Tune Qwen medium prefill batching
This commit is contained in:
@@ -23,6 +23,8 @@ VISION_PROJECTOR_FILE=qwen/mmproj-BF16.gguf
|
||||
|
||||
FAST_CONTEXT=76800
|
||||
MEDIUM_CONTEXT=160000
|
||||
MEDIUM_BATCH_SIZE=2048
|
||||
MEDIUM_UBATCH_SIZE=128
|
||||
LARGE_CONTEXT=192000
|
||||
ULTRA_CONTEXT=262144
|
||||
UNCENSORED_CONTEXT=80000
|
||||
|
||||
Reference in New Issue
Block a user