/**
* @fileoverview NvidiaAdapter - NVIDIA NIM API adapters
* @module backends/nvidia-adapter
*
* Adapters for NVIDIA cloud backends:
* - NVIDIA DeepSeek (reasoning)
* - Qwen3 Coder 480B (coding)
*
* Smart AI Bridge v2.0.0
*/
import { BackendAdapter } from './backend-adapter.js';
const NVIDIA_BASE_URL = 'https://integrate.api.nvidia.com/v1/chat/completions';
/**
* Calculate dynamic timeout based on requested max_tokens
* @param {number} maxTokens - Maximum tokens to generate
* @param {boolean} thinking - Whether thinking/reasoning mode is enabled
* @returns {number} Timeout in milliseconds
*/
function calculateDynamicTimeout(maxTokens, thinking = false) {
const baseMs = maxTokens * 40;
const withThinking = thinking ? baseMs * 1.5 : baseMs;
return Math.min(Math.max(60000, withThinking), 600000);
}
/**
* NVIDIA DeepSeek adapter (reasoning model)
*/
class NvidiaDeepSeekAdapter extends BackendAdapter {
constructor(config = {}) {
super({
name: 'nvidia_deepseek',
type: 'nvidia',
url: NVIDIA_BASE_URL,
apiKey: config.apiKey || process.env.NVIDIA_API_KEY,
maxTokens: config.maxTokens || 8192,
timeout: config.timeout || 120000,
streaming: true,
...config
});
this.model = 'deepseek-ai/deepseek-v3.1-terminus';
this.fallbackModel = 'deepseek-ai/deepseek-v3.2';
}
async makeRequest(prompt, options = {}) {
if (!this.config.apiKey) {
throw new Error('NVIDIA_API_KEY not configured');
}
try {
return await this._executeRequest(prompt, this.model, options);
} catch (error) {
const isTimeout = error.name === 'TimeoutError' || error.message.includes('timeout') || error.message.includes('aborted');
const isServerError = error.message.includes('500') || error.message.includes('502') || error.message.includes('503') || error.message.includes('Internal Server Error');
if (isTimeout || isServerError) {
const reason = isTimeout ? 'timed out' : 'server error';
console.error(`[SAB] NVIDIA DeepSeek ${reason}, falling back to V3.1-terminus...`);
return await this._executeRequest(prompt, this.fallbackModel, options, true);
}
throw error;
}
}
async _executeRequest(prompt, model, options = {}, isFallback = false) {
const isTerminus = model.includes('terminus');
const body = {
model: model,
messages: [{ role: 'user', content: prompt }],
max_tokens: options.maxTokens || this.config.maxTokens,
temperature: isTerminus ? 0.2 : (options.temperature || 1),
top_p: isTerminus ? 0.7 : 0.95,
stream: false
};
if (options.thinking !== false) {
body.extra_body = { chat_template_kwargs: { thinking: true } };
}
const requestedTokens = options.maxTokens || this.config.maxTokens;
const baseTimeout = isFallback ? 60000 : calculateDynamicTimeout(requestedTokens, options.thinking);
const timeout = options.timeout
|| (process.env.NVIDIA_TIMEOUT ? parseInt(process.env.NVIDIA_TIMEOUT) : null)
|| baseTimeout;
const response = await fetch(this.config.url, {
method: 'POST',
headers: this.buildHeaders(),
body: JSON.stringify(body),
signal: AbortSignal.timeout(timeout)
});
if (!response.ok) {
const error = await response.text();
throw new Error(`NVIDIA DeepSeek error (${model}): ${response.status} - ${error}`);
}
const data = await response.json();
const result = this.parseResponse(data);
result.metadata = result.metadata || {};
result.metadata.model = model;
result.metadata.wasFallback = isFallback;
return result;
}
async checkHealth() {
const startTime = Date.now();
for (const model of [this.model, this.fallbackModel]) {
try {
const response = await fetch(this.config.url, {
method: 'POST',
headers: this.buildHeaders(),
body: JSON.stringify({
model: model,
messages: [{ role: 'user', content: 'ping' }],
max_tokens: 5
}),
signal: AbortSignal.timeout(10000)
});
if (response.ok) {
this.lastHealth = {
healthy: true,
latency: Date.now() - startTime,
checkedAt: new Date(),
error: null,
activeModel: model
};
return this.lastHealth;
}
} catch (error) {
continue;
}
}
this.lastHealth = {
healthy: false,
latency: Date.now() - startTime,
checkedAt: new Date(),
error: 'Both V3.2 and V3.1-terminus unavailable'
};
return this.lastHealth;
}
}
/**
* Qwen3 Coder 480B adapter
*/
class NvidiaQwenAdapter extends BackendAdapter {
constructor(config = {}) {
super({
name: 'nvidia_qwen',
type: 'nvidia',
url: NVIDIA_BASE_URL,
apiKey: config.apiKey || process.env.NVIDIA_API_KEY,
maxTokens: config.maxTokens || 32768,
timeout: config.timeout || 60000,
streaming: false,
...config
});
this.model = 'qwen/qwen3-coder-480b-a35b-instruct';
}
async makeRequest(prompt, options = {}) {
if (!this.config.apiKey) {
throw new Error('NVIDIA_API_KEY not configured');
}
const body = {
model: this.model,
messages: [{ role: 'user', content: prompt }],
max_tokens: options.maxTokens || this.config.maxTokens,
temperature: options.temperature || 0.7,
top_p: options.top_p || 0.8,
stream: false
};
const requestedTokens = options.maxTokens || this.config.maxTokens;
const timeout = options.timeout
|| (process.env.NVIDIA_TIMEOUT ? parseInt(process.env.NVIDIA_TIMEOUT) : null)
|| calculateDynamicTimeout(requestedTokens, options.thinking);
const response = await fetch(this.config.url, {
method: 'POST',
headers: this.buildHeaders(),
body: JSON.stringify(body),
signal: AbortSignal.timeout(timeout)
});
if (!response.ok) {
const error = await response.text();
throw new Error(`NVIDIA Qwen error: ${response.status} - ${error}`);
}
const data = await response.json();
return this.parseResponse(data);
}
async checkHealth() {
const startTime = Date.now();
try {
const response = await fetch(this.config.url, {
method: 'POST',
headers: this.buildHeaders(),
body: JSON.stringify({
model: this.model,
messages: [{ role: 'user', content: 'ping' }],
max_tokens: 5
}),
signal: AbortSignal.timeout(8000)
});
this.lastHealth = {
healthy: response.ok,
latency: Date.now() - startTime,
checkedAt: new Date(),
error: response.ok ? null : `Status ${response.status}`
};
return this.lastHealth;
} catch (error) {
this.lastHealth = {
healthy: false,
latency: Date.now() - startTime,
checkedAt: new Date(),
error: error.message
};
return this.lastHealth;
}
}
}
export {
NvidiaDeepSeekAdapter,
NvidiaQwenAdapter
};
