[Rust Frontend] Pass effective reasoning_parser_kwargs for structured output (#46360)

Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
This commit is contained in:
Bugen Zhao
2026-06-24 03:13:44 +00:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 4ed8eaafb0
commit ce9f64020b
19 changed files with 165 additions and 34 deletions
+10 -1
View File
@@ -50,7 +50,7 @@ mod request;
mod stream;
use vllm_engine_core_client::EngineCoreClient;
use vllm_engine_core_client::protocol::ModelDtype;
use vllm_engine_core_client::protocol::{ModelDtype, ReasoningParserKwargs};
use vllm_llm::Llm;
use vllm_text::{Prompt, TextLlm, TextRequest};
@@ -179,6 +179,14 @@ impl ChatLlm {
},
)?;
let rendered = self.backend.chat_renderer().render(&request)?;
let reasoning_parser_kwargs =
request
.sampling_params
.structured_outputs
.is_some()
.then(|| ReasoningParserKwargs {
chat_template_kwargs: rendered.effective_template_kwargs.clone(),
});
let (prompt, mm_features) = multimodal::finalize_rendered_prompt(
&request,
@@ -199,6 +207,7 @@ impl ChatLlm {
cache_salt: request.cache_salt,
add_special_tokens: request.add_special_tokens,
data_parallel_rank: request.data_parallel_rank,
reasoning_parser_kwargs,
lora_request: request.lora_request,
};
let decoded_stream = self.text.generate(text_request).await?.map_err(Error::from).boxed();