Fix some typing issues found by mypy==1.18.2 (#26596)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

tests/v1/core/test_scheduler.py

@@ -807,8 +807,10 @@ def test_schedule_spec_decoding_stats(spec_tokens, output_tokens, expected):
         engine_core_outputs[0].scheduler_stats if engine_core_outputs else None
     )
     if expected[0] == 0:
+        assert scheduler_stats is not None
         assert scheduler_stats.spec_decoding_stats is None
     else:
+        assert scheduler_stats is not None
         assert scheduler_stats.spec_decoding_stats is not None
         stats = scheduler_stats.spec_decoding_stats
         assert stats.num_drafts == expected[0]

vllm/model_executor/layers/linear.py

@@ -1229,10 +1229,10 @@ class QKVParallelLinear(ColumnParallelLinear):
 
             param_data = param_data.narrow(output_dim, shard_offset, shard_size)
             if loaded_shard_id == "q":
-                shard_id = self.tp_rank
+                shard_rank = self.tp_rank
             else:
-                shard_id = self.tp_rank // self.num_kv_head_replicas
+                shard_rank = self.tp_rank // self.num_kv_head_replicas
-            start_idx = shard_id * shard_size
+            start_idx = shard_rank * shard_size
 
             if not is_sharded_weight:
                 loaded_weight = loaded_weight.narrow(output_dim, start_idx, shard_size)

vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_24.py

@@ -49,16 +49,16 @@ class CompressedTensors24(CompressedTensorsScheme):
         self.quantized = quantized
         self.weight_quant = weight_quant
         self.input_quant = input_quant
-        self.model_compressor = (
+        model_compressor = ModelCompressor.from_compression_config(
-            ModelCompressor.from_compression_config(model_compression_config)
+            model_compression_config
-            if model_compression_config is not None
-            else None
         )
         self.do_sparse_decompress = (
-            self.model_compressor is not None
+            model_compressor is not None
-            and self.model_compressor.sparsity_config.format
+            and model_compressor.sparsity_config.format
             == CompressionFormat.sparse_24_bitmask.value
         )
+        if self.do_sparse_decompress:
+            self.model_compressor = model_compressor
 
         if (
             quantized

vllm/model_executor/layers/resampler.py

@@ -200,12 +200,10 @@ class BaseResampler(nn.Module):
         self.ln_q = norm_layer(embed_dim)
         self.ln_kv = norm_layer(embed_dim)
         self.do_post_projection = do_post_projection
-        self.ln_post = norm_layer(embed_dim) if do_post_projection else None
+        if self.do_post_projection:
-        self.proj = (
+            self.ln_post = norm_layer(embed_dim)
-            nn.Parameter((embed_dim**-0.5) * torch.empty(embed_dim, embed_dim))
+            data = (embed_dim**-0.5) * torch.empty(embed_dim, embed_dim)
-            if do_post_projection
+            self.proj = nn.Parameter(data=data)
-            else None
-        )
 
     def _repeat(self, query, N: int):
         return query.unsqueeze(1).repeat(1, N, 1)

vllm/model_executor/model_loader/bitsandbytes_loader.py

@@ -542,8 +542,7 @@ class BitsAndBytesModelLoader(BaseModelLoader):
             )
 
         quant_config = getattr(model_config.hf_config, "quantization_config", None)
-        if quant_config is not None:
+        if quant_config and (quant_method := quant_config.get("quant_method")):
-            quant_method = quant_config.get("quant_method")
             if quant_method == "bitsandbytes":
                 self.pre_quant = True
             else:
@@ -558,7 +557,7 @@ class BitsAndBytesModelLoader(BaseModelLoader):
                 "Prequant BitsAndBytes models with tensor parallelism is not "
                 "supported. Please try with pipeline parallelism."
             )
-        if self.pre_quant:
+        if quant_config and self.pre_quant:
             self.load_8bit = quant_config.get("load_in_8bit", False)
 
     def _initialize_loader_state(

vllm/transformers_utils/processors/ovis2_5.py

@@ -397,6 +397,8 @@ class Ovis2_5Processor(ProcessorMixin):
                     images.append(image)
             elif isinstance(video, list):
                 images = video
+        else:
+            raise ValueError("Either images or video should be provided.")
         min_pixels = min(
             max_pixels if max_pixels is not None else MAX_PIXELS,
             min_pixels if min_pixels is not None else MIN_PIXELS,