@infernetprotocol/nim-adapter

NVIDIA NIM / build.nvidia.com adapter for the Infernet Protocol.

Treats NVIDIA's OpenAI-compatible inference endpoint as a fallback "virtual provider." When the P2P network has no live providers available for a chat job, we stream from NIM instead so the UX never breaks during the bootstrap phase.

Environment

NVIDIA_NIM_API_KEY=...                                  # required
NVIDIA_NIM_API_URL=https://integrate.api.nvidia.com/v1  # optional override
NVIDIA_NIM_DEFAULT_MODEL=meta/llama-3.3-70b-instruct    # optional override

Grab a free API key at build.nvidia.com → click any model → Get API Key on the right sidebar.

Usage

import { streamChatCompletion, isNimConfigured, nimVirtualProvider } from "@infernetprotocol/nim-adapter";

if (isNimConfigured()) {
    for await (const ev of streamChatCompletion({
        messages: [{ role: "user", content: "hi" }],
        maxTokens: 256,
        temperature: 0.7
    })) {
        if (ev.type === "token") process.stdout.write(ev.data.text);
        if (ev.type === "done")  console.log();
    }
}

Yielded events:

• { type: 'meta', data: { provider: 'nvidia-nim', model, started_at } }
• { type: 'token', data: { text } } (repeating)
• { type: 'done', data: { text, finished_at } }
• { type: 'error', data: { message } } — terminal

Same shape as Infernet's own job_events stream, so the chat SSE route can relay NIM output without an adapter layer on the server side.