diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 6a0cb7271f2c8e5d83af19d9937b87c967aee2e3..a13b0d6f26a0f9f792ed7e79a2a8a75bff5fd0f5 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -151,6 +151,9 @@ jobs:
- name: Pack @llamaindex/groq
run: pnpm pack --pack-destination ${{ runner.temp }}
working-directory: packages/llm/groq
+ - name: Pack @llamaindex/ollama
+ run: pnpm pack --pack-destination ${{ runner.temp }}
+ working-directory: packages/llm/ollama
- name: Pack @llamaindex/core
run: pnpm pack --pack-destination ${{ runner.temp }}
working-directory: packages/core
diff --git a/packages/llamaindex/package.json b/packages/llamaindex/package.json
index 201d6c6327991397c77145e8a1844e1cd2210887..d90b61321249c52f0b6bacf353edf2c0528585e3 100644
--- a/packages/llamaindex/package.json
+++ b/packages/llamaindex/package.json
@@ -34,6 +34,7 @@
"@llamaindex/core": "workspace:*",
"@llamaindex/env": "workspace:*",
"@llamaindex/groq": "workspace:*",
+ "@llamaindex/ollama": "workspace:*",
"@llamaindex/openai": "workspace:*",
"@mistralai/mistralai": "^1.0.4",
"@mixedbread-ai/sdk": "^2.2.11",
diff --git a/packages/llamaindex/src/embeddings/OllamaEmbedding.ts b/packages/llamaindex/src/embeddings/OllamaEmbedding.ts
index f6323c1495566453a43a23df661aef76d5a50773..2bd40a48eeaa17652ef2d5a46e15fc3e0aa9b2b1 100644
--- a/packages/llamaindex/src/embeddings/OllamaEmbedding.ts
+++ b/packages/llamaindex/src/embeddings/OllamaEmbedding.ts
@@ -1,5 +1,5 @@
import type { BaseEmbedding } from "@llamaindex/core/embeddings";
-import { Ollama } from "../llm/ollama.js";
+import { Ollama } from "@llamaindex/ollama";
/**
* OllamaEmbedding is an alias for Ollama that implements the BaseEmbedding interface.
diff --git a/packages/llamaindex/src/internal/deps/ollama.d.ts b/packages/llamaindex/src/internal/deps/ollama.d.ts
deleted file mode 100644
index 39e6218f7c9c8e33073b2ceae6f00fe7eb484882..0000000000000000000000000000000000000000
--- a/packages/llamaindex/src/internal/deps/ollama.d.ts
+++ /dev/null
@@ -1,264 +0,0 @@
-type Fetch = typeof fetch;
-interface Config {
- host: string;
- fetch?: Fetch;
- proxy?: boolean;
-}
-interface Options {
- numa: boolean;
- num_ctx: number;
- num_batch: number;
- main_gpu: number;
- low_vram: boolean;
- f16_kv: boolean;
- logits_all: boolean;
- vocab_only: boolean;
- use_mmap: boolean;
- use_mlock: boolean;
- embedding_only: boolean;
- num_thread: number;
- num_keep: number;
- seed: number;
- num_predict: number;
- top_k: number;
- top_p: number;
- tfs_z: number;
- typical_p: number;
- repeat_last_n: number;
- temperature: number;
- repeat_penalty: number;
- presence_penalty: number;
- frequency_penalty: number;
- mirostat: number;
- mirostat_tau: number;
- mirostat_eta: number;
- penalize_newline: boolean;
- stop: string[];
-}
-interface GenerateRequest {
- model: string;
- prompt: string;
- system?: string;
- template?: string;
- context?: number[];
- stream?: boolean;
- raw?: boolean;
- format?: string;
- images?: Uint8Array[] | string[];
- keep_alive?: string | number;
- options?: Partial<Options>;
-}
-interface Message {
- role: string;
- content: string;
- images?: Uint8Array[] | string[];
-}
-interface ChatRequest {
- model: string;
- messages?: Message[];
- stream?: boolean;
- format?: string;
- keep_alive?: string | number;
- options?: Partial<Options>;
-}
-interface PullRequest {
- model: string;
- insecure?: boolean;
- stream?: boolean;
-}
-interface PushRequest {
- model: string;
- insecure?: boolean;
- stream?: boolean;
-}
-interface CreateRequest {
- model: string;
- path?: string;
- modelfile?: string;
- stream?: boolean;
-}
-interface DeleteRequest {
- model: string;
-}
-interface CopyRequest {
- source: string;
- destination: string;
-}
-interface ShowRequest {
- model: string;
- system?: string;
- template?: string;
- options?: Partial<Options>;
-}
-interface EmbeddingsRequest {
- model: string;
- prompt: string;
- keep_alive?: string | number;
- options?: Partial<Options>;
-}
-interface GenerateResponse {
- model: string;
- created_at: Date;
- response: string;
- done: boolean;
- context: number[];
- total_duration: number;
- load_duration: number;
- prompt_eval_count: number;
- prompt_eval_duration: number;
- eval_count: number;
- eval_duration: number;
-}
-interface ChatResponse {
- model: string;
- created_at: Date;
- message: Message;
- done: boolean;
- total_duration: number;
- load_duration: number;
- prompt_eval_count: number;
- prompt_eval_duration: number;
- eval_count: number;
- eval_duration: number;
-}
-interface EmbeddingsResponse {
- embedding: number[];
-}
-interface ProgressResponse {
- status: string;
- digest: string;
- total: number;
- completed: number;
-}
-interface ModelResponse {
- name: string;
- modified_at: Date;
- size: number;
- digest: string;
- details: ModelDetails;
-}
-interface ModelDetails {
- parent_model: string;
- format: string;
- family: string;
- families: string[];
- parameter_size: string;
- quantization_level: string;
-}
-interface ShowResponse {
- license: string;
- modelfile: string;
- parameters: string;
- template: string;
- system: string;
- details: ModelDetails;
- messages: Message[];
-}
-interface ListResponse {
- models: ModelResponse[];
-}
-interface ErrorResponse {
- error: string;
-}
-interface StatusResponse {
- status: string;
-}
-
-declare class Ollama {
- protected readonly config: Config;
- protected readonly fetch: Fetch;
- private abortController;
- constructor(config?: Partial<Config>);
- abort(): void;
- protected processStreamableRequest<T extends object>(
- endpoint: string,
- request: {
- stream?: boolean;
- } & Record<string, any>,
- ): Promise<T | AsyncGenerator<T>>;
- encodeImage(image: Uint8Array | string): Promise<string>;
- generate(
- request: GenerateRequest & {
- stream: true;
- },
- ): Promise<AsyncGenerator<GenerateResponse>>;
- generate(
- request: GenerateRequest & {
- stream?: false;
- },
- ): Promise<GenerateResponse>;
- chat(
- request: ChatRequest & {
- stream: true;
- },
- ): Promise<AsyncGenerator<ChatResponse>>;
- chat(
- request: ChatRequest & {
- stream?: false;
- },
- ): Promise<ChatResponse>;
- create(
- request: CreateRequest & {
- stream: true;
- },
- ): Promise<AsyncGenerator<ProgressResponse>>;
- create(
- request: CreateRequest & {
- stream?: false;
- },
- ): Promise<ProgressResponse>;
- pull(
- request: PullRequest & {
- stream: true;
- },
- ): Promise<AsyncGenerator<ProgressResponse>>;
- pull(
- request: PullRequest & {
- stream?: false;
- },
- ): Promise<ProgressResponse>;
- push(
- request: PushRequest & {
- stream: true;
- },
- ): Promise<AsyncGenerator<ProgressResponse>>;
- push(
- request: PushRequest & {
- stream?: false;
- },
- ): Promise<ProgressResponse>;
- delete(request: DeleteRequest): Promise<StatusResponse>;
- copy(request: CopyRequest): Promise<StatusResponse>;
- list(): Promise<ListResponse>;
- show(request: ShowRequest): Promise<ShowResponse>;
- embeddings(request: EmbeddingsRequest): Promise<EmbeddingsResponse>;
-}
-declare const _default: Ollama;
-
-export {
- Ollama,
- _default as default,
- type ChatRequest,
- type ChatResponse,
- type Config,
- type CopyRequest,
- type CreateRequest,
- type DeleteRequest,
- type EmbeddingsRequest,
- type EmbeddingsResponse,
- type ErrorResponse,
- type Fetch,
- type GenerateRequest,
- type GenerateResponse,
- type ListResponse,
- type Message,
- type ModelDetails,
- type ModelResponse,
- type Options,
- type ProgressResponse,
- type PullRequest,
- type PushRequest,
- type ShowRequest,
- type ShowResponse,
- type StatusResponse,
-};
diff --git a/packages/llamaindex/src/internal/deps/ollama.js b/packages/llamaindex/src/internal/deps/ollama.js
deleted file mode 100644
index db189e10a1cc3196352378efe1f2b7f753aa6ab7..0000000000000000000000000000000000000000
--- a/packages/llamaindex/src/internal/deps/ollama.js
+++ /dev/null
@@ -1,462 +0,0 @@
-// generate from "tsup ./src/browser.js --format esm --dts"
-var __defProp = Object.defineProperty;
-var __getOwnPropSymbols = Object.getOwnPropertySymbols;
-var __hasOwnProp = Object.prototype.hasOwnProperty;
-var __propIsEnum = Object.prototype.propertyIsEnumerable;
-var __knownSymbol = (name, symbol) => {
- return (symbol = Symbol[name]) ? symbol : Symbol.for("Symbol." + name);
-};
-var __defNormalProp = (obj, key, value) =>
- key in obj
- ? __defProp(obj, key, {
- enumerable: true,
- configurable: true,
- writable: true,
- value,
- })
- : (obj[key] = value);
-var __spreadValues = (a, b) => {
- for (var prop in b || (b = {}))
- if (__hasOwnProp.call(b, prop)) __defNormalProp(a, prop, b[prop]);
- if (__getOwnPropSymbols)
- for (var prop of __getOwnPropSymbols(b)) {
- if (__propIsEnum.call(b, prop)) __defNormalProp(a, prop, b[prop]);
- }
- return a;
-};
-var __async = (__this, __arguments, generator) => {
- return new Promise((resolve, reject) => {
- var fulfilled = (value) => {
- try {
- step(generator.next(value));
- } catch (e) {
- reject(e);
- }
- };
- var rejected = (value) => {
- try {
- step(generator.throw(value));
- } catch (e) {
- reject(e);
- }
- };
- var step = (x) =>
- x.done
- ? resolve(x.value)
- : Promise.resolve(x.value).then(fulfilled, rejected);
- step((generator = generator.apply(__this, __arguments)).next());
- });
-};
-var __await = function (promise, isYieldStar) {
- this[0] = promise;
- this[1] = isYieldStar;
-};
-var __asyncGenerator = (__this, __arguments, generator) => {
- var resume = (k, v, yes, no) => {
- try {
- var x = generator[k](v),
- isAwait = (v = x.value) instanceof __await,
- done = x.done;
- Promise.resolve(isAwait ? v[0] : v)
- .then((y) =>
- isAwait
- ? resume(
- k === "return" ? k : "next",
- v[1] ? { done: y.done, value: y.value } : y,
- yes,
- no,
- )
- : yes({ value: y, done }),
- )
- .catch((e) => resume("throw", e, yes, no));
- } catch (e) {
- no(e);
- }
- };
- var method = (k) =>
- (it[k] = (x) => new Promise((yes, no) => resume(k, x, yes, no)));
- var it = {};
- return (
- (generator = generator.apply(__this, __arguments)),
- (it[__knownSymbol("asyncIterator")] = () => it),
- method("next"),
- method("throw"),
- method("return"),
- it
- );
-};
-var __forAwait = (obj, it, method) =>
- (it = obj[__knownSymbol("asyncIterator")])
- ? it.call(obj)
- : ((obj = obj[__knownSymbol("iterator")]()),
- (it = {}),
- (method = (key, fn) =>
- (fn = obj[key]) &&
- (it[key] = (arg) =>
- new Promise(
- (yes, no, done) => (
- (arg = fn.call(obj, arg)),
- (done = arg.done),
- Promise.resolve(arg.value).then(
- (value) => yes({ value, done }),
- no,
- )
- ),
- ))),
- method("next"),
- method("return"),
- it);
-
-// src/version.ts
-var version = "0.0.0";
-
-// src/utils.ts
-var ResponseError = class _ResponseError extends Error {
- constructor(error, status_code) {
- super(error);
- this.error = error;
- this.status_code = status_code;
- this.name = "ResponseError";
- if (Error.captureStackTrace) {
- Error.captureStackTrace(this, _ResponseError);
- }
- }
-};
-var checkOk = (response) =>
- __async(void 0, null, function* () {
- var _a;
- if (!response.ok) {
- let message = `Error ${response.status}: ${response.statusText}`;
- let errorData = null;
- if (
- (_a = response.headers.get("content-type")) == null
- ? void 0
- : _a.includes("application/json")
- ) {
- try {
- errorData = yield response.json();
- message = errorData.error || message;
- } catch (error) {
- console.log("Failed to parse error response as JSON");
- }
- } else {
- try {
- console.log("Getting text from response");
- const textResponse = yield response.text();
- message = textResponse || message;
- } catch (error) {
- console.log("Failed to get text from error response");
- }
- }
- throw new ResponseError(message, response.status);
- }
- });
-function getPlatform() {
- if (typeof window !== "undefined" && window.navigator) {
- return `${window.navigator.platform.toLowerCase()} Browser/${navigator.userAgent};`;
- } else if (typeof process !== "undefined") {
- return `${process.arch} ${process.platform} Node.js/${process.version}`;
- }
- return "";
-}
-var fetchWithHeaders = (_0, _1, ..._2) =>
- __async(void 0, [_0, _1, ..._2], function* (fetch2, url, options = {}) {
- const defaultHeaders = {
- "Content-Type": "application/json",
- Accept: "application/json",
- "User-Agent": `ollama-js/${version} (${getPlatform()})`,
- };
- if (!options.headers) {
- options.headers = {};
- }
- options.headers = __spreadValues(
- __spreadValues({}, defaultHeaders),
- options.headers,
- );
- return fetch2(url, options);
- });
-var get = (fetch2, host) =>
- __async(void 0, null, function* () {
- const response = yield fetchWithHeaders(fetch2, host);
- yield checkOk(response);
- return response;
- });
-var post = (fetch2, host, data, options) =>
- __async(void 0, null, function* () {
- const isRecord = (input) => {
- return (
- input !== null && typeof input === "object" && !Array.isArray(input)
- );
- };
- const formattedData = isRecord(data) ? JSON.stringify(data) : data;
- const response = yield fetchWithHeaders(fetch2, host, {
- method: "POST",
- body: formattedData,
- signal: options == null ? void 0 : options.signal,
- });
- yield checkOk(response);
- return response;
- });
-var del = (fetch2, host, data) =>
- __async(void 0, null, function* () {
- const response = yield fetchWithHeaders(fetch2, host, {
- method: "DELETE",
- body: JSON.stringify(data),
- });
- yield checkOk(response);
- return response;
- });
-var parseJSON = function (itr) {
- return __asyncGenerator(this, null, function* () {
- var _a;
- const decoder = new TextDecoder("utf-8");
- let buffer = "";
- const reader = itr.getReader();
- while (true) {
- const { done, value: chunk } = yield new __await(reader.read());
- if (done) {
- break;
- }
- buffer += decoder.decode(chunk);
- const parts = buffer.split("\n");
- buffer = (_a = parts.pop()) != null ? _a : "";
- for (const part of parts) {
- try {
- yield JSON.parse(part);
- } catch (error) {
- console.warn("invalid json: ", part);
- }
- }
- }
- for (const part of buffer.split("\n").filter((p) => p !== "")) {
- try {
- yield JSON.parse(part);
- } catch (error) {
- console.warn("invalid json: ", part);
- }
- }
- });
-};
-var formatHost = (host) => {
- if (!host) {
- return "http://127.0.0.1:11434";
- }
- let isExplicitProtocol = host.includes("://");
- if (host.startsWith(":")) {
- host = `http://127.0.0.1${host}`;
- isExplicitProtocol = false;
- }
- if (!isExplicitProtocol) {
- host = `http://${host}`;
- }
- const url = new URL(host);
- let port = url.port;
- if (!port) {
- if (!isExplicitProtocol) {
- port = "11434";
- } else {
- port = url.protocol === "https:" ? "443" : "80";
- }
- }
- let formattedHost = `${url.protocol}//${url.hostname}:${port}${url.pathname}`;
- if (formattedHost.endsWith("/")) {
- formattedHost = formattedHost.slice(0, -1);
- }
- return formattedHost;
-};
-
-// src/browser.ts
-// import "whatwg-fetch";
-var Ollama = class {
- constructor(config) {
- var _a;
- this.config = {
- host: "",
- };
- if (!(config == null ? void 0 : config.proxy)) {
- this.config.host = formatHost(
- (_a = config == null ? void 0 : config.host) != null
- ? _a
- : "http://127.0.0.1:11434",
- );
- }
- this.fetch = fetch;
- if ((config == null ? void 0 : config.fetch) != null) {
- this.fetch = config.fetch;
- }
- this.abortController = new AbortController();
- }
- // Abort any ongoing requests to Ollama
- abort() {
- this.abortController.abort();
- this.abortController = new AbortController();
- }
- processStreamableRequest(endpoint, request) {
- return __async(this, null, function* () {
- var _a;
- request.stream = (_a = request.stream) != null ? _a : false;
- const response = yield post(
- this.fetch,
- `${this.config.host}/api/${endpoint}`,
- __spreadValues({}, request),
- { signal: this.abortController.signal },
- );
- if (!response.body) {
- throw new Error("Missing body");
- }
- const itr = parseJSON(response.body);
- if (request.stream) {
- return (function () {
- return __asyncGenerator(this, null, function* () {
- try {
- for (
- var iter = __forAwait(itr), more, temp, error;
- (more = !(temp = yield new __await(iter.next())).done);
- more = false
- ) {
- const message = temp.value;
- if ("error" in message) {
- throw new Error(message.error);
- }
- yield message;
- if (message.done || message.status === "success") {
- return;
- }
- }
- } catch (temp) {
- error = [temp];
- } finally {
- try {
- more &&
- (temp = iter.return) &&
- (yield new __await(temp.call(iter)));
- } finally {
- if (error) throw error[0];
- }
- }
- throw new Error(
- "Did not receive done or success response in stream.",
- );
- });
- })();
- } else {
- const message = yield itr.next();
- if (!message.value.done && message.value.status !== "success") {
- throw new Error("Expected a completed response.");
- }
- return message.value;
- }
- });
- }
- encodeImage(image) {
- return __async(this, null, function* () {
- if (typeof image !== "string") {
- const uint8Array = new Uint8Array(image);
- const numberArray = Array.from(uint8Array);
- const base64String = btoa(String.fromCharCode.apply(null, numberArray));
- return base64String;
- }
- return image;
- });
- }
- generate(request) {
- return __async(this, null, function* () {
- if (request.images) {
- request.images = yield Promise.all(
- request.images.map(this.encodeImage.bind(this)),
- );
- }
- return this.processStreamableRequest("generate", request);
- });
- }
- chat(request) {
- return __async(this, null, function* () {
- if (request.messages) {
- for (const message of request.messages) {
- if (message.images) {
- message.images = yield Promise.all(
- message.images.map(this.encodeImage.bind(this)),
- );
- }
- }
- }
- return this.processStreamableRequest("chat", request);
- });
- }
- create(request) {
- return __async(this, null, function* () {
- return this.processStreamableRequest("create", {
- name: request.model,
- stream: request.stream,
- modelfile: request.modelfile,
- });
- });
- }
- pull(request) {
- return __async(this, null, function* () {
- return this.processStreamableRequest("pull", {
- name: request.model,
- stream: request.stream,
- insecure: request.insecure,
- });
- });
- }
- push(request) {
- return __async(this, null, function* () {
- return this.processStreamableRequest("push", {
- name: request.model,
- stream: request.stream,
- insecure: request.insecure,
- });
- });
- }
- delete(request) {
- return __async(this, null, function* () {
- yield del(this.fetch, `${this.config.host}/api/delete`, {
- name: request.model,
- });
- return { status: "success" };
- });
- }
- copy(request) {
- return __async(this, null, function* () {
- yield post(
- this.fetch,
- `${this.config.host}/api/copy`,
- __spreadValues({}, request),
- );
- return { status: "success" };
- });
- }
- list() {
- return __async(this, null, function* () {
- const response = yield get(this.fetch, `${this.config.host}/api/tags`);
- const listResponse = yield response.json();
- return listResponse;
- });
- }
- show(request) {
- return __async(this, null, function* () {
- const response = yield post(
- this.fetch,
- `${this.config.host}/api/show`,
- __spreadValues({}, request),
- );
- const showResponse = yield response.json();
- return showResponse;
- });
- }
- embeddings(request) {
- return __async(this, null, function* () {
- const response = yield post(
- this.fetch,
- `${this.config.host}/api/embeddings`,
- __spreadValues({}, request),
- );
- const embeddingsResponse = yield response.json();
- return embeddingsResponse;
- });
- }
-};
-var browser_default = new Ollama();
-export { Ollama, browser_default as default };
diff --git a/packages/llamaindex/src/internal/deps/ollama.license b/packages/llamaindex/src/internal/deps/ollama.license
deleted file mode 100644
index 49bd8b185208cba029f60faf87388482ab647775..0000000000000000000000000000000000000000
--- a/packages/llamaindex/src/internal/deps/ollama.license
+++ /dev/null
@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2023 Saul
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
diff --git a/packages/llamaindex/src/llm/index.ts b/packages/llamaindex/src/llm/index.ts
index 61698abcb587e633ab194121d0f94d2d7b85e676..5596440a0ce24099adeb6d692f4299e67c5855e8 100644
--- a/packages/llamaindex/src/llm/index.ts
+++ b/packages/llamaindex/src/llm/index.ts
@@ -23,7 +23,7 @@ export { Portkey } from "./portkey.js";
export * from "./replicate_ai.js";
// Note: The type aliases for replicate are to simplify usage for Llama 2 (we're using replicate for Llama 2 support)
export { DeepInfra } from "./deepinfra.js";
-export { Ollama, type OllamaParams } from "./ollama.js";
+export * from "./ollama.js";
export {
ALL_AVAILABLE_REPLICATE_MODELS,
DeuceChatStrategy,
diff --git a/packages/llamaindex/src/llm/ollama.ts b/packages/llamaindex/src/llm/ollama.ts
index 1fea1b7d0467dcb86e7dcc97d532d3ab0e8edd27..0047c5bc9ee5ba43ce94887c74e81791c0600f24 100644
--- a/packages/llamaindex/src/llm/ollama.ts
+++ b/packages/llamaindex/src/llm/ollama.ts
@@ -1,253 +1 @@
-import { BaseEmbedding } from "@llamaindex/core/embeddings";
-import type {
- ChatResponse,
- ChatResponseChunk,
- CompletionResponse,
- LLM,
- LLMChatParamsNonStreaming,
- LLMChatParamsStreaming,
- LLMCompletionParamsNonStreaming,
- LLMCompletionParamsStreaming,
- LLMMetadata,
-} from "@llamaindex/core/llms";
-import { extractText, streamConverter } from "@llamaindex/core/utils";
-import {
- Ollama as OllamaBase,
- type Config,
- type CopyRequest,
- type CreateRequest,
- type DeleteRequest,
- type EmbeddingsRequest,
- type EmbeddingsResponse,
- type GenerateRequest,
- type ListResponse,
- type ChatResponse as OllamaChatResponse,
- type GenerateResponse as OllamaGenerateResponse,
- type Options,
- type ProgressResponse,
- type PullRequest,
- type PushRequest,
- type ShowRequest,
- type ShowResponse,
- type StatusResponse,
-} from "../internal/deps/ollama.js";
-
-const messageAccessor = (part: OllamaChatResponse): ChatResponseChunk => {
- return {
- raw: part,
- delta: part.message.content,
- };
-};
-
-const completionAccessor = (
- part: OllamaGenerateResponse,
-): CompletionResponse => {
- return { text: part.response, raw: part };
-};
-
-export type OllamaParams = {
- model: string;
- config?: Partial<Config>;
- options?: Partial<Options>;
-};
-
-/**
- * This class both implements the LLM and Embedding interfaces.
- */
-export class Ollama
- extends BaseEmbedding
- implements LLM, Omit<OllamaBase, "chat">
-{
- readonly hasStreaming = true;
-
- ollama: OllamaBase;
-
- // https://ollama.ai/library
- model: string;
-
- options: Partial<Omit<Options, "num_ctx" | "top_p" | "temperature">> &
- Pick<Options, "num_ctx" | "top_p" | "temperature"> = {
- num_ctx: 4096,
- top_p: 0.9,
- temperature: 0.7,
- };
-
- constructor(params: OllamaParams) {
- super();
- this.model = params.model;
- this.ollama = new OllamaBase(params.config);
- if (params.options) {
- this.options = {
- ...this.options,
- ...params.options,
- };
- }
- }
-
- get metadata(): LLMMetadata {
- const { temperature, top_p, num_ctx } = this.options;
- return {
- model: this.model,
- temperature: temperature,
- topP: top_p,
- maxTokens: undefined,
- contextWindow: num_ctx,
- tokenizer: undefined,
- };
- }
-
- chat(
- params: LLMChatParamsStreaming,
- ): Promise<AsyncIterable<ChatResponseChunk>>;
- chat(params: LLMChatParamsNonStreaming): Promise<ChatResponse>;
- async chat(
- params: LLMChatParamsNonStreaming | LLMChatParamsStreaming,
- ): Promise<ChatResponse | AsyncIterable<ChatResponseChunk>> {
- const { messages, stream } = params;
- const payload = {
- model: this.model,
- messages: messages.map((message) => ({
- role: message.role,
- content: extractText(message.content),
- })),
- stream: !!stream,
- options: {
- ...this.options,
- },
- };
- if (!stream) {
- const chatResponse = await this.ollama.chat({
- ...payload,
- stream: false,
- });
-
- return {
- message: {
- role: "assistant",
- content: chatResponse.message.content,
- },
- raw: chatResponse,
- };
- } else {
- const stream = await this.ollama.chat({
- ...payload,
- stream: true,
- });
- return streamConverter(stream, messageAccessor);
- }
- }
-
- complete(
- params: LLMCompletionParamsStreaming,
- ): Promise<AsyncIterable<CompletionResponse>>;
- complete(
- params: LLMCompletionParamsNonStreaming,
- ): Promise<CompletionResponse>;
- async complete(
- params: LLMCompletionParamsStreaming | LLMCompletionParamsNonStreaming,
- ): Promise<CompletionResponse | AsyncIterable<CompletionResponse>> {
- const { prompt, stream } = params;
- const payload = {
- model: this.model,
- prompt: extractText(prompt),
- stream: !!stream,
- options: {
- ...this.options,
- },
- };
- if (!stream) {
- const response = await this.ollama.generate({
- ...payload,
- stream: false,
- });
- return {
- text: response.response,
- raw: response,
- };
- } else {
- const stream = await this.ollama.generate({
- ...payload,
- stream: true,
- });
- return streamConverter(stream, completionAccessor);
- }
- }
-
- private async getEmbedding(prompt: string): Promise<number[]> {
- const payload = {
- model: this.model,
- prompt,
- options: {
- ...this.options,
- },
- };
- const response = await this.ollama.embeddings({
- ...payload,
- });
- return response.embedding;
- }
-
- async getTextEmbedding(text: string): Promise<number[]> {
- return this.getEmbedding(text);
- }
-
- // Inherited from OllamaBase
-
- push(
- request: PushRequest & { stream: true },
- ): Promise<AsyncGenerator<ProgressResponse, any, unknown>>;
- push(
- request: PushRequest & { stream?: false | undefined },
- ): Promise<ProgressResponse>;
- push(request: any): any {
- return this.ollama.push(request);
- }
- abort(): void {
- return this.ollama.abort();
- }
- encodeImage(image: string | Uint8Array): Promise<string> {
- return this.ollama.encodeImage(image);
- }
- generate(
- request: GenerateRequest & { stream: true },
- ): Promise<AsyncGenerator<OllamaGenerateResponse>>;
- generate(
- request: GenerateRequest & { stream?: false | undefined },
- ): Promise<OllamaGenerateResponse>;
- generate(request: any): any {
- return this.ollama.generate(request);
- }
- create(
- request: CreateRequest & { stream: true },
- ): Promise<AsyncGenerator<ProgressResponse>>;
- create(
- request: CreateRequest & { stream?: false | undefined },
- ): Promise<ProgressResponse>;
- create(request: any): any {
- return this.ollama.create(request);
- }
- pull(
- request: PullRequest & { stream: true },
- ): Promise<AsyncGenerator<ProgressResponse>>;
- pull(
- request: PullRequest & { stream?: false | undefined },
- ): Promise<ProgressResponse>;
- pull(request: any): any {
- return this.ollama.pull(request);
- }
- delete(request: DeleteRequest): Promise<StatusResponse> {
- return this.ollama.delete(request);
- }
- copy(request: CopyRequest): Promise<StatusResponse> {
- return this.ollama.copy(request);
- }
- list(): Promise<ListResponse> {
- return this.ollama.list();
- }
- show(request: ShowRequest): Promise<ShowResponse> {
- return this.ollama.show(request);
- }
- embeddings(request: EmbeddingsRequest): Promise<EmbeddingsResponse> {
- return this.ollama.embeddings(request);
- }
-}
+export { Ollama, type OllamaParams } from "@llamaindex/ollama";
diff --git a/packages/llm/ollama/package.json b/packages/llm/ollama/package.json
new file mode 100644
index 0000000000000000000000000000000000000000..40bf722b93dc3afdbf507c652bbe46b1966d3bb5
--- /dev/null
+++ b/packages/llm/ollama/package.json
@@ -0,0 +1,41 @@
+{
+ "name": "@llamaindex/ollama",
+ "description": "Ollama Adapter for LlamaIndex",
+ "version": "0.0.1",
+ "type": "module",
+ "main": "./dist/index.cjs",
+ "module": "./dist/index.js",
+ "exports": {
+ ".": {
+ "require": {
+ "types": "./dist/index.d.cts",
+ "default": "./dist/index.cjs"
+ },
+ "import": {
+ "types": "./dist/index.d.ts",
+ "default": "./dist/index.js"
+ }
+ }
+ },
+ "files": [
+ "dist"
+ ],
+ "repository": {
+ "type": "git",
+ "url": "https://github.com/run-llama/LlamaIndexTS.git",
+ "directory": "packages/llm/openai"
+ },
+ "scripts": {
+ "build": "bunchee",
+ "dev": "bunchee --watch"
+ },
+ "devDependencies": {
+ "bunchee": "5.3.2"
+ },
+ "dependencies": {
+ "@llamaindex/core": "workspace:*",
+ "@llamaindex/env": "workspace:*",
+ "ollama": "^0.5.9",
+ "remeda": "^2.12.0"
+ }
+}
diff --git a/packages/llm/ollama/src/index.ts b/packages/llm/ollama/src/index.ts
new file mode 100644
index 0000000000000000000000000000000000000000..1bdcd4b81af422e56ac4fe09ce13a1fc1faead57
--- /dev/null
+++ b/packages/llm/ollama/src/index.ts
@@ -0,0 +1,172 @@
+import { BaseEmbedding } from "@llamaindex/core/embeddings";
+import type {
+ ChatResponse,
+ ChatResponseChunk,
+ CompletionResponse,
+ LLM,
+ LLMChatParamsNonStreaming,
+ LLMChatParamsStreaming,
+ LLMCompletionParamsNonStreaming,
+ LLMCompletionParamsStreaming,
+ LLMMetadata,
+} from "@llamaindex/core/llms";
+import { extractText, streamConverter } from "@llamaindex/core/utils";
+import {
+ Ollama as OllamaBase,
+ type Config,
+ type ChatResponse as OllamaChatResponse,
+ type GenerateResponse as OllamaGenerateResponse,
+ type Options,
+} from "ollama/browser";
+
+const messageAccessor = (part: OllamaChatResponse): ChatResponseChunk => {
+ return {
+ raw: part,
+ delta: part.message.content,
+ };
+};
+
+const completionAccessor = (
+ part: OllamaGenerateResponse,
+): CompletionResponse => {
+ return { text: part.response, raw: part };
+};
+
+export type OllamaParams = {
+ model: string;
+ config?: Partial<Config>;
+ options?: Partial<Options>;
+};
+
+export class Ollama extends BaseEmbedding implements LLM {
+ public readonly ollama: OllamaBase;
+
+ // https://ollama.ai/library
+ model: string;
+
+ options: Partial<Omit<Options, "num_ctx" | "top_p" | "temperature">> &
+ Pick<Options, "num_ctx" | "top_p" | "temperature"> = {
+ num_ctx: 4096,
+ top_p: 0.9,
+ temperature: 0.7,
+ };
+
+ constructor(params: OllamaParams) {
+ super();
+ this.model = params.model;
+ this.ollama = new OllamaBase(params.config);
+ if (params.options) {
+ this.options = {
+ ...this.options,
+ ...params.options,
+ };
+ }
+ }
+
+ get metadata(): LLMMetadata {
+ const { temperature, top_p, num_ctx } = this.options;
+ return {
+ model: this.model,
+ temperature: temperature,
+ topP: top_p,
+ maxTokens: this.options.num_ctx,
+ contextWindow: num_ctx,
+ tokenizer: undefined,
+ };
+ }
+
+ chat(
+ params: LLMChatParamsStreaming,
+ ): Promise<AsyncIterable<ChatResponseChunk>>;
+ chat(params: LLMChatParamsNonStreaming): Promise<ChatResponse>;
+ async chat(
+ params: LLMChatParamsNonStreaming | LLMChatParamsStreaming,
+ ): Promise<ChatResponse | AsyncIterable<ChatResponseChunk>> {
+ const { messages, stream } = params;
+ const payload = {
+ model: this.model,
+ messages: messages.map((message) => ({
+ role: message.role,
+ content: extractText(message.content),
+ })),
+ stream: !!stream,
+ options: {
+ ...this.options,
+ },
+ };
+ if (!stream) {
+ const chatResponse = await this.ollama.chat({
+ ...payload,
+ stream: false,
+ });
+
+ return {
+ message: {
+ role: "assistant",
+ content: chatResponse.message.content,
+ },
+ raw: chatResponse,
+ };
+ } else {
+ const stream = await this.ollama.chat({
+ ...payload,
+ stream: true,
+ });
+ return streamConverter(stream, messageAccessor);
+ }
+ }
+
+ complete(
+ params: LLMCompletionParamsStreaming,
+ ): Promise<AsyncIterable<CompletionResponse>>;
+ complete(
+ params: LLMCompletionParamsNonStreaming,
+ ): Promise<CompletionResponse>;
+ async complete(
+ params: LLMCompletionParamsStreaming | LLMCompletionParamsNonStreaming,
+ ): Promise<CompletionResponse | AsyncIterable<CompletionResponse>> {
+ const { prompt, stream } = params;
+ const payload = {
+ model: this.model,
+ prompt: extractText(prompt),
+ stream: !!stream,
+ options: {
+ ...this.options,
+ },
+ };
+ if (!stream) {
+ const response = await this.ollama.generate({
+ ...payload,
+ stream: false,
+ });
+ return {
+ text: response.response,
+ raw: response,
+ };
+ } else {
+ const stream = await this.ollama.generate({
+ ...payload,
+ stream: true,
+ });
+ return streamConverter(stream, completionAccessor);
+ }
+ }
+
+ private async getEmbedding(prompt: string): Promise<number[]> {
+ const payload = {
+ model: this.model,
+ prompt,
+ options: {
+ ...this.options,
+ },
+ };
+ const response = await this.ollama.embeddings({
+ ...payload,
+ });
+ return response.embedding;
+ }
+
+ async getTextEmbedding(text: string): Promise<number[]> {
+ return this.getEmbedding(text);
+ }
+}
diff --git a/packages/llm/ollama/tsconfig.json b/packages/llm/ollama/tsconfig.json
new file mode 100644
index 0000000000000000000000000000000000000000..5a94aa033116766429d5c2d42f337c5b1393ba86
--- /dev/null
+++ b/packages/llm/ollama/tsconfig.json
@@ -0,0 +1,18 @@
+{
+ "extends": "../../../tsconfig.json",
+ "compilerOptions": {
+ "target": "ESNext",
+ "module": "ESNext",
+ "moduleResolution": "bundler",
+ "outDir": "./lib"
+ },
+ "include": ["./src"],
+ "references": [
+ {
+ "path": "../../llamaindex/tsconfig.json"
+ },
+ {
+ "path": "../../env/tsconfig.json"
+ }
+ ]
+}
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml
index 056bf0720ccad02eee799760773cef272908e50b..e3d2abce4fadcdf35abc4a0ee9443708cbec7165 100644
--- a/pnpm-lock.yaml
+++ b/pnpm-lock.yaml
@@ -541,6 +541,9 @@ importers:
'@llamaindex/groq':
specifier: workspace:*
version: link:../llm/groq
+ '@llamaindex/ollama':
+ specifier: workspace:*
+ version: link:../llm/ollama
'@llamaindex/openai':
specifier: workspace:*
version: link:../llm/openai
@@ -922,6 +925,25 @@ importers:
specifier: 5.3.2
version: 5.3.2(typescript@5.6.2)
+ packages/llm/ollama:
+ dependencies:
+ '@llamaindex/core':
+ specifier: workspace:*
+ version: link:../../core
+ '@llamaindex/env':
+ specifier: workspace:*
+ version: link:../../env
+ ollama:
+ specifier: ^0.5.9
+ version: 0.5.9
+ remeda:
+ specifier: ^2.12.0
+ version: 2.12.0
+ devDependencies:
+ bunchee:
+ specifier: 5.3.2
+ version: 5.3.2(typescript@5.6.2)
+
packages/llm/openai:
dependencies:
'@llamaindex/core':
@@ -8885,6 +8907,9 @@ packages:
ohash@1.1.3:
resolution: {integrity: sha512-zuHHiGTYTA1sYJ/wZN+t5HKZaH23i4yI1HMwbuXm24Nid7Dv0KcuRlKoNKS9UNfAVSBlnGLcuQrnOKWOZoEGaw==}
+ ollama@0.5.9:
+ resolution: {integrity: sha512-F/KZuDRC+ZsVCuMvcOYuQ6zj42/idzCkkuknGyyGVmNStMZ/sU3jQpvhnl4SyC0+zBzLiKNZJnJeuPFuieWZvQ==}
+
on-finished@2.4.1:
resolution: {integrity: sha512-oVlzkg3ENAhCk2zdv7IJwd/QUD4z2RxRwpkcGY8psCVcCYZNq4wYnVWALHM+brtuJjePWiYF/ClmuDr8Ch5+kg==}
engines: {node: '>= 0.8'}
@@ -22795,6 +22820,10 @@ snapshots:
ohash@1.1.3: {}
+ ollama@0.5.9:
+ dependencies:
+ whatwg-fetch: 3.6.20
+
on-finished@2.4.1:
dependencies:
ee-first: 1.1.1
diff --git a/tsconfig.json b/tsconfig.json
index c6db3c526a2195073d43276818c7d56c911c39b3..a03509a020cd58c7ac7585d02def554ff69a2444 100644
--- a/tsconfig.json
+++ b/tsconfig.json
@@ -32,6 +32,9 @@
{
"path": "./packages/llm/groq/tsconfig.json"
},
+ {
+ "path": "./packages/llm/ollama/tsconfig.json"
+ },
{
"path": "./packages/cloud/tsconfig.json"
},