test: use max_batch_size=2 for test_free_generation

Parameterize _make_engine_config with max_batch_size (default=1) so that
test_free_generation can create its own pipeline with max_batch_size=2,
while all other tests continue using the shared pipe fixture with
max_batch_size=1.

Author: windreamer

tests/test_lmdeploy/test_mtp_guided_decoding.py

@@ -76,10 +76,10 @@ def _make_spec_config():
     return SpeculativeConfig(method='qwen3_5_mtp', num_speculative_tokens=1)
 
 
-def _make_engine_config():
+def _make_engine_config(max_batch_size=1):
     """PytorchEngineConfig suitable for MTP + guided decoding tests."""
     return PytorchEngineConfig(
-        max_batch_size=1,
+        max_batch_size=max_batch_size,
         session_len=1024,
         cache_max_entry_count=0.1,
     )
@@ -242,11 +242,20 @@ def test_mixed_json_and_free(self, pipe):
 class TestSpecDecodeNoGuided:
     """MTP speculative decoding without guided decoding — baseline sanity."""
 
-    def test_free_generation(self, pipe):
-        gen_config = GenerationConfig(max_new_tokens=50)
-        response = pipe(PROMPT, gen_config=gen_config)
-        assert response.text, 'Free generation should produce text'
-        assert response.generate_token_len > 0
+    def test_free_generation(self):
+        p = pipeline(
+            MTP_MODEL_ID,
+            backend_config=_make_engine_config(max_batch_size=2),
+            speculative_config=_make_spec_config(),
+            log_level='INFO',
+        )
+        try:
+            gen_config = GenerationConfig(max_new_tokens=50)
+            response = p(PROMPT, gen_config=gen_config)
+            assert response.text, 'Free generation should produce text'
+            assert response.generate_token_len > 0
+        finally:
+            p.close()
 
 
 # ---------------------------------------------------------------------------