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, 12:15 UTC
[Bugfix] LoRA - Retire unused maxnreg LoRA kernel argument (#17677)
Varun Sundar Rabindranath committed
May 6, 2025 at 06:04 UTC
90bd2ae172007b7e84e15eea45b37826d5da1a67
2 files changed
-4
vllm/lora/ops/triton_ops/lora_expand_op.py
-2
@@ -204,7 +204,6 @@ def _lora_expand(
204
NUM_WARPS = 4
205
NUM_CTAS = 1
206
NUM_STAGES = 2
207
- MAX_NREG = None
207
208
EVEN_K = K % BLOCK_K == 0 # type: ignore
209
@@ -258,7 +257,6 @@ def _lora_expand(
257
num_warps=NUM_WARPS,
258
num_ctas=NUM_CTAS,
259
num_stages=NUM_STAGES,
261
- maxnreg=MAX_NREG,
260
)
261
262
return
vllm/lora/ops/triton_ops/lora_shrink_op.py
-2
@@ -168,7 +168,6 @@ def _lora_shrink(
168
NUM_WARPS = 4
169
NUM_CTAS = 1
170
NUM_STAGES = 2
171
- MAX_NREG = None
171
172
EVEN_K = K % (BLOCK_K * SPLIT_K) == 0 # type: ignore
173
@@ -213,7 +212,6 @@ def _lora_shrink(
212
num_warps=NUM_WARPS,
213
num_ctas=NUM_CTAS,
214
num_stages=NUM_STAGES,
216
- maxnreg=MAX_NREG,
215
)
216
217
return