feat(providers): Add vision and document processing capabilities to providers
This commit is contained in:
@@ -6,18 +6,21 @@ import type { ChatOptions, ChatResponse, ChatMessage } from './abstract.classes.
|
||||
export interface IOllamaProviderOptions {
|
||||
baseUrl?: string;
|
||||
model?: string;
|
||||
visionModel?: string; // Model to use for vision tasks (e.g. 'llava')
|
||||
}
|
||||
|
||||
export class OllamaProvider extends MultiModalModel {
|
||||
private options: IOllamaProviderOptions;
|
||||
private baseUrl: string;
|
||||
private model: string;
|
||||
private visionModel: string;
|
||||
|
||||
constructor(optionsArg: IOllamaProviderOptions = {}) {
|
||||
super();
|
||||
this.options = optionsArg;
|
||||
this.baseUrl = optionsArg.baseUrl || 'http://localhost:11434';
|
||||
this.model = optionsArg.model || 'llama2';
|
||||
this.visionModel = optionsArg.visionModel || 'llava';
|
||||
}
|
||||
|
||||
async start() {
|
||||
@@ -167,4 +170,83 @@ export class OllamaProvider extends MultiModalModel {
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
throw new Error('Audio generation is not supported by Ollama.');
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
const base64Image = optionsArg.image.toString('base64');
|
||||
|
||||
const response = await fetch(`${this.baseUrl}/api/chat`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.visionModel,
|
||||
messages: [{
|
||||
role: 'user',
|
||||
content: optionsArg.prompt,
|
||||
images: [base64Image]
|
||||
}],
|
||||
stream: false
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.statusText}`);
|
||||
}
|
||||
|
||||
const result = await response.json();
|
||||
return result.message.content;
|
||||
}
|
||||
|
||||
public async document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: ChatMessage[];
|
||||
}): Promise<{ message: any }> {
|
||||
// Convert PDF documents to images using SmartPDF
|
||||
const smartpdfInstance = new plugins.smartpdf.SmartPdf();
|
||||
let documentImageBytesArray: Uint8Array[] = [];
|
||||
|
||||
for (const pdfDocument of optionsArg.pdfDocuments) {
|
||||
const documentImageArray = await smartpdfInstance.convertPDFToPngBytes(pdfDocument);
|
||||
documentImageBytesArray = documentImageBytesArray.concat(documentImageArray);
|
||||
}
|
||||
|
||||
// Convert images to base64
|
||||
const base64Images = documentImageBytesArray.map(bytes => Buffer.from(bytes).toString('base64'));
|
||||
|
||||
// Send request to Ollama with images
|
||||
const response = await fetch(`${this.baseUrl}/api/chat`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.visionModel,
|
||||
messages: [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory,
|
||||
{
|
||||
role: 'user',
|
||||
content: optionsArg.userMessage,
|
||||
images: base64Images
|
||||
}
|
||||
],
|
||||
stream: false
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.statusText}`);
|
||||
}
|
||||
|
||||
const result = await response.json();
|
||||
return {
|
||||
message: {
|
||||
role: 'assistant',
|
||||
content: result.message.content
|
||||
}
|
||||
};
|
||||
}
|
||||
}
|
Reference in New Issue
Block a user