Repository navigation
feat(embeddings): native Ollama embedding provider #146
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,155 @@ | ||
| import { EmbeddingProvider } from './types.js'; | ||
|
|
||
| /** Common Ollama embedding models and their vector sizes (for LanceDB dimension checks). */ | ||
| export const OLLAMA_MODEL_DIMENSIONS: Record<string, number> = { | ||
| 'nomic-embed-text': 768, | ||
| 'mxbai-embed-large': 1024, | ||
| 'all-minilm': 384, | ||
| 'snowflake-arctic-embed:137m': 768, | ||
| 'snowflake-arctic-embed:335m': 1024 | ||
| }; | ||
|
|
||
| interface OllamaEmbedResponse { | ||
| embeddings: number[][]; | ||
| } | ||
|
|
||
| interface OllamaShowResponse { | ||
| modelfile?: string; | ||
| parameters?: string; | ||
| } | ||
|
|
||
| function normalizeOllamaHost(host: string): string { | ||
| return host.replace(/\/+$/, ''); | ||
| } | ||
|
|
||
| export function getOllamaModelDimensions(modelName: string): number { | ||
| const direct = OLLAMA_MODEL_DIMENSIONS[modelName]; | ||
| if (direct !== undefined) { | ||
| return direct; | ||
| } | ||
| const baseName = modelName.split(':')[0] ?? modelName; | ||
| const fromBase = OLLAMA_MODEL_DIMENSIONS[baseName]; | ||
| if (fromBase !== undefined) { | ||
| return fromBase; | ||
| } | ||
| const fromEnv = process.env.OLLAMA_EMBEDDING_DIMENSIONS; | ||
| if (fromEnv) { | ||
| const parsed = parseInt(fromEnv, 10); | ||
| if (!Number.isNaN(parsed) && parsed > 0) { | ||
| return parsed; | ||
| } | ||
| } | ||
| return 768; | ||
|
There was a problem hiding this comment.
For an Ollama model outside the hardcoded map that returns vectors other than 768 dimensions, a full build succeeds because Useful? React with 👍 / 👎. |
||
| } | ||
|
|
||
| /** | ||
| * Ollama embedding provider — HTTP client for a local Ollama server (`/api/embed`). | ||
| */ | ||
| export class OllamaEmbeddingProvider implements EmbeddingProvider { | ||
| readonly name = 'ollama'; | ||
| private resolvedDimensions: number; | ||
|
|
||
| constructor( | ||
| readonly modelName: string, | ||
| private host: string = 'http://localhost:11434' | ||
| ) { | ||
| this.host = normalizeOllamaHost(host); | ||
| this.resolvedDimensions = getOllamaModelDimensions(modelName); | ||
| } | ||
|
|
||
| get dimensions(): number { | ||
| return this.resolvedDimensions; | ||
| } | ||
|
|
||
| async initialize(): Promise<void> { | ||
| let response: Response; | ||
| try { | ||
| response = await fetch(`${this.host}/api/show`, { | ||
| method: 'POST', | ||
| headers: { 'Content-Type': 'application/json' }, | ||
| body: JSON.stringify({ name: this.modelName }) | ||
| }); | ||
| } catch (error) { | ||
| throw new Error( | ||
| `Cannot reach Ollama at ${this.host}. Start the server (ollama serve) or set OLLAMA_HOST. ${String(error)}` | ||
| ); | ||
| } | ||
|
|
||
| if (response.status === 404) { | ||
| throw new Error( | ||
| `Ollama model "${this.modelName}" is not available. Pull it first: ollama pull ${this.modelName}` | ||
| ); | ||
| } | ||
|
|
||
| if (!response.ok) { | ||
| const error = await response.text(); | ||
| throw new Error(`Ollama API Error ${response.status}: ${error}`); | ||
| } | ||
|
|
||
| const show = (await response.json()) as OllamaShowResponse; | ||
| const fromModelfile = parseEmbeddingLengthFromShow(show); | ||
| if (fromModelfile !== undefined) { | ||
| this.resolvedDimensions = fromModelfile; | ||
| } | ||
| } | ||
|
|
||
| isReady(): boolean { | ||
| return true; | ||
| } | ||
|
|
||
| async embed(text: string): Promise<number[]> { | ||
| const batch = await this.embedBatch([text]); | ||
| return batch[0]; | ||
| } | ||
|
|
||
| async embedBatch(texts: string[]): Promise<number[][]> { | ||
| if (!texts.length) return []; | ||
|
|
||
| try { | ||
| const response = await fetch(`${this.host}/api/embed`, { | ||
| method: 'POST', | ||
| headers: { 'Content-Type': 'application/json' }, | ||
| body: JSON.stringify({ | ||
| model: this.modelName, | ||
| input: texts.length === 1 ? texts[0] : texts | ||
| }) | ||
| }); | ||
|
Comment on lines
+109
to
+116
Contributor
There was a problem hiding this comment.
Neither new HTTP request has an application timeout. If Ollama stops answering while loading or running a model, indexing keeps waiting rather than reporting a useful failure. Add a configurable deadline to both Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time! |
||
|
|
||
| if (!response.ok) { | ||
| const error = await response.text(); | ||
| if (response.status === 404 && error.includes('not found')) { | ||
| throw new Error( | ||
| `Ollama model "${this.modelName}" is not available. Pull it first: ollama pull ${this.modelName}` | ||
| ); | ||
| } | ||
| throw new Error(`Ollama API Error ${response.status}: ${error}`); | ||
| } | ||
|
|
||
| const data = (await response.json()) as OllamaEmbedResponse; | ||
| if (!Array.isArray(data.embeddings) || data.embeddings.length !== texts.length) { | ||
| throw new Error( | ||
| `Ollama returned ${data.embeddings?.length ?? 0} embeddings for ${texts.length} inputs` | ||
| ); | ||
| } | ||
|
|
||
| if (data.embeddings[0]?.length && data.embeddings[0].length !== this.resolvedDimensions) { | ||
| this.resolvedDimensions = data.embeddings[0].length; | ||
| } | ||
|
|
||
| return data.embeddings; | ||
| } catch (error) { | ||
| console.error('Ollama Embedding Failed:', error); | ||
| throw error; | ||
| } | ||
| } | ||
| } | ||
|
|
||
| function parseEmbeddingLengthFromShow(show: OllamaShowResponse): number | undefined { | ||
| const haystack = `${show.modelfile ?? ''}\n${show.parameters ?? ''}`; | ||
| const match = haystack.match(/embedding[_\s-]?length[\s=:]+(\d+)/i); | ||
| if (!match) { | ||
| return undefined; | ||
| } | ||
| const parsed = parseInt(match[1] ?? '', 10); | ||
| return Number.isNaN(parsed) || parsed <= 0 ? undefined : parsed; | ||
| } | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,39 @@ | ||
| import { afterEach, describe, expect, it, vi } from 'vitest'; | ||
|
|
||
| describe('DEFAULT_EMBEDDING_CONFIG environment', () => { | ||
| afterEach(() => { | ||
| vi.unstubAllEnvs(); | ||
| vi.resetModules(); | ||
| }); | ||
|
|
||
| it('uses OPENAI_BASE_URL when set', async () => { | ||
| vi.stubEnv('EMBEDDING_PROVIDER', 'openai'); | ||
| vi.stubEnv('OPENAI_BASE_URL', 'http://127.0.0.1:11434/v1/'); | ||
| const { DEFAULT_EMBEDDING_CONFIG } = await import('../src/embeddings/types.js'); | ||
| expect(DEFAULT_EMBEDDING_CONFIG.apiEndpoint).toBe('http://127.0.0.1:11434/v1'); | ||
| }); | ||
|
|
||
| it('defaults to the official OpenAI endpoint when OPENAI_BASE_URL is unset', async () => { | ||
| vi.stubEnv('EMBEDDING_PROVIDER', 'openai'); | ||
| vi.unstubAllEnvs(); | ||
| vi.stubEnv('EMBEDDING_PROVIDER', 'openai'); | ||
| const { DEFAULT_EMBEDDING_CONFIG } = await import('../src/embeddings/types.js'); | ||
| expect(DEFAULT_EMBEDDING_CONFIG.apiEndpoint).toBe('https://api.openai.com/v1'); | ||
| }); | ||
|
|
||
| it('uses OLLAMA_HOST when provider is ollama', async () => { | ||
| vi.stubEnv('EMBEDDING_PROVIDER', 'ollama'); | ||
| vi.stubEnv('OLLAMA_HOST', 'http://127.0.0.1:11434/'); | ||
| const { DEFAULT_EMBEDDING_CONFIG } = await import('../src/embeddings/types.js'); | ||
| expect(DEFAULT_EMBEDDING_CONFIG.provider).toBe('ollama'); | ||
| expect(DEFAULT_EMBEDDING_CONFIG.apiEndpoint).toBe('http://127.0.0.1:11434'); | ||
| expect(DEFAULT_EMBEDDING_CONFIG.model).toBe('nomic-embed-text'); | ||
| }); | ||
|
|
||
| it('resolveEmbeddingModel picks provider-specific defaults', async () => { | ||
| const { resolveEmbeddingModel } = await import('../src/embeddings/types.js'); | ||
| expect(resolveEmbeddingModel('ollama', undefined)).toBe('nomic-embed-text'); | ||
| expect(resolveEmbeddingModel('openai', undefined)).toBe('text-embedding-3-small'); | ||
| expect(resolveEmbeddingModel('transformers', undefined)).toBe('Xenova/bge-small-en-v1.5'); | ||
| }); | ||
| }); |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
getConfiguredDimensionskeeps using a fixed guess after Ollama returns the actual vector size. If a custom model returns 1024 values andOLLAMA_EMBEDDING_DIMENSIONSis unset, a full build succeeds. A later incremental refresh compares those stored vectors against 768 and throwsIndexCorruptedError.Use the resolved vector size for the incremental check, or require an explicit size before building an unknown model.
Knowledge Base Used: Index storage and integrity