Parameters

Sampling parameters pass through to the upstream model under each protocol's native field names. The tables below map the common parameters to the three protocols; "—" means the protocol has no such parameter. Moyi API does not validate value ranges; out-of-range values are handled by the upstream model.

Temperature

Controls randomness; lower values are more deterministic.

Anthropic OpenAI Gemini
temperature temperature generationConfig.temperature

Top P

Nucleus sampling: sample only from the tokens whose cumulative probability reaches this value.

Anthropic OpenAI Gemini
top_p top_p generationConfig.topP

Top K

Sample only from the K most likely tokens.

Anthropic OpenAI Gemini
top_k top_k (passed through; not part of the standard OpenAI API, effect depends on the model) generationConfig.topK

Frequency Penalty

Penalizes tokens by how often they already appear in the output.

Anthropic OpenAI Gemini
frequency_penalty

Presence Penalty

Penalizes tokens that have appeared at all in the output.

Anthropic OpenAI Gemini
presence_penalty

Seed

Fixes the random seed so identical requests produce output as close to identical as possible.

Anthropic OpenAI Gemini
seed generationConfig.seed

Max Tokens

Maximum output tokens.

Anthropic OpenAI Gemini
max_tokens (required) max_tokens generationConfig.maxOutputTokens
Ready? Three steps to startLog in to the console · top up · create an API key
DiscordGet community help instantly
Parameters · Docs