fix lint

serve/mlc_serve/model/dummy_model.py

@@ -11,7 +11,6 @@
     DecodeRequest,
     KVCache,
     PrefillRequest,
-    RequestType,
     SequenceId,
     TextGenerationResult,
 )
@@ -98,18 +97,17 @@ def get_max_new_tokens(self) -> int:
 class DummyTextGenerator:
     def generate(
         self,
-        requests: list[RequestType],
+        requests: list[Union[PrefillRequest, DecodeRequest]e],
         kv_cache: DummyCache,
     ) -> list[TextGenerationResult]:
         result = []
         for req in requests:
-            # TODO(vvchernov): support other types of Request
             if isinstance(req, DecodeRequest):
                 seq_id = req.sequence_id
                 request_id = req.sequence_id.request_id
                 if req.sequence_id.sequence_index > 0:
                     raise RuntimeError("Multiple generated sequences not supported")
-            else:  # PrefillRequest
+            else:
                 seq_id = SequenceId(req.request_id, 0)
                 request_id = req.request_id
 

serve/mlc_serve/model/model_common.py

@@ -351,11 +351,11 @@ def sample_from_logits(
     assert logits.shape[0] == len(requests)
 
     sampling_params = [req.sampling_params for req in requests]
+    outputs: List[TextGenerationResult] = []
 
     try:
         next_tokens, logprob_infos = sample(logits, sampling_params, vocab_size)
         assert next_tokens is not None
-        outputs = []
         for i, (sequence_id, new_token) in enumerate(zip(sequence_ids, next_tokens)):
             update_tokens_frequency(requests[i], new_token)
             outputs = append_text_gen_res(
@@ -369,7 +369,6 @@ def sample_from_logits(
         return outputs
     except RuntimeError:
         # Fallback to per-token sampling in case some logits values are corrupted.
-        outputs = []
         err_msg = (
             "Error from sampling: probability tensor contains either `inf`, `nan`"
             " or element < 0"