[Rust Frontend][CI] Unblock more end-to-end test cases (#47735)

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
2026-07-07 08:27:20 +00:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent ba50b9763f
commit e55cc59e52
14 changed files with 105 additions and 40 deletions
+6 -2
View File
@@ -456,13 +456,17 @@ pub struct ServerUnsupportedArgs {
/// Enable the `/tokenizer_info` endpoint. May expose chat
/// templates and other tokenizer configuration.
///
/// Accepted as a no-op: the Rust frontend serves `/tokenize` and
/// `/detokenize`, but does not implement `/tokenizer_info` yet.
#[arg(
long,
visible_alias = "no-enable-tokenizer-info-endpoint",
default_missing_value = "true",
num_args = 0..=1
num_args = 0..=1,
hide = true
)]
pub enable_tokenizer_info_endpoint: Option<Unsupported>,
pub enable_tokenizer_info_endpoint: Option<Noop>,
/// If set to True, log model outputs (generations).
/// Requires `--enable-log-requests`. As with `--enable-log-requests`,
@@ -29,7 +29,9 @@ pub struct GenerateRequest {
impl Normalizable for GenerateRequest {}
/// Mirrors the Python vLLM `GenerateResponseChoice` class.
#[serde_with::skip_serializing_none]
///
/// Do not skip serializing `None` fields here: non-streaming response types
/// should serialize `None` as explicit `null`.
#[derive(Debug, Clone, Serialize)]
pub(super) struct GenerateResponseChoice {
pub index: u32,
@@ -58,7 +60,6 @@ pub(super) struct GenerateStreamResponse {
}
/// Mirrors the Python vLLM `GenerateResponse` class.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub(super) struct GenerateResponse {
pub request_id: String,
@@ -68,7 +69,6 @@ pub(super) struct GenerateResponse {
}
/// Mirrors the Python vLLM `Logprob` class used in prompt-logprobs payloads.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub(super) struct GenerateLogprob {
pub logprob: f32,
@@ -211,7 +211,7 @@ async fn collect_chat_completion(
Some(prefix) => Some(format!("{prefix}{}", message.text())),
None => Some(message.text()).filter(|t| !t.is_empty()),
},
tool_calls: Some(tool_calls).filter(|calls| !calls.is_empty()),
tool_calls,
reasoning: if include_reasoning { reasoning } else { None },
},
logprobs,
@@ -830,7 +830,7 @@ mod tests {
let message = ChatCompletionMessage {
role: AssistantRole,
content: Some("answer".to_string()),
tool_calls: None,
tool_calls: Vec::new(),
reasoning: Some("inner".to_string()),
};
let message_json = serde_json::to_value(message).expect("message serializes");
@@ -331,7 +331,9 @@ impl Normalizable for ChatCompletionRequest {
}
/// Mirrors the Python vLLM `ChatCompletionResponse` class.
#[serde_with::skip_serializing_none]
///
/// Do not skip serializing `None` fields here: non-streaming response types
/// should serialize `None` as explicit `null`.
#[derive(Debug, Clone, Serialize)]
pub(super) struct ChatCompletionResponse {
pub id: String,
@@ -347,7 +349,6 @@ pub(super) struct ChatCompletionResponse {
}
/// Mirrors the Python vLLM `ChatCompletionResponseChoice` class.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub(super) struct ChatCompletionChoice {
pub index: u32,
@@ -370,12 +371,12 @@ impl fmt::Display for AssistantRole {
}
/// Mirrors the Python vLLM response `ChatMessage` class.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub(super) struct ChatCompletionMessage {
pub role: AssistantRole,
pub content: Option<String>,
pub tool_calls: Option<Vec<ToolCall>>,
#[serde(skip_serializing_if = "Vec::is_empty")]
pub tool_calls: Vec<ToolCall>,
pub reasoning: Option<String>,
}
@@ -196,7 +196,9 @@ impl Normalizable for CompletionRequest {
}
/// Mirrors the Python vLLM `CompletionResponse` class.
#[serde_with::skip_serializing_none]
///
/// Do not skip serializing `None` fields here: non-streaming response types
/// should serialize `None` as explicit `null`.
#[derive(Debug, Clone, Serialize)]
pub(super) struct CompletionResponse {
pub id: String,
@@ -210,7 +212,6 @@ pub(super) struct CompletionResponse {
}
/// Mirrors the Python vLLM `CompletionResponseChoice` class.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub(super) struct CompletionChoice {
pub index: u32,
@@ -311,7 +311,9 @@ pub enum MessageContent {
// ============================================================================
/// Mirrors the Python vLLM `UsageInfo` class.
#[serde_with::skip_serializing_none]
///
/// Do not skip serializing `None` fields here: non-streaming response types
/// should serialize `None` as explicit `null`.
#[derive(Debug, Clone, Serialize)]
pub struct Usage {
pub prompt_tokens: usize,
@@ -402,14 +404,12 @@ pub struct LogProbs {
}
/// Mirrors the Python vLLM `ChatCompletionLogProbs` class.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub struct ChatLogProbs {
pub content: Option<Vec<ChatLogProbsContent>>,
}
/// Mirrors the Python vLLM `ChatCompletionLogProbsContent` class.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub struct ChatLogProbsContent {
pub token: String,
@@ -419,7 +419,6 @@ pub struct ChatLogProbsContent {
}
/// Mirrors the Python vLLM `ChatCompletionLogProb` class.
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Serialize)]
pub struct TopLogProb {
pub token: String,
@@ -436,7 +435,6 @@ pub struct ErrorResponse {
pub error: ErrorDetail,
}
#[serde_with::skip_serializing_none]
#[derive(Debug, Clone, Deserialize, Serialize)]
pub struct ErrorDetail {
pub message: String,
+49 -8
View File
@@ -2191,6 +2191,28 @@ async fn non_stream_chat_returns_json_response() {
assert_eq!(json["usage"]["prompt_tokens"], 22);
assert_eq!(json["usage"]["completion_tokens"], 3);
assert_eq!(json["usage"]["total_tokens"], 25);
// Unset optional fields are serialized as explicit `null` on
// non-streaming responses...
let response_object = json.as_object().expect("response object");
let choice = json["choices"][0].as_object().expect("choice object");
let message = choice["message"].as_object().expect("message object");
for (object, key) in [
(response_object, "system_fingerprint"),
(response_object, "prompt_token_ids"),
(response_object, "kv_transfer_params"),
(choice, "logprobs"),
(choice, "stop_reason"),
(choice, "token_ids"),
(message, "reasoning"),
] {
assert!(
object.contains_key(key) && object[key].is_null(),
"expected explicit null `{key}`: {json}"
);
}
// ...except `tool_calls`, which Python pops from the payload when empty.
assert!(!message.contains_key("tool_calls"), "{json}");
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
@@ -2956,6 +2978,25 @@ async fn non_stream_completions_return_json_response() {
assert_eq!(json["choices"][0]["text"], "hi");
assert_eq!(json["choices"][0]["finish_reason"], "stop");
assert_eq!(json["usage"]["completion_tokens"], 3);
// Unset optional fields are serialized as explicit `null` on
// non-streaming responses.
let response_object = json.as_object().expect("response object");
let choice = json["choices"][0].as_object().expect("choice object");
for (object, key) in [
(response_object, "system_fingerprint"),
(response_object, "kv_transfer_params"),
(choice, "logprobs"),
(choice, "stop_reason"),
(choice, "prompt_logprobs"),
(choice, "token_ids"),
(choice, "prompt_token_ids"),
] {
assert!(
object.contains_key(key) && object[key].is_null(),
"expected explicit null `{key}`: {json}"
);
}
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
@@ -4371,10 +4412,10 @@ async fn include_reasoning_false_suppresses_reasoning_in_non_stream_chat() {
let json: serde_json::Value = serde_json::from_str(&text).expect("decode json");
assert_eq!(json["choices"][0]["message"]["content"], "answer");
// Suppressed fields are serialized as explicit `null` on non-streaming
// responses.
assert!(
json["choices"][0]["message"]
.as_object()
.is_some_and(|message| !message.contains_key("reasoning")),
json["choices"][0]["message"]["reasoning"].is_null(),
"{text}"
);
}
@@ -4476,14 +4517,14 @@ async fn include_reasoning_false_suppresses_non_stream_output_metadata() {
let choice = json["choices"][0].as_object().expect("choice object");
assert_eq!(json["choices"][0]["message"]["content"], "answer");
// Suppressed fields are serialized as explicit `null` on non-streaming
// responses.
assert!(
json["choices"][0]["message"]
.as_object()
.is_some_and(|message| !message.contains_key("reasoning")),
json["choices"][0]["message"]["reasoning"].is_null(),
"{text}"
);
assert!(!choice.contains_key("logprobs"), "{text}");
assert!(!choice.contains_key("token_ids"), "{text}");
assert!(choice["logprobs"].is_null(), "{text}");
assert!(choice["token_ids"].is_null(), "{text}");
assert!(json["prompt_token_ids"].is_array(), "{text}");
}
+2 -1
View File
@@ -102,12 +102,13 @@ pub struct DetokenizeRequest {
pub tokens: Vec<u32>,
}
/// Do not skip serializing `None` fields here: non-streaming response types
/// should serialize `None` as explicit `null`.
#[derive(Debug, Clone, Serialize)]
pub struct TokenizeResponse {
pub count: usize,
pub max_model_len: u32,
pub tokens: Vec<u32>,
#[serde(skip_serializing_if = "Option::is_none")]
pub token_strs: Option<Vec<String>>,
}
+1 -1
View File
@@ -10,7 +10,7 @@ pub enum TokenIdsError {
#[error("allowed_token_ids should not be empty")]
EmptyAllowedTokenIds,
#[error(
"token_id(s) {token_ids:?} in {parameter} contain out-of-vocab token ids. \
"token_id(s) {token_ids:?} in {parameter} are out of vocabulary. \
Vocabulary size: {vocab_size}"
)]
OutOfVocab {