fix: bump up bitsandbytes version to support torch==2.6.0 upgrade

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

Author: kmehant

plugins/accelerated-peft/requirements.txt

@@ -5,10 +5,9 @@
 accelerate >= 0.29
 
 # bitsandbytes for the BNB plugin
-# - lower bound is because bnb is missing quant_state
-# - upper bound is because of segmentation faults
-# see https://github.com/foundation-model-stack/fms-acceleration/issues/17
-bitsandbytes >=0.41,<=0.43.3
+# exact version is needed 0.45.1 for torch upgrade to 2.6
+
+bitsandbytes == 0.45.1
 
 # Used to manage the thread limit in functions for converting old 
 # GPTQ models to new GPTQ model format that support symmetrical=False