lib/src/limits.js

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
exports.TokenLimits = void 0;
class TokenLimits {
    maxTokens;
    requestTokens;
    responseTokens;
    knowledgeCutOff;
    constructor(model = 'llama-3.1-8b-instant') {
        this.knowledgeCutOff = '2023-12-01';
        // Default for modern large context models
        this.maxTokens = 128000;
        this.responseTokens = 16000;
        if (model.includes('kimi-k2')) {
            this.maxTokens = 256000;
            this.responseTokens = 16000;
        }
        else if (model.includes('qwen') || model.includes('llama-4')) {
            this.maxTokens = 128000;
            this.responseTokens = 16000;
        }
        else if (model.includes('gpt-oss-120b')) {
            this.maxTokens = 128000;
            this.responseTokens = 16000;
        }
        else if (model.includes('3.2-11b')) {
            this.maxTokens = 128000;
            this.responseTokens = 16000;
        }
        else if (model.includes('llama3') ||
            model.includes('llama-3') ||
            model.includes('groq') ||
            model.includes('mixtral')) {
            this.maxTokens = 128000;
            this.responseTokens = 16000;
        }
        else {
            this.maxTokens = 128000;
            this.responseTokens = 4000;
        }
        // Provide some margin
        this.requestTokens = this.maxTokens - this.responseTokens - 500;
    }
    string() {
        return `max_tokens=${this.maxTokens}, request_tokens=${this.requestTokens}, response_tokens=${this.responseTokens}`;
    }
}
exports.TokenLimits = TokenLimits;
//# sourceMappingURL=limits.js.map