Struct ChatCompletionsRequest 

pub struct ChatCompletionsRequest {

    pub model: String,
    pub messages: Vec<ChatMessage>,
    pub max_tokens: Option<u32>,
    pub max_completion_tokens: Option<u32>,
    pub temperature: Option<f32>,
    pub top_p: Option<f32>,
    pub top_k: Option<i64>,
    pub min_p: Option<f32>,
    pub repetition_penalty: Option<f32>,
    pub n: Option<u32>,
    pub stream: Option<bool>,
    pub ignore_eos: Option<bool>,
    pub stop: Option<Vec<String>>,
    pub presence_penalty: Option<f32>,
    pub frequency_penalty: Option<f32>,
    pub logit_bias: Option<HashMap<String, f32>>,
    pub logprobs: Option<bool>,
    pub top_logprobs: Option<u32>,
    pub user: Option<String>,
    pub seed: Option<u64>,
    pub response_format: Option<OpenAiResponseFormat>,
    pub tools: Option<Vec<ChatTool>>,
    pub tool_choice: Option<ToolChoice>,
    pub stream_options: Option<StreamOptions>,
    pub functions: Option<Vec<ChatFunction>>,
    pub function_call: Option<FunctionCallChoice>,
    pub metadata: Option<HashMap<String, Value>>,
    pub chat_template_kwargs: Option<HashMap<String, Value>>,
}

Chat completions request (OpenAI compatible)

Fields

model: String

Model to use for completion

messages: Vec<ChatMessage>

List of messages

max_tokens: Option<u32>

Maximum number of tokens to generate

max_completion_tokens: Option<u32>

Newer OpenAI chat field replacing max_tokens for completion budget. When both are supplied, Ferrum uses this value.

temperature: Option<f32>

Temperature for sampling

top_p: Option<f32>

Top-p for nucleus sampling

top_k: Option<i64>

vLLM-compatible top-k sampling extension. Values -1 and 0 disable top-k filtering.

min_p: Option<f32>

vLLM-compatible minimum probability sampling extension. A value of 0 disables minimum-probability filtering.

repetition_penalty: Option<f32>

vLLM-compatible repetition penalty extension.

n: Option<u32>

Number of completions to generate

stream: Option<bool>

Whether to stream responses

ignore_eos: Option<bool>

vLLM-compatible extension for benchmark/throughput workloads. When true, Ferrum ignores model EOS tokens and stops only on the requested token budget or explicit user stop sequences.

stop: Option<Vec<String>>

Stop sequences

presence_penalty: Option<f32>

Presence penalty

frequency_penalty: Option<f32>

Frequency penalty

logit_bias: Option<HashMap<String, f32>>

Logit bias

logprobs: Option<bool>

Return log probabilities. Ferrum rejects this until implemented so clients get an explicit OpenAI-style error instead of silent ignore.

top_logprobs: Option<u32>

Number of top log probabilities to return.

user: Option<String>

User identifier

seed: Option<u64>

Random seed

response_format: Option<OpenAiResponseFormat>

Response format constraint (e.g., {"type": "json_object"})

tools: Option<Vec<ChatTool>>

OpenAI tool definitions. Function tools are parsed, carried through structured request data, and can shape model-emitted tool-call JSON. Tool execution itself stays caller-owned.

tool_choice: Option<ToolChoice>

OpenAI tool selection policy.

stream_options: Option<StreamOptions>

Streaming response options.

functions: Option<Vec<ChatFunction>>

Legacy OpenAI functions compatibility.

function_call: Option<FunctionCallChoice>

Legacy OpenAI function-call selector.

metadata: Option<HashMap<String, Value>>

Ferrum extension metadata. Used for opt-in product features such as metadata.ferrum_session_id when callers prefer body metadata over the X-Ferrum-Session header.

chat_template_kwargs: Option<HashMap<String, Value>>

vLLM-compatible chat-template variables. Ferrum forwards supported values to the model-provided chat template; templates that do not read a variable are unaffected.

Trait Implementations

impl Clone for ChatCompletionsRequest

fn clone(&self) -> ChatCompletionsRequest

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ChatCompletionsRequest

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ChatCompletionsRequest

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ChatCompletionsRequest

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.