From 675704ac01e8ab1b986f777f7ccc4ac72159eb7b Mon Sep 17 00:00:00 2001 From: Madeesh Kannan Date: Fri, 31 Oct 2025 17:58:42 +0100 Subject: [PATCH] [Bugfix] Allow 64-bit integer values for LoRA IDs to avoid overflow/truncation (#27876) Signed-off-by: Madeesh Kannan --- vllm/v1/worker/gpu_input_batch.py | 2 +- vllm/v1/worker/tpu_input_batch.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/v1/worker/gpu_input_batch.py b/vllm/v1/worker/gpu_input_batch.py index bc7578cbd97cd..fe834db115e70 100644 --- a/vllm/v1/worker/gpu_input_batch.py +++ b/vllm/v1/worker/gpu_input_batch.py @@ -204,7 +204,7 @@ class InputBatch: self.num_accepted_tokens_cpu = self.num_accepted_tokens_cpu_tensor.numpy() # lora related - self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int32) + self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int64) self.lora_id_to_request_ids: dict[int, set[str]] = {} self.lora_id_to_lora_request: dict[int, LoRARequest] = {} diff --git a/vllm/v1/worker/tpu_input_batch.py b/vllm/v1/worker/tpu_input_batch.py index 74e8225b2f4b8..d3fb17054c1a7 100644 --- a/vllm/v1/worker/tpu_input_batch.py +++ b/vllm/v1/worker/tpu_input_batch.py @@ -139,7 +139,7 @@ class InputBatch: self.min_tokens: dict[int, tuple[int, set[int]]] = {} # lora related - self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int32) + self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int64) self.lora_id_to_request_ids: dict[int, set[str]] = {} self.lora_id_to_lora_request: dict[int, LoRARequest] = {}