From 4d7bef05ec206a79dd6be058abf867565f920196 Mon Sep 17 00:00:00 2001 From: Jiwoo Park Date: Wed, 28 Jan 2026 16:58:45 +0900 Subject: [PATCH] fix(attn): remove invalid attention param --- vllm_rbln/attention/backends/flash_attention.py | 3 +-- vllm_rbln/v1/attention/backends/flash_attention.py | 6 +----- 2 files changed, 2 insertions(+), 7 deletions(-) diff --git a/vllm_rbln/attention/backends/flash_attention.py b/vllm_rbln/attention/backends/flash_attention.py index 3d0db93cb..945cd5445 100644 --- a/vllm_rbln/attention/backends/flash_attention.py +++ b/vllm_rbln/attention/backends/flash_attention.py @@ -14,7 +14,7 @@ import math from dataclasses import dataclass -from typing import Any, Dict, List, Optional, Tuple, Type +from typing import Dict, List, Optional, Tuple, Type import torch from vllm.attention.backends.abstract import (AttentionBackend, AttentionImpl, @@ -492,7 +492,6 @@ def __init__( alibi_slopes: Optional[List[float]], sliding_window: Optional[int], kv_cache_dtype: str, - blocksparse_params: Optional[Dict[str, Any]] = None, logits_soft_cap: Optional[float] = None, attn_type: str = AttentionType.DECODER, kv_sharing_target_layer_name: Optional[str] = None, diff --git a/vllm_rbln/v1/attention/backends/flash_attention.py b/vllm_rbln/v1/attention/backends/flash_attention.py index 94a9cad97..da6ef6fb8 100644 --- a/vllm_rbln/v1/attention/backends/flash_attention.py +++ b/vllm_rbln/v1/attention/backends/flash_attention.py @@ -15,7 +15,7 @@ import math from dataclasses import dataclass -from typing import TYPE_CHECKING, Any, Optional +from typing import TYPE_CHECKING, Optional import torch from vllm.attention.backends.abstract import (AttentionBackend, AttentionImpl, @@ -787,7 +787,6 @@ def __init__( alibi_slopes: Optional[list[float]], sliding_window: Optional[int], kv_cache_dtype: str, - blocksparse_params: Optional[dict[str, Any]] = None, logits_soft_cap: Optional[float] = None, attn_type: str = AttentionType.DECODER, kv_sharing_target_layer_name: Optional[str] = None, @@ -799,9 +798,6 @@ def __init__( if kv_sharing_target_layer_name is not None: raise NotImplementedError("KV sharing is not supported in RBLN.") - if blocksparse_params is not None: - raise ValueError("RBLN Attention Backend does not " - "support block-sparse attention.") if logits_soft_cap is not None: logger.warning_once( "RBLN Attention Backend does not support logits soft cap. "