A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 00:30 UTC
[Bugfix] Allow 64-bit integer values for LoRA IDs to avoid overflow/truncation (#27876)
Signed-off-by: Madeesh Kannan <shadeMe@users.noreply.github.com>
Madeesh Kannan committed
Oct 31, 2025 at 17:58 UTC
675704ac01e8ab1b986f777f7ccc4ac72159eb7b
2 files changed
+2
-2
vllm/v1/worker/gpu_input_batch.py
+1
-1
@@ -204,7 +204,7 @@ class InputBatch:
204
self.num_accepted_tokens_cpu = self.num_accepted_tokens_cpu_tensor.numpy()
205
206
# lora related
207
- self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int32)
207
+ self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int64)
208
self.lora_id_to_request_ids: dict[int, set[str]] = {}
209
self.lora_id_to_lora_request: dict[int, LoRARequest] = {}
210
vllm/v1/worker/tpu_input_batch.py
+1
-1
@@ -139,7 +139,7 @@ class InputBatch:
139
self.min_tokens: dict[int, tuple[int, set[int]]] = {}
140
141
# lora related
142
- self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int32)
142
+ self.request_lora_mapping = np.zeros((self.max_num_reqs,), dtype=np.int64)
143
self.lora_id_to_request_ids: dict[int, set[str]] = {}
144
self.lora_id_to_lora_request: dict[int, LoRARequest] = {}
145