ReliableLLM

Setup

from SimplerLLM.language.llm import LLM, LLMProvider, ReliableLLM

primary = LLM.create(provider=LLMProvider.OPENAI, model_name="gpt-4o")
secondary = LLM.create(provider=LLMProvider.ANTHROPIC, model_name="claude-sonnet-4-20250514")

reliable = ReliableLLM(primary_llm=primary, secondary_llm=secondary)

Parameter	Type	Default	Description
primary_llm	LLM	Required	Primary provider
secondary_llm	LLM	Required	Fallback provider
verbose	bool	False	Enable debug logging
skip_validation	bool	False	Skip provider validation
lazy_validation	bool	False	Defer validation to first call

Validation Modes

By default, both providers are validated at initialization.

# Lazy validation - validates on first request
reliable = ReliableLLM(
    primary_llm=primary,
    secondary_llm=secondary,
    lazy_validation=True
)

# Skip validation entirely
reliable = ReliableLLM(
    primary_llm=primary,
    secondary_llm=secondary,
    skip_validation=True
)

Usage

Basic Generation

response = reliable.generate_response(
    prompt="What is Python?",
    max_tokens=200
)

If the primary provider fails, the secondary is used automatically.

Track Which Provider Was Used

response, provider, model = reliable.generate_response(
    prompt="What is Python?",
    return_provider=True
)

print(f"Response from: {provider.name} ({model})")

Async Generation

import asyncio

async def main():
    response = await reliable.generate_response_async(
        prompt="Explain machine learning",
        max_tokens=300
    )
    return response

result = asyncio.run(main())

Full Response with Metadata

response, provider, model = reliable.generate_response(
    prompt="What is 2+2?",
    return_provider=True,
    full_response=True
)

print(f"Text: {response.generated_text}")
print(f"Tokens: {response.input_token_count} in, {response.output_token_count} out")
print(f"Provider: {provider.name}")

Provider Capabilities

Parameters are automatically filtered based on the provider. Provider-specific parameters are ignored when not supported.

Provider	Supported Parameters
OpenAI	reasoning_effort, timeout, web_search, images, detail
Anthropic	thinking_budget, prompt_caching, cached_input, web_search, images
Gemini	thinking_budget, thinking_level, prompt_caching, cache_id, response_schema, timeout, web_search, images
DeepSeek	thinking, images
Perplexity	timeout, search_domain_filter, search_recency_filter, return_images, return_related_questions, images
OpenRouter	reasoning_effort, timeout, site_url, site_name, images, detail
Ollama	images

Universal parameters (prompt, messages, system_prompt, temperature, max_tokens, top_p, json_mode, full_response) work with all providers.

Note: Enable verbose=True to see which parameters are filtered for each provider.