[Bugfix] Fix use_cascade_attention handling for Alibi-based models on vllm/v1 (#15211)

Signed-off-by: h-sugi <h.sugi@ieee.org>
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

vllm/utils.py

@@ -61,7 +61,7 @@ import vllm.envs as envs
 from vllm.logger import enable_trace_function_call, init_logger
 
 if TYPE_CHECKING:
-    from vllm.config import VllmConfig
+    from vllm.config import ModelConfig, VllmConfig
 
 logger = init_logger(__name__)
 
@@ -2498,6 +2498,18 @@ def cprofile(save_file: Optional[str] = None, enabled: bool = True):
     return decorator
 
 
+# Only relevant for models using ALiBi (e.g, MPT)
+def check_use_alibi(model_config: ModelConfig) -> bool:
+    return (getattr(model_config.hf_text_config, "alibi", False)  # Falcon
+            or ("BloomForCausalLM" in getattr(model_config.hf_config,
+                                              "architectures", []))  # Bloom
+            or getattr(model_config.hf_text_config, "position_encoding_type",
+                       "") == "alibi"  # codellm_1b_alibi
+            or
+            (hasattr(model_config.hf_text_config, "attn_config")  # MPT
+             and model_config.hf_text_config.attn_config.get("alibi", False)))
+
+
 def sha256(input) -> int:
     """Hash any picklable Python object using SHA-256.
 

vllm/v1/worker/gpu_model_runner.py

@@ -25,7 +25,7 @@ from vllm.multimodal.utils import group_mm_inputs_by_modality
 from vllm.sampling_params import SamplingType
 from vllm.sequence import IntermediateTensors
 from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, DeviceMemoryProfiler,
-                        LayerBlockType, LazyLoader, cdiv,
+                        LayerBlockType, LazyLoader, cdiv, check_use_alibi,
                         is_pin_memory_available)
 from vllm.v1.attention.backends.flash_attn import FlashAttentionMetadata
 from vllm.v1.core.encoder_cache_manager import compute_encoder_budget
@@ -223,6 +223,9 @@ class GPUModelRunner(LoRAModelRunnerMixin):
                 device="cpu",
                 pin_memory=self.pin_memory)
 
+        # Only relevant for models using ALiBi (e.g, MPT)
+        self.use_alibi = check_use_alibi(model_config)
+
         self.inputs_embeds = torch.zeros(
             (self.max_num_tokens, self.hidden_size),
             dtype=self.dtype,
@@ -689,7 +692,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
             query_lens=num_scheduled_tokens,
             num_query_heads=self.num_query_heads,
             num_kv_heads=self.num_kv_heads,
-            use_alibi=False,  # FIXME
+            use_alibi=self.use_alibi,
             use_sliding_window=self.window_size is not None,
             num_sms=self.num_sms,
         )