Adjust tokens budget calculation in deep thinking tool by reducing an additional 1000 tokens

Author: JegernOUTT

refact-agent/engine/src/tools/tool_deep_thinking.rs

@@ -28,7 +28,7 @@ async fn _make_prompt(
     let caps = try_load_caps_quickly_if_not_present(gcx.clone(), 0).await.map_err(|x| x.message)?;
     let tokenizer = cached_tokenizers::cached_tokenizer(caps, gcx.clone(), subchat_params.subchat_model.to_string()).await
         .map_err(|e| ScratchError::new(StatusCode::INTERNAL_SERVER_ERROR, format!("Error loading tokenizer: {}", e))).map_err(|x| x.message)?;
-    let mut tokens_budget: i64 = (subchat_params.subchat_n_ctx - subchat_params.subchat_max_new_tokens - subchat_params.subchat_tokens_for_rag) as i64;
+    let mut tokens_budget: i64 = (subchat_params.subchat_n_ctx - subchat_params.subchat_max_new_tokens - subchat_params.subchat_tokens_for_rag - 1000) as i64;
     let final_message = format!("***Problem:***\n{problem_statement}\n\n***Problem context:***\n");
     tokens_budget -= count_tokens(&tokenizer.read().unwrap(), &final_message) as i64;
     let mut context = "".to_string();