pub mod engine;
pub mod error;
pub mod http;
pub mod prompt;
pub mod request;
pub mod response;
pub mod sse;
pub mod toolcall;
pub mod wire;
pub use engine::{
Completion, Engine, EngineConfig, ModelEngine, NullSink, Plan, StreamError, TokenSink,
common_prefix, hush_stream_aborts,
};
pub use error::{ErrorDetail, ErrorResponse, RETRY_HEADERS, ServerError};
pub use http::{
BODY_BYTES_PER_CONTEXT_TOKEN, MIN_BODY_BYTES, ServeConfig, ServeError,
max_body_bytes_for_context, serve, serve_with,
};
pub use prompt::{Prompt, check_context};
pub use request::{
ChatCompletionRequest, Content, ContentPart, Message, MessageRole, StreamOptions,
StringOrArray, ToolCall,
};
pub use response::{
ChatCompletion, ChatCompletionChunk, ChunkBuilder, Delta, FinishReason, Health, ModelList,
ResponseMessage, ToolCallDelta, Usage,
};
pub use sse::{DONE_SENTINEL, sse_data, sse_done, sse_event};
pub use toolcall::{
CallStream, Extracted, ParsedCall, Step, call_id, extract, wire_call, wire_calls,
};
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn a_request_walks_the_whole_pure_path() -> Result<(), ServerError> {
let request = ChatCompletionRequest::from_json(
r#"{"model":"qwen3-30b-a3b",
"messages":[
{"role":"system","content":"be terse"},
{"role":"user","content":[{"type":"text","text":"2+2?"}]}
],
"stream":true,
"stream_options":{"include_usage":true},
"max_completion_tokens":8}"#,
)?;
request.validate()?;
assert!(request.stream());
assert!(request.include_usage());
assert_eq!(request.max_new_tokens(), Some(8));
let mapped = prompt::chat_messages(&request.messages)?;
assert_eq!(mapped.len(), 2);
assert_eq!(mapped[1].content, "2+2?");
check_context(64, 8, 4096)?;
let builder = ChunkBuilder::new("chatcmpl-1", 0, &request.model);
let stream = format!(
"{}{}{}{}{}",
sse_event(&builder.role())?,
sse_event(&builder.content("4"))?,
sse_event(&builder.finish(FinishReason::Stop))?,
sse_event(&builder.usage(Usage::new(20, 1)))?,
sse_done(),
);
assert!(stream.ends_with("data: [DONE]\n\n"));
assert_eq!(stream.matches("\n\n").count(), 5);
Ok(())
}
}