@llm-translate/cli 1.0.0-next.1

package/src/providers/interface.ts

@@ -0,0 +1,109 @@
+import type { ProviderName } from '../types/index.js';
+
+// ============================================================================
+// Chat Message Types
+// ============================================================================
+
+export type ChatRole = 'system' | 'user' | 'assistant';
+
+/**
+ * Content part with optional cache control for prompt caching
+ */
+export interface CacheableTextPart {
+  type: 'text';
+  text: string;
+  /** Enable prompt caching for this content part (Claude only) */
+  cacheControl?: { type: 'ephemeral' };
+}
+
+export interface ChatMessage {
+  role: ChatRole;
+  /** Content can be a string or array of cacheable parts */
+  content: string | CacheableTextPart[];
+}
+
+// ============================================================================
+// Request/Response Types
+// ============================================================================
+
+export interface ChatRequest {
+  messages: ChatMessage[];
+  model?: string;
+  temperature?: number;
+  maxTokens?: number;
+}
+
+export interface ChatResponse {
+  content: string;
+  usage: {
+    inputTokens: number;
+    outputTokens: number;
+    /** Tokens read from cache (90% cost reduction) */
+    cacheReadTokens?: number;
+    /** Tokens written to cache (25% cost increase for first write) */
+    cacheWriteTokens?: number;
+  };
+  model: string;
+  finishReason: 'stop' | 'length' | 'error';
+}
+
+// ============================================================================
+// Model Information
+// ============================================================================
+
+export interface ModelInfo {
+  maxContextTokens: number;
+  supportsStreaming: boolean;
+  costPer1kInput?: number;
+  costPer1kOutput?: number;
+}
+
+// ============================================================================
+// Provider Configuration
+// ============================================================================
+
+export interface ProviderConfig {
+  apiKey?: string;
+  baseUrl?: string;
+  defaultModel?: string;
+  timeout?: number;
+}
+
+// ============================================================================
+// LLM Provider Interface
+// ============================================================================
+
+export interface LLMProvider {
+  readonly name: ProviderName;
+
+  /**
+   * The default model used by this provider
+   */
+  readonly defaultModel: string;
+
+  /**
+   * Send a chat request and receive a complete response
+   */
+  chat(request: ChatRequest): Promise<ChatResponse>;
+
+  /**
+   * Send a chat request and receive a streaming response
+   */
+  stream(request: ChatRequest): AsyncIterable<string>;
+
+  /**
+   * Count the number of tokens in a text string
+   */
+  countTokens(text: string): number;
+
+  /**
+   * Get information about a specific model
+   */
+  getModelInfo(model?: string): ModelInfo;
+}
+
+// ============================================================================
+// Provider Factory Type
+// ============================================================================
+
+export type ProviderFactory = (config: ProviderConfig) => LLMProvider;

package/src/providers/ollama.ts

@@ -0,0 +1,379 @@
+import { createOpenAI } from '@ai-sdk/openai';
+import { generateText, streamText } from 'ai';
+import type { ProviderName } from '../types/index.js';
+import type {
+  LLMProvider,
+  ProviderConfig,
+  ChatRequest,
+  ChatResponse,
+  ModelInfo,
+} from './interface.js';
+import { TranslationError, ErrorCode } from '../errors.js';
+import { estimateTokens } from '../utils/tokens.js';
+
+// ============================================================================
+// Model Information
+// ============================================================================
+
+// Common Ollama models with their context sizes
+// Note: These are estimates - actual limits depend on model variant and system memory
+const MODEL_INFO: Record<string, ModelInfo> = {
+  // Llama 3.x models
+  'llama3.3': {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  'llama3.2': {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  'llama3.1': {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  'llama3': {
+    maxContextTokens: 8192,
+    supportsStreaming: true,
+  },
+  // Llama 2 models
+  llama2: {
+    maxContextTokens: 4096,
+    supportsStreaming: true,
+  },
+  'llama2:13b': {
+    maxContextTokens: 4096,
+    supportsStreaming: true,
+  },
+  'llama2:70b': {
+    maxContextTokens: 4096,
+    supportsStreaming: true,
+  },
+  // Mistral models
+  mistral: {
+    maxContextTokens: 32768,
+    supportsStreaming: true,
+  },
+  'mistral-nemo': {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  mixtral: {
+    maxContextTokens: 32768,
+    supportsStreaming: true,
+  },
+  // Qwen models
+  qwen2: {
+    maxContextTokens: 32768,
+    supportsStreaming: true,
+  },
+  'qwen2.5': {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  'qwen2.5-coder': {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  // Gemma models
+  gemma2: {
+    maxContextTokens: 8192,
+    supportsStreaming: true,
+  },
+  gemma: {
+    maxContextTokens: 8192,
+    supportsStreaming: true,
+  },
+  // Phi models
+  phi3: {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  'phi3:mini': {
+    maxContextTokens: 128000,
+    supportsStreaming: true,
+  },
+  // Code models
+  codellama: {
+    maxContextTokens: 16384,
+    supportsStreaming: true,
+  },
+  'deepseek-coder': {
+    maxContextTokens: 16384,
+    supportsStreaming: true,
+  },
+  // Other popular models
+  'neural-chat': {
+    maxContextTokens: 8192,
+    supportsStreaming: true,
+  },
+  vicuna: {
+    maxContextTokens: 2048,
+    supportsStreaming: true,
+  },
+};
+
+// Default to llama3.2 for better multilingual support
+const DEFAULT_MODEL = 'llama3.2';
+const DEFAULT_BASE_URL = 'http://localhost:11434';
+
+// ============================================================================
+// Ollama Provider Implementation
+// ============================================================================
+
+export class OllamaProvider implements LLMProvider {
+  readonly name: ProviderName = 'ollama';
+  readonly defaultModel: string;
+  private readonly client: ReturnType<typeof createOpenAI>;
+  private readonly baseUrl: string;
+
+  constructor(config: ProviderConfig = {}) {
+    this.baseUrl =
+      config.baseUrl ??
+      process.env['OLLAMA_BASE_URL'] ??
+      DEFAULT_BASE_URL;
+
+    // Ollama uses OpenAI-compatible API at /v1 endpoint
+    this.client = createOpenAI({
+      apiKey: 'ollama', // Ollama doesn't require an API key
+      baseURL: `${this.baseUrl}/v1`,
+    });
+
+    this.defaultModel = config.defaultModel ?? DEFAULT_MODEL;
+  }
+
+  async chat(request: ChatRequest): Promise<ChatResponse> {
+    const model = request.model ?? this.defaultModel;
+
+    try {
+      await this.ensureModelAvailable(model);
+
+      const messages = this.convertMessages(request.messages);
+
+      const result = await generateText({
+        model: this.client(model),
+        messages,
+        temperature: request.temperature ?? 0,
+        maxTokens: request.maxTokens ?? 4096,
+      });
+
+      return {
+        content: result.text,
+        usage: {
+          inputTokens: result.usage?.promptTokens ?? 0,
+          outputTokens: result.usage?.completionTokens ?? 0,
+        },
+        model,
+        finishReason: mapFinishReason(result.finishReason),
+      };
+    } catch (error) {
+      throw this.handleError(error, model);
+    }
+  }
+
+  /**
+   * Convert messages to Vercel AI SDK format
+   * Ollama doesn't support cache control, so we simplify content
+   */
+  private convertMessages(
+    messages: Array<{
+      role: 'system' | 'user' | 'assistant';
+      content: string | Array<{ type: 'text'; text: string }>;
+    }>
+  ) {
+    return messages.map((msg) => {
+      // If content is an array of parts, concatenate text
+      if (Array.isArray(msg.content)) {
+        return {
+          role: msg.role,
+          content: msg.content.map((part) => part.text).join(''),
+        };
+      }
+      return { role: msg.role, content: msg.content };
+    });
+  }
+
+  async *stream(request: ChatRequest): AsyncIterable<string> {
+    const model = request.model ?? this.defaultModel;
+
+    try {
+      await this.ensureModelAvailable(model);
+
+      const messages = this.convertMessages(request.messages);
+
+      const result = streamText({
+        model: this.client(model),
+        messages,
+        temperature: request.temperature ?? 0,
+        maxTokens: request.maxTokens ?? 4096,
+      });
+
+      for await (const chunk of result.textStream) {
+        yield chunk;
+      }
+    } catch (error) {
+      throw this.handleError(error, model);
+    }
+  }
+
+  countTokens(text: string): number {
+    // Use estimation for Ollama models
+    return estimateTokens(text);
+  }
+
+  getModelInfo(model?: string): ModelInfo {
+    const modelName = model ?? this.defaultModel;
+
+    // Try exact match first
+    if (MODEL_INFO[modelName]) {
+      return MODEL_INFO[modelName];
+    }
+
+    // Try base model name (e.g., "llama3.2:7b" -> "llama3.2")
+    const baseModel = modelName.split(':')[0] ?? modelName;
+    if (baseModel && MODEL_INFO[baseModel]) {
+      return MODEL_INFO[baseModel];
+    }
+
+    // Default fallback
+    return {
+      maxContextTokens: 4096,
+      supportsStreaming: true,
+    };
+  }
+
+  /**
+   * Check if the Ollama server is running and the model is available
+   */
+  private async ensureModelAvailable(model: string): Promise<void> {
+    try {
+      const response = await fetch(`${this.baseUrl}/api/tags`);
+
+      if (!response.ok) {
+        throw new TranslationError(ErrorCode.PROVIDER_ERROR, {
+          provider: 'ollama',
+          message: `Ollama server not responding at ${this.baseUrl}`,
+        });
+      }
+
+      const data = (await response.json()) as {
+        models?: Array<{ name: string }>;
+      };
+      const models = data.models ?? [];
+      const modelNames = models.map((m) => m.name);
+
+      // Check for exact match or base model match
+      const modelExists = modelNames.some(
+        (name) => name === model || name.startsWith(`${model}:`)
+      );
+
+      if (!modelExists) {
+        throw new TranslationError(ErrorCode.PROVIDER_ERROR, {
+          provider: 'ollama',
+          model,
+          availableModels: modelNames.slice(0, 10), // Show first 10
+          message: `Model "${model}" not found. Pull it with: ollama pull ${model}`,
+        });
+      }
+    } catch (error) {
+      if (error instanceof TranslationError) {
+        throw error;
+      }
+
+      // Connection refused or other network error
+      throw new TranslationError(ErrorCode.PROVIDER_ERROR, {
+        provider: 'ollama',
+        baseUrl: this.baseUrl,
+        message: `Cannot connect to Ollama server at ${this.baseUrl}. Is Ollama running?`,
+      });
+    }
+  }
+
+  private handleError(error: unknown, model?: string): TranslationError {
+    if (error instanceof TranslationError) {
+      return error;
+    }
+
+    const errorMessage =
+      error instanceof Error ? error.message : String(error);
+
+    // Check for connection errors
+    if (
+      errorMessage.includes('ECONNREFUSED') ||
+      errorMessage.includes('fetch failed') ||
+      errorMessage.includes('network')
+    ) {
+      return new TranslationError(ErrorCode.PROVIDER_ERROR, {
+        provider: 'ollama',
+        baseUrl: this.baseUrl,
+        message: `Cannot connect to Ollama server at ${this.baseUrl}. Is Ollama running?`,
+      });
+    }
+
+    // Check for model not found
+    if (
+      errorMessage.includes('model') &&
+      errorMessage.includes('not found')
+    ) {
+      return new TranslationError(ErrorCode.PROVIDER_ERROR, {
+        provider: 'ollama',
+        model,
+        message: `Model "${model}" not found. Pull it with: ollama pull ${model}`,
+      });
+    }
+
+    // Check for context length errors
+    if (
+      errorMessage.includes('context') ||
+      errorMessage.includes('too long')
+    ) {
+      return new TranslationError(ErrorCode.CHUNK_TOO_LARGE, {
+        provider: 'ollama',
+        model,
+        message: errorMessage,
+      });
+    }
+
+    // Check for out of memory
+    if (
+      errorMessage.includes('out of memory') ||
+      errorMessage.includes('OOM')
+    ) {
+      return new TranslationError(ErrorCode.PROVIDER_ERROR, {
+        provider: 'ollama',
+        model,
+        message: 'Out of memory. Try a smaller model or reduce chunk size.',
+      });
+    }
+
+    return new TranslationError(ErrorCode.PROVIDER_ERROR, {
+      provider: 'ollama',
+      message: errorMessage,
+    });
+  }
+}
+
+// ============================================================================
+// Helper Functions
+// ============================================================================
+
+function mapFinishReason(
+  reason: string | null | undefined
+): 'stop' | 'length' | 'error' {
+  switch (reason) {
+    case 'stop':
+      return 'stop';
+    case 'length':
+    case 'max_tokens':
+      return 'length';
+    default:
+      return 'error';
+  }
+}
+
+// ============================================================================
+// Factory Function
+// ============================================================================
+
+export function createOllamaProvider(config: ProviderConfig = {}): LLMProvider {
+  return new OllamaProvider(config);
+}