YandexGPT API Integration: IAM, Models, RAG for 152-FZ

When launching AI features in the Russian market, you face a choice: use foreign APIs with a risk of blocking or local models. YandexGPT from Yandex is one option, but its integration requires understanding IAM authentication, model selection, and building a RAG pipeline. We have implemented such integrations for several projects, including with government customers. Our engineers have 10+ years of experience in ML and NLP, so we share proven solutions.

Here's a typical pain point: you deploy an MVP on a foreign API, and a month later you are blocked due to sanctions. With YandexGPT, this problem does not exist — servers are located in Russia, data does not leave the country. But to ensure a smooth integration, you will need to handle IAM tokens, context windows, and asynchronous calls. Contact us to discuss your scenario — we will help with model selection and configuration.

How to set up IAM authentication?

YandexGPT API requires authentication via an IAM token or a service account API key. The token lives 12 hours, the API key is perpetual. For production, use automatic token refresh via the SDK. According to Yandex Cloud Documentation, the IAM token must be refreshed every 12 hours. Example setup:

import requests
import json

FOLDER_ID = "your-folder-id"
IAM_TOKEN = "your-iam-token"  # Refreshed every 12 hours
# Or API_KEY for service account

How to work with the API: synchronous and asynchronous requests?

Synchronous call via REST API:

def yandexgpt_chat(
    prompt: str,
    model: str = "yandexgpt",
    temperature: float = 0.1,
    max_tokens: int = 2000,
) -> str:
    url = "https://llm.api.cloud.yandex.net/foundationModels/v1/completion"

    headers = {
        "Authorization": f"Api-Key {API_KEY}",
        "x-folder-id": FOLDER_ID,
    }

    body = {
        "modelUri": f"gpt://{FOLDER_ID}/{model}",
        "completionOptions": {
            "stream": False,
            "temperature": temperature,
            "maxTokens": max_tokens,
        },
        "messages": [
            {"role": "user", "text": prompt}
        ]
    }

    response = requests.post(url, headers=headers, json=body)
    response.raise_for_status()

    return response.json()["result"]["alternatives"][0]["message"]["text"]

With system prompt:

def yandexgpt_with_system(system: str, user_prompt: str) -> str:
    url = "https://llm.api.cloud.yandex.net/foundationModels/v1/completion"

    body = {
        "modelUri": f"gpt://{FOLDER_ID}/yandexgpt",
        "completionOptions": {"stream": False, "temperature": 0.1, "maxTokens": 2000},
        "messages": [
            {"role": "system", "text": system},
            {"role": "user", "text": user_prompt}
        ]
    }

    response = requests.post(
        url,
        headers={"Authorization": f"Api-Key {API_KEY}", "x-folder-id": FOLDER_ID},
        json=body,
    )
    return response.json()["result"]["alternatives"][0]["message"]["text"]

Asynchronous calls via the official YandexGPT SDK:

from yandex_cloud_ml_sdk import YCloudML

sdk = YCloudML(folder_id=FOLDER_ID, auth=API_KEY)
model = sdk.models.completions("yandexgpt")

# Synchronous
result = model.configure(temperature=0.5).run("Tell me about Moscow")

# Async
result = await model.configure(temperature=0.5).run_async("Request")

# Streaming
for event in model.configure(temperature=0.5).run_stream("Long request"):
    print(event.alternatives[0].text, end="")

Model configuration: parameters and available options

For more details on parameters, see the official Yandex Cloud documentation.

What are Yandex embeddings and why are they needed?

Yandex embeddings are vector representations of text, used for semantic search and RAG. Two types: text-search-doc for document indexing and text-search-query for search queries. Example retrieval:

def get_yandex_embedding(text: str, embedding_type: str = "text-search-doc") -> list[float]:
    response = requests.post(
        "https://llm.api.cloud.yandex.net/foundationModels/v1/textEmbedding",
        headers={"Authorization": f"Api-Key {API_KEY}", "x-folder-id": FOLDER_ID},
        json={
            "modelUri": f"emb://{FOLDER_ID}/{embedding_type}",
            "text": text,
        }
    )
    return response.json()["embedding"]

From practice: RAG pipeline for a state enterprise

One client — a state enterprise with strict data localization requirements. We deployed an automatic response system for citizens' inquiries. YandexGPT was chosen because:

• data does not leave the Russian Federation (152-FZ compliance);
• integration with Yandex SpeechKit for voice input;
• high quality in Russian.

We configured IAM authentication, implemented asynchronous requests, and built a RAG pipeline with Yandex embeddings and pgvector. Response time decreased by 60%, and infrastructure costs — by 40% due to prompt optimization and caching. Savings in query processing time allowed fast ROI. Get a consultation for your project — we will evaluate YandexGPT applicability and design the architecture.

Process

1. Analytics: audit your project, identify YandexGPT usage scenarios.
2. Design: integration architecture, model selection, security setup.
3. Implementation: coding, authentication configuration, integration with existing services.
4. Testing: load testing, p99 latency checks, response quality.
5. Deployment: production rollout, monitoring, documentation.

Estimated timelines

• Basic REST integration: 1 to 2 days.
• SDK integration with async/streaming: 2 to 3 days.
• Full RAG pipeline with embeddings: 1 to 2 weeks.

Cost is calculated individually — contact us for a project evaluation.

What is included

• Configured API access with IAM authentication.
• Ready integration code (REST/SDK) in Python.
• Documentation for use and maintenance.
• Training your team on YandexGPT.
• Technical support during implementation.

Typical mistakes when integrating YandexGPT

• Incorrect IAM token refresh: the token lives 12 hours, it must be updated automatically.
• Ignoring the context window: trim history for long requests.
• Lack of error handling: API may return 429 (rate limit) — implement retry with exponential backoff.
• Suboptimal prompts: for Russian, use clear instructions and examples (few-shot).

Our certified engineers guarantee quality and SLA 99.9%. Contact us to discuss your project — get a free consultation.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.