Use the min context limit for Answer input token limit.

Bug: 372535307
Change-Id: I3c8ff10d896e309c34b997d3d22793ccbd5fbf1b
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5950645


Commit-Queue: Steven Holte <holte@chromium.org>
Reviewed-by: Jun Zou <junzou@chromium.org>
Auto-Submit: Steven Holte <holte@chromium.org>
Reviewed-by: Zekun Jiang <zekunjiang@google.com>
Reviewed-by: Orin Jaworski <orinj@chromium.org>
Cr-Commit-Position: refs/heads/main@{#1372250}

components/history_embeddings/ml_answerer.cc

@@ -141,8 +141,7 @@ class MlAnswerer::SessionManager {
                              base::OnceCallback<void(int)> session_added_cb,
                              std::vector<ModelInput> inputs) {
     HistoryAnswerRequest request;
-    // TODO(crbug.com/372535307): use actual model limit.
-    int token_limit = 1024;
+    int token_limit = session->GetTokenLimits().min_context_tokens;
     // Reserve space for preamble text.
     int token_count = kPreambleTokenBufferSize;
 

components/history_embeddings/ml_answerer_unittest.cc

@@ -43,8 +43,23 @@ class HistoryEmbeddingsMlAnswererTest : public testing::Test {
  public:
   void SetUp() override {
     ml_answerer_ = std::make_unique<MlAnswerer>(&model_executor_);
+    token_limits_ = {
+        .min_context_tokens = 1024,
+    };
+
+
+    ON_CALL(session_1_, GetTokenLimits())
+        .WillByDefault([&]() -> optimization_guide::TokenLimits& {
+          return GetTokenLimits();
+        });
+    ON_CALL(session_2_, GetTokenLimits())
+        .WillByDefault([&]() -> optimization_guide::TokenLimits& {
+          return GetTokenLimits();
+        });
   }
 
+  optimization_guide::TokenLimits& GetTokenLimits() { return token_limits_; }
+
  protected:
   optimization_guide::StreamingResponse MakeResponse(
       const std::string& answer_text,
@@ -61,6 +76,7 @@ class HistoryEmbeddingsMlAnswererTest : public testing::Test {
   testing::NiceMock<MockModelExecutor> model_executor_;
   std::unique_ptr<MlAnswerer> ml_answerer_;
   testing::NiceMock<optimization_guide::MockSession> session_1_, session_2_;
+  optimization_guide::TokenLimits token_limits_;
 };
 
 TEST_F(HistoryEmbeddingsMlAnswererTest, ComputeAnswerNoSession) {