
-        response: {
+        rawResponse: {
             model: string;
             stop: true;
             content: string;
@@ -204,7 +213,7 @@ export declare class LlamaCppCompletionModel<CONTEXT_WINDOW_SIZE extends number
         };
     }>;
     restoreGeneratedTexts(rawResponse: unknown): {
-        response: {
+        rawResponse: {
             model: string;
             stop: true;
             content: string;
@@ -263,8 +272,8 @@ export declare class LlamaCppCompletionModel<CONTEXT_WINDOW_SIZE extends number
             totalTokens: number;
         };
     };
-    processTextGenerationResponse(response: LlamaCppTextGenerationResponse): {
-        response: {
+    processTextGenerationResponse(rawResponse: LlamaCppTextGenerationResponse): {
+        rawResponse: {
             model: string;
             stop: true;
             content: string;
@@ -376,11 +385,12 @@ export declare class LlamaCppCompletionModel<CONTEXT_WINDOW_SIZE extends number
         content: string;
     }>>>;
     extractTextDelta(delta: unknown): string;
+    asStructureGenerationModel<INPUT_PROMPT, LlamaCppPrompt>(promptTemplate: StructureFromTextPromptTemplate<INPUT_PROMPT, LlamaCppPrompt> | FlexibleStructureFromTextPromptTemplate<INPUT_PROMPT, unknown>): StructureFromTextStreamingModel<INPUT_PROMPT, unknown, TextStreamingModel<unknown, TextGenerationModelSettings>> | StructureFromTextStreamingModel<INPUT_PROMPT, LlamaCppPrompt, TextStreamingModel<LlamaCppPrompt, TextGenerationModelSettings>>;
+    withJsonOutput(): this;
+    private get promptTemplateProvider();
     withTextPrompt(): PromptTemplateTextStreamingModel<string, LlamaCppCompletionPrompt, LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>, this>;
-    /**
-     * Maps the prompt for a text version of the Llama.cpp prompt template (without image support).
-     */
-    withTextPromptTemplate<INPUT_PROMPT>(promptTemplate: TextGenerationPromptTemplate<INPUT_PROMPT, string>): PromptTemplateTextStreamingModel<INPUT_PROMPT, string, LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>, PromptTemplateTextStreamingModel<string, LlamaCppCompletionPrompt, LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>, this>>;
+    withInstructionPrompt(): PromptTemplateTextStreamingModel<InstructionPrompt, LlamaCppCompletionPrompt, LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>, this>;
+    withChatPrompt(): PromptTemplateTextStreamingModel<ChatPrompt, LlamaCppCompletionPrompt, LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>, this>;
     /**
      * Maps the prompt for the full Llama.cpp prompt template (incl. image support).
      */

package/model-provider/llamacpp/LlamaCppCompletionModel.js

@@ -5,12 +5,15 @@ import { zodSchema } from "../../core/schema/ZodSchema.js";
 import { parseJSON } from "../../core/schema/parseJSON.js";
 import { validateTypes } from "../../core/schema/validateTypes.js";
 import { AbstractModel } from "../../model-function/AbstractModel.js";
+import { StructureFromTextStreamingModel } from "../../model-function/generate-structure/StructureFromTextStreamingModel.js";
 import { PromptTemplateTextStreamingModel } from "../../model-function/generate-text/PromptTemplateTextStreamingModel.js";
 import { textGenerationModelProperties, } from "../../model-function/generate-text/TextGenerationModel.js";
 import { AsyncQueue } from "../../util/AsyncQueue.js";
 import { parseEventSourceStream } from "../../util/streaming/parseEventSourceStream.js";
 import { LlamaCppApiConfiguration } from "./LlamaCppApiConfiguration.js";
 import { failedLlamaCppCallResponseHandler } from "./LlamaCppError.js";
+import { json } from "./LlamaCppGrammars.js";
+import { Text } from "./LlamaCppPrompt.js";
 import { LlamaCppTokenizer } from "./LlamaCppTokenizer.js";
 export class LlamaCppCompletionModel extends AbstractModel {
     constructor(settings = {}) {
@@ -136,23 +139,23 @@ export class LlamaCppCompletionModel extends AbstractModel {
             schema: zodSchema(llamaCppTextGenerationResponseSchema),
         }));
     }
-    processTextGenerationResponse(response) {
+    processTextGenerationResponse(rawResponse) {
         return {
-            response,
+            rawResponse,
             textGenerationResults: [
                 {
-                    text: response.content,
-                    finishReason: response.stopped_eos || response.stopped_word
+                    text: rawResponse.content,
+                    finishReason: rawResponse.stopped_eos || rawResponse.stopped_word
                         ? "stop"
-                        : response.stopped_limit
+                        : rawResponse.stopped_limit
                             ? "length"
                             : "unknown",
                 },
             ],
             usage: {
-                promptTokens: response.tokens_evaluated,
-                completionTokens: response.tokens_predicted,
-                totalTokens: response.tokens_evaluated + response.tokens_predicted,
+                promptTokens: rawResponse.tokens_evaluated,
+                completionTokens: rawResponse.tokens_predicted,
+                totalTokens: rawResponse.tokens_evaluated + rawResponse.tokens_predicted,
             },
         };
     }
@@ -164,27 +167,34 @@ export class LlamaCppCompletionModel extends AbstractModel {
     extractTextDelta(delta) {
         return delta.content;
     }
+    asStructureGenerationModel(promptTemplate) {
+        return "adaptModel" in promptTemplate
+            ? new StructureFromTextStreamingModel({
+                model: promptTemplate.adaptModel(this),
+                template: promptTemplate,
+            })
+            : new StructureFromTextStreamingModel({
+                model: this,
+                template: promptTemplate,
+            });
+    }
+    withJsonOutput() {
+        // don't override the grammar if it's already set (to support more restrictive grammars)
+        return this.settings.grammar == null
+            ? this.withSettings({ grammar: json })
+            : this;
+    }
+    get promptTemplateProvider() {
+        return this.settings.promptTemplate ?? Text;
+    }
     withTextPrompt() {
-        return this.withPromptTemplate({
-            format(prompt) {
-                return { text: prompt };
-            },
-            stopSequences: [],
-        });
+        return this.withPromptTemplate(this.promptTemplateProvider.text());
     }
-    /**
-     * Maps the prompt for a text version of the Llama.cpp prompt template (without image support).
-     */
-    withTextPromptTemplate(promptTemplate) {
-        return new PromptTemplateTextStreamingModel({
-            model: this.withTextPrompt().withSettings({
-                stopSequences: [
-                    ...(this.settings.stopSequences ?? []),
-                    ...promptTemplate.stopSequences,
-                ],
-            }),
-            promptTemplate,
-        });
+    withInstructionPrompt() {
+        return this.withPromptTemplate(this.promptTemplateProvider.instruction());
+    }
+    withChatPrompt() {
+        return this.withPromptTemplate(this.promptTemplateProvider.chat());
     }
     /**
      * Maps the prompt for the full Llama.cpp prompt template (incl. image support).

package/model-provider/llamacpp/LlamaCppFacade.cjs

@@ -1,6 +1,29 @@
 "use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || function (mod) {
+    if (mod && mod.__esModule) return mod;
+    var result = {};
+    if (mod != null) for (var k in mod) if (k !== "default" && Object.prototype.hasOwnProperty.call(mod, k)) __createBinding(result, mod, k);
+    __setModuleDefault(result, mod);
+    return result;
+};
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.Tokenizer = exports.TextEmbedder = exports.TextGenerator = exports.Api = void 0;
+exports.prompt = exports.grammar = exports.Tokenizer = exports.TextEmbedder = exports.CompletionTextGenerator = exports.Api = void 0;
 const LlamaCppApiConfiguration_js_1 = require("./LlamaCppApiConfiguration.cjs");
 const LlamaCppCompletionModel_js_1 = require("./LlamaCppCompletionModel.cjs");
 const LlamaCppTextEmbeddingModel_js_1 = require("./LlamaCppTextEmbeddingModel.cjs");
@@ -13,10 +36,10 @@ function Api(settings) {
     return new LlamaCppApiConfiguration_js_1.LlamaCppApiConfiguration(settings);
 }
 exports.Api = Api;
-function TextGenerator(settings = {}) {
+function CompletionTextGenerator(settings = {}) {
     return new LlamaCppCompletionModel_js_1.LlamaCppCompletionModel(settings);
 }
-exports.TextGenerator = TextGenerator;
+exports.CompletionTextGenerator = CompletionTextGenerator;
 function TextEmbedder(settings = {}) {
     return new LlamaCppTextEmbeddingModel_js_1.LlamaCppTextEmbeddingModel(settings);
 }
@@ -25,3 +48,8 @@ function Tokenizer(api = new LlamaCppApiConfiguration_js_1.LlamaCppApiConfigurat
     return new LlamaCppTokenizer_js_1.LlamaCppTokenizer(api);
 }
 exports.Tokenizer = Tokenizer;
+/**
+ * GBNF grammars. You can use them in the `grammar` option of the `TextGenerator` model.
+ */
+exports.grammar = __importStar(require("./LlamaCppGrammars.cjs"));
+exports.prompt = __importStar(require("./LlamaCppPrompt.cjs"));

package/model-provider/llamacpp/LlamaCppFacade.d.ts

@@ -9,6 +9,11 @@ import { LlamaCppTokenizer } from "./LlamaCppTokenizer.js";
  * It calls the API at http://127.0.0.1:8080 by default.
  */
 export declare function Api(settings: PartialBaseUrlPartsApiConfigurationOptions): LlamaCppApiConfiguration;
-export declare function TextGenerator<CONTEXT_WINDOW_SIZE extends number>(settings?: LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>): LlamaCppCompletionModel<CONTEXT_WINDOW_SIZE>;
+export declare function CompletionTextGenerator<CONTEXT_WINDOW_SIZE extends number>(settings?: LlamaCppCompletionModelSettings<CONTEXT_WINDOW_SIZE>): LlamaCppCompletionModel<CONTEXT_WINDOW_SIZE>;
 export declare function TextEmbedder(settings?: LlamaCppTextEmbeddingModelSettings): LlamaCppTextEmbeddingModel;
 export declare function Tokenizer(api?: ApiConfiguration): LlamaCppTokenizer;
+/**
+ * GBNF grammars. You can use them in the `grammar` option of the `TextGenerator` model.
+ */
+export * as grammar from "./LlamaCppGrammars.js";
+export * as prompt from "./LlamaCppPrompt.js";

package/model-provider/llamacpp/LlamaCppFacade.js

@@ -9,7 +9,7 @@ import { LlamaCppTokenizer } from "./LlamaCppTokenizer.js";
 export function Api(settings) {
     return new LlamaCppApiConfiguration(settings);
 }
-export function TextGenerator(settings = {}) {
+export function CompletionTextGenerator(settings = {}) {
     return new LlamaCppCompletionModel(settings);
 }
 export function TextEmbedder(settings = {}) {
@@ -18,3 +18,8 @@ export function TextEmbedder(settings = {}) {
 export function Tokenizer(api = new LlamaCppApiConfiguration()) {
     return new LlamaCppTokenizer(api);
 }
+/**
+ * GBNF grammars. You can use them in the `grammar` option of the `TextGenerator` model.
+ */
+export * as grammar from "./LlamaCppGrammars.js";
+export * as prompt from "./LlamaCppPrompt.js";

package/model-provider/llamacpp/LlamaCppGrammars.cjs

@@ -0,0 +1,84 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.list = exports.jsonArray = exports.json = void 0;
+/**
+ * GBNF grammar for JSON.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/json.gbnf
+ */
+exports.json = `
+root   ::= object
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\\"" (
+    [^"\\\\] |
+    "\\\\" (["\\\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+# Optional space: by convention, applied in this grammar after literal chars when allowed
+ws ::= ([ \t\n] ws)?
+`;
+/**
+ * GBNF grammar for JSON array outputs. Restricts whitespace at the end of the array.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/json_arr.gbnf
+ */
+exports.jsonArray = `
+root   ::= arr
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+arr  ::=
+  "[\n" ws (
+            value
+    (",\n" ws value)*
+  )? "]"
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\\"" (
+    [^"\\\\] |
+    "\\\\" (["\\\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+# Optional space: by convention, applied in this grammar after literal chars when allowed
+ws ::= ([ \t\n] ws)?
+`;
+/**
+ * GBNF grammar for list outputs. List items are separated by newlines and start with `- `.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/list.gbnf
+ */
+exports.list = `
+root ::= item+
+
+# Excludes various line break characters
+item ::= "- " [^\r\n\x0b\x0c\x85\u2028\u2029]+ "\n"
+`;

package/model-provider/llamacpp/LlamaCppGrammars.d.ts

@@ -0,0 +1,18 @@
+/**
+ * GBNF grammar for JSON.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/json.gbnf
+ */
+export declare const json: string;
+/**
+ * GBNF grammar for JSON array outputs. Restricts whitespace at the end of the array.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/json_arr.gbnf
+ */
+export declare const jsonArray: string;
+/**
+ * GBNF grammar for list outputs. List items are separated by newlines and start with `- `.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/list.gbnf
+ */
+export declare const list: string;

package/model-provider/llamacpp/LlamaCppGrammars.js

@@ -0,0 +1,81 @@
+/**
+ * GBNF grammar for JSON.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/json.gbnf
+ */
+export const json = `
+root   ::= object
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\\"" (
+    [^"\\\\] |
+    "\\\\" (["\\\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+# Optional space: by convention, applied in this grammar after literal chars when allowed
+ws ::= ([ \t\n] ws)?
+`;
+/**
+ * GBNF grammar for JSON array outputs. Restricts whitespace at the end of the array.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/json_arr.gbnf
+ */
+export const jsonArray = `
+root   ::= arr
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+arr  ::=
+  "[\n" ws (
+            value
+    (",\n" ws value)*
+  )? "]"
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\\"" (
+    [^"\\\\] |
+    "\\\\" (["\\\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+# Optional space: by convention, applied in this grammar after literal chars when allowed
+ws ::= ([ \t\n] ws)?
+`;
+/**
+ * GBNF grammar for list outputs. List items are separated by newlines and start with `- `.
+ *
+ * @see https://github.com/ggerganov/llama.cpp/blob/master/grammars/list.gbnf
+ */
+export const list = `
+root ::= item+
+
+# Excludes various line break characters
+item ::= "- " [^\r\n\x0b\x0c\x85\u2028\u2029]+ "\n"
+`;