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 15, 18:30 UTC
[Misc][Docs] fix the comments of KV_T and CACHE_T in CALL_RESHAPE_AND_CACHE_XX macros (#14347)
Serena committed
Mar 18, 2025 at 20:50 UTC
64fc2193dc777f764b5e6b3dde7bdef340ade1c5
1 file changed
+6
-6
csrc/cache_kernels.cu
+6
-6
@@ -350,8 +350,8 @@ __global__ void concat_and_cache_mla_kernel(
350
351
} // namespace vllm
352
353
-// KV_T is the stored data type of kv-cache.
354
-// CACHE_T is the data type of key and value tensors.
353
+// KV_T is the data type of key and value tensors.
354
+// CACHE_T is the stored data type of kv-cache.
355
// KV_DTYPE is the real data type of kv-cache.
356
#define CALL_RESHAPE_AND_CACHE(KV_T, CACHE_T, KV_DTYPE) \
357
vllm::reshape_and_cache_kernel<KV_T, CACHE_T, KV_DTYPE> \
@@ -393,8 +393,8 @@ void reshape_and_cache(
393
CALL_RESHAPE_AND_CACHE)
394
}
395
396
-// KV_T is the stored data type of kv-cache.
397
-// CACHE_T is the data type of key and value tensors.
396
+// KV_T is the data type of key and value tensors.
397
+// CACHE_T is the stored data type of kv-cache.
398
// KV_DTYPE is the real data type of kv-cache.
399
#define CALL_RESHAPE_AND_CACHE_FLASH(KV_T, CACHE_T, KV_DTYPE) \
400
vllm::reshape_and_cache_flash_kernel<KV_T, CACHE_T, KV_DTYPE> \
@@ -446,8 +446,8 @@ void reshape_and_cache_flash(
446
CALL_RESHAPE_AND_CACHE_FLASH);
447
}
448
449
-// KV_T is the stored data type of kv-cache.
450
-// CACHE_T is the data type of key and value tensors.
449
+// KV_T is the data type of key and value tensors.
450
+// CACHE_T is the stored data type of kv-cache.
451
// KV_DTYPE is the real data type of kv-cache.
452
#define CALL_CONCAT_AND_CACHE_MLA(KV_T, CACHE_T, KV_DTYPE) \
453
vllm::concat_and_cache_mla_kernel<KV_T, CACHE_T, KV_DTYPE> \