How Twinkle Eval Handles SSL Verification and API Timeout Configuration

Twinkle Eval centralizes SSL verification and API timeout controls in config.yaml, applying secure defaults of 600 seconds for timeouts and enabled SSL verification before passing these values directly to the underlying HTTP client.

The ai-twinkle/eval repository provides a flexible evaluation framework for Large Language Models (LLMs) that requires robust network configuration handling. Understanding how Twinkle Eval manages SSL verification and API timeout configuration is essential for deploying the tool in both development and production environments with varying security and performance requirements.

Configuration Defaults in ConfigurationManager

When the configuration loads, the ConfigurationManager._apply_defaults method in twinkle_eval/config.py injects sensible defaults for network security parameters. This ensures the system operates securely even when users omit specific settings in their configuration files.

SSL Verification Settings

By default, Twinkle Eval enables strict SSL certificate verification to maintain secure connections. The disable_ssl_verify parameter defaults to False in twinkle_eval/config.py lines 75‑80, meaning SSL verification remains active unless explicitly disabled. This default protects against man-in-the-middle attacks while allowing flexibility for testing environments with self-signed certificates.

API Timeout Configuration

The framework sets a generous default timeout of 600 seconds (10 minutes) for API requests. This value, also established in twinkle_eval/config.py lines 75‑80, prevents indefinite hanging while accommodating slow LLM responses. Users can override this based on their specific latency requirements and network conditions.

Passing Settings to the LLM Client

The OpenAIModel class in twinkle_eval/models.py translates configuration values into concrete HTTP client behavior. Lines 48‑59 demonstrate how the system constructs both an httpx.Client and an OpenAI client using the validated configuration parameters.

When disable_ssl_verify is set to True, the code instantiates the HTTP client with verify=False, bypassing certificate validation. Conversely, the default False value maintains standard SSL verification. The timeout value passes directly to the OpenAI client constructor via its timeout argument, ensuring consistent request duration limits across all API calls.

Validating Configuration Parameters

Before reaching the client construction stage, twinkle_eval/validators.py enforces type safety and logical constraints on both parameters. The validation logic at lines 85‑87 confirms that timeout is a positive number, while separate checks verify that disable_ssl_verify is strictly a boolean value. This early validation prevents runtime errors and ensures malformed configurations fail fast during initialization.

Practical Configuration Examples

Configure these settings through the central config.yaml file to control network behavior without modifying source code.

# config.yaml – custom SSL & timeout settings

llm_api:
  type: openai
  api_key: YOUR_API_KEY
  base_url: https://api.openai.com/v1
  disable_ssl_verify: true      # skip SSL certificate verification

  timeout: 120                  # 2‑minute request timeout

  max_retries: 2

Once configured, the framework automatically applies these settings when loading the LLM instance:

# Using the configured LLM (no extra code needed)

from twinkle_eval.config import load_config

cfg = load_config()                     # reads the YAML above

llm = cfg["llm_instance"]              # OpenAIModel created by the factory

response = llm.call("Explain quantum entanglement?")
print(response.choices[0].message.content)

Changing disable_ssl_verify to false re‑enables normal certificate checks, while adjusting timeout controls how long the client will wait for a response before raising an error.

Summary

• Centralized configuration: Both SSL verification and timeout settings reside in config.yaml, managed by ConfigurationManager._apply_defaults in twinkle_eval/config.py.
• Secure defaults: SSL verification defaults to enabled (False for disable_ssl_verify) and timeouts default to 600 seconds.
• Client integration: OpenAIModel in twinkle_eval/models.py passes these values directly to httpx.Client and the OpenAI client constructor.
• Validation layer: twinkle_eval/validators.py enforces boolean types for SSL settings and positive numeric values for timeouts at lines 85‑87.
• Flexible deployment: Users can disable SSL verification for testing or adjust timeouts for specific network conditions without code changes.

Frequently Asked Questions

How do I disable SSL certificate verification in Twinkle Eval?

Set disable_ssl_verify: true in your config.yaml under the llm_api section. According to the source code in twinkle_eval/models.py, this passes verify=False to the underlying httpx.Client, bypassing certificate validation for environments with self-signed certificates.

What is the default API timeout and how can I change it?

The default timeout is 600 seconds (10 minutes) as defined in twinkle_eval/config.py. Override this by specifying a timeout value in seconds within your config.yaml file. The OpenAIModel class passes this value directly to the OpenAI client constructor.

Why does Twinkle Eval validate configuration parameters before use?

The validators.py module checks that timeout is a positive number and disable_ssl_verify is a boolean to prevent runtime errors. This validation occurs at initialization, ensuring that malformed configurations fail immediately rather than causing cryptic HTTP errors during LLM calls.

Can I configure different timeout values for different LLM providers?

The current implementation in ai-twinkle/eval uses a unified timeout setting passed through the central configuration. While the base config.yaml structure supports provider-specific settings, the OpenAIModel implementation in twinkle_eval/models.py applies a single timeout value to all HTTP clients created from that configuration.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}