NVIDIA: Llama Nemotron Rerank VL 1B V2 (free)

Характеристики

Контекстное окно: 10K токенов
Провайдер: 10240
Архитектура: text+image->rerank

Модальности

Входящие

text
image

Исходящие

rerank

Цены

Оплата по факту за использованные токены и операции. Цены в рублях (за 1 млн токенов или за единицу — как указано).

Тип	Стоимость	Описание
Входящие токены	0.00 ₽ / 1M	Токены запроса и контекста, обработанные моделью
Исходящие токены	0.00 ₽ / 1M	Токены ответа, сгенерированные моделью

Цены могут меняться в зависимости от курса валют. Данные по тарифам поступают от поставщика услуг и обновляются каждые сутки.

Начать использовать

Описание

Here's a thinking process:

Analyze User Input:

Source text: "Llama Nemotron Rerank VL 1B V2 is a 1.7B multimodal reranking model from NVIDIA. It evaluates the relevance of document images and text against user queries, designed for vision RAG..."
Task: Translate into Russian
Constraints:
- Keep model names and technical terms in the "familiar for Russian-speaking developers" style (usually means transliterating/keeping English terms, or using standard Russian AI/ML terminology, but keeping model names as-is in English/Latin script is typical)
- Output only the text translation, no introduction, no quotes around the whole text
Model name in source: NVIDIA: Llama Nemotron Rerank VL 1B V2 (free) (nvidia/llama-nemotron-rerank-vl-1b-v2:free)
Model name in target: Should keep the naming convention familiar to Russian devs. Typically, model names like "Llama Nemotron Rerank VL 1B V2" are kept in English/Latin, maybe with slight transliteration if needed, but standard practice in Russian tech community is to keep model names in English. I'll keep the model name as given, or adapt it slightly while keeping the technical terms intact.
Technical terms: "multimodal reranking model", "document images and text", "user queries", "vision RAG" - these should be translated or adapted appropriately, but keeping the "familiar for Russian-speaking developers" vibe means either keeping them in English or using standard Russian equivalents that devs use. Usually, "reranking" is often left as "reranking" or translated as "ранжирование" or "переранжирование". "Vision RAG" might become "визуальный RAG" or kept as "RAG". I'll translate the descriptive part but keep model names and key acronyms/terms in English/Latin as is customary.

Let me re-read the constraint: "Сохраняй имена моделей и технические термины в привычном для русскоязычных разработчиков виде." This means: Keep model names and technical terms in the format familiar to Russian-speaking developers. This typically means: model names in English (Latin script), technical terms either in English or with standard Russian translations that are widely accepted. I'll keep the model name exactly as provided, and translate the description, adapting technical terms to natural Russian AI terminology while keeping the model name intact.

Source description: "Llama Nemotron Rerank VL 1B V2 is a 1.7B multimodal reranking model from NVIDIA. It evaluates the relevance of document images and text against user queries, designed for vision RAG..."

Let's break it down:

"Llama Nemotron Rerank VL 1B V2 is a 1.7B multimodal reranking model from NVIDIA." → "Llama Nemotron Rerank VL 1B V2 — это модель reranking multimодального типа объемом 1.7B от NVIDIA." Or keep "multimodal" as "многомодальный", "reranking" as "ранжирование" or "reranking". Russian devs often say "reranking" or "ранжирование". I'll use "ранжирование" or keep "reranking" if it's a proper noun/term. I think translating to "ранжирование" is fine, but to be safe and "familiar", I might keep "reranking" in parentheses or use the Russian equivalent. The prompt says "technical terms in familiar format". I'll use standard Russian AI terms: "многомодальная", "ранжирование", "визуальный RAG", but keep model name and "1.7B", "NVIDIA".

Let's craft the translation carefully: "Llama Nemotron Rerank VL 1B V2 — это модель reranking multimодального типа объемом 1.7B от NVIDIA. Она оценивает релевантность документов, изображений и текста в relation к запросам пользователей, предназначена для vision RAG..."

Wait, "vision RAG" - I'll keep "RAG" as is, maybe add "визуальный" or just "RAG". Standard in Russian: "RAG" is often kept, or translated as "поиск с уточнением", but keeping "RAG" is fine.

Let's look at the exact phrasing expected: "Ответь только текстом перевода, без вступления и без кавычек вокруг всего текста."

So I should output just the translated text, no quotes at the start/end, no "Translation:".

Model name: NVIDIA: Llama Nemotron Rerank VL 1B V2 (free) (nvidia/llama-nemotron-rerank-vl-1b-v2:free) - I should keep this exactly as is, or integrate it naturally. The prompt says "Keep model names... in familiar format". I'll keep the model name block as given, probably at the start or as part of the text. But the source text starts with "Модель: NVIDIA: Llama Nemotron Rerank VL 1B V2 (free) (nvidia/llama-nemotron-rerank-vl-1b-v2:free)" and then the description. The user gave: "Модель: NVIDIA: Llama Nemotron Rerank VL 1B V2 (free) (nvidia/llama-nemotron-rerank-vl-1b-v2:free)" followed by "Оригинал: ..." Wait, the user message is: "Модель: NVIDIA: Llama Nemotron Rerank VL 1B V2 (free) (nvidia/llama-nemotron-rerank-vl-1b-v2:free)

Оригинал: Llama Nemotron Rerank VL 1B V2 is a 1.7B multimodal reranking model from NVIDIA. It evaluates the relevance of document images and text against user queries, designed f…

API и примеры кода

Единый API совместим с OpenAI SDK. Используйте ключ из личного кабинета.

main.py

from openai import OpenAI

client = OpenAI(
    base_url="https://routerapi.ru/api/v1",
    api_key="YOUR_API_KEY",
)

response = client.chat.completions.create(
    model="nvidia/llama-nemotron-rerank-vl-1b-v2:free",
    messages=[{"role": "user", "content": "Привет!"}],
)

print(response.choices[0].message.content)

terminal

curl -X POST "https://routerapi.ru/api/v1/chat/completions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nvidia/llama-nemotron-rerank-vl-1b-v2:free",
    "messages": [
      {"role": "user", "content": "Привет!"}
    ]
  }'

request.php

<?php
declare(strict_types=1);

$response = (new \GuzzleHttp\Client())->post(
    'https://routerapi.ru/api/v1/chat/completions',
    [
        'headers' => [
            'Authorization' => 'Bearer YOUR_API_KEY',
            'Content-Type'  => 'application/json',
        ],
        'json' => [
            'model'    => 'nvidia/llama-nemotron-rerank-vl-1b-v2:free',
            'messages' => [
                ['role' => 'user', 'content' => 'Привет!'],
            ],
        ],
    ]
);

$data = json_decode((string) $response->getBody(), true);
echo $data['choices'][0]['message']['content'];

main.go

package main

import (
    "bytes"
    "encoding/json"
    "fmt"
    "io"
    "net/http"
)

func main() {
    body, _ := json.Marshal(map[string]any{
        "model": "nvidia/llama-nemotron-rerank-vl-1b-v2:free",
        "messages": []map[string]string{
            {"role": "user", "content": "Привет!"},
        },
    })

    req, _ := http.NewRequest("POST",
        "https://routerapi.ru/api/v1/chat/completions",
        bytes.NewReader(body))
    req.Header.Set("Authorization", "Bearer YOUR_API_KEY")
    req.Header.Set("Content-Type", "application/json")

    resp, _ := http.DefaultClient.Do(req)
    defer resp.Body.Close()

    var result map[string]any
    raw, _ := io.ReadAll(resp.Body)
    json.Unmarshal(raw, &result)

    choices := result["choices"].([]any)
    msg := choices[0].(map[string]any)["message"].(map[string]any)
    fmt.Println(msg["content"])
}

Совместимо с OpenAI Python SDK, Node.js SDK, curl и любым HTTP-клиентом. Документация API · Все модели