Tune Qwen medium prefill batching
This commit is contained in:
@@ -77,6 +77,8 @@ VISION_PROJECTOR_SHA256=83ee4f4f205fa514161778c41df1ea14144faa0f713510893b63c239
|
||||
|
||||
FAST_CONTEXT=76800
|
||||
MEDIUM_CONTEXT=160000
|
||||
MEDIUM_BATCH_SIZE=2048
|
||||
MEDIUM_UBATCH_SIZE=128
|
||||
MEDIUM_TENSOR_SPLIT=90,10
|
||||
LARGE_CONTEXT=192000
|
||||
LARGE_TENSOR_SPLIT=86,14
|
||||
|
||||
Reference in New Issue
Block a user