[Rust Frontend] Fix DeepSeek V3.2 continue_final_message rendering (#45155)

Signed-off-by: reidliu41 <reid201711@gmail.com>

rust/src/chat/src/renderer/deepseek_v32/encoding.rs

@@ -49,6 +49,7 @@ pub(super) fn render_request(request: &ChatRequest) -> Result<String> {
     let last_user_render_index =
         find_last_user_render_index(request.messages.as_slice(), render_offset);
     let last_user_actual_index = find_last_user_actual_index(request.messages.as_slice());
+    let continue_final_message = request.chat_options.continue_final_message();
     let mut prompt = String::from(BOS_TOKEN);
 
     if request.tool_parsing_enabled() {
@@ -66,6 +67,7 @@ pub(super) fn render_request(request: &ChatRequest) -> Result<String> {
             last_user_actual_index,
             thinking_mode,
             drop_thinking,
+            continue_final_message,
         )?;
     }
 
@@ -96,6 +98,7 @@ fn render_message(
     last_user_actual_index: usize,
     thinking_mode: ThinkingMode,
     drop_thinking: bool,
+    continue_final_message: bool,
 ) -> Result<()> {
     let render_index = message_index as isize + render_offset;
     let opens_thinking = render_index == last_user_render_index;
@@ -125,9 +128,7 @@ fn render_message(
                 thinking_mode,
                 drop_thinking,
             ),
-            // TODO: Respect `continue_final_message` and map it to DeepSeek's
-            // prefix-style final-assistant continuation behavior.
-            false,
+            continue_final_message && message_index + 1 == messages.len(),
         ),
         ChatMessage::ToolResponse { content, .. } => render_tool_message(
             out,

rust/src/chat/src/renderer/deepseek_v32/tests.rs

@@ -404,6 +404,24 @@ fn assistant_after_last_user_requires_reasoning_or_tool_calls() {
     expect!["chat template error: invalid DeepSeek V3.2 assistant message after last user message: expected reasoning or tool calls"]
         .assert_eq(&error.to_report_string());
 }
+
+#[test]
+fn continue_final_assistant_omits_final_eos() {
+    let mut request = ChatRequest {
+        messages: vec![
+            ChatMessage::user("write"),
+            ChatMessage::assistant_text("partial answer"),
+        ],
+        ..ChatRequest::for_test()
+    };
+    request.chat_options.generation_prompt_mode = GenerationPromptMode::ContinueFinalAssistant;
+
+    let rendered = render_request(&request);
+
+    expect!["<｜begin▁of▁sentence｜><｜User｜>write<｜Assistant｜></think>partial answer"]
+        .assert_eq(&rendered);
+}
+
 #[test]
 fn render_rejects_multimodal_input() {
     let request = ChatRequest {