Share your thoughts, 1 month free Claude Pro on usSee more

Preservation of General Capabilities on HellaSwag, WinoGrande, IFEval, MMLU

20.4HellaSwag Delta

WiSE-FT

Updated 14d ago

Evaluation Results

Method	Links
WiSE-FT 2026.05		20.4	3.4	50	7	20.2
WiSE-FT 2026.05		14.7	3.2	37.8	7.2	15.7
L2 Reg 2026.05		10.8	0.8	11.8	3.9	6.4
SFT 2026.05		10.4	0.3	21.7	3.6	9
FLOW 2026.05		8.4	2.5	36.9	6	12.2
DFT 2026.05		7.8	2.4	9	2.2	4.2
SFT 2026.05		7.3	2	9.4	19.6	9.6
FLOW 2026.05		6.4	0.7	13.7	12.7	8
L2 Reg 2026.05		5.5	0.2	1.9	14.3	5.4
TALR 2026.05		5.5	0.6	12.3	2	4.8
LoRA 2026.05		4.4	6.9	9.1	9.2	7.4
FINCH 2026.05		4.2	2.2	0	2.6	1.2
SFT (small lr) 2026.05		3.8	2.2	3.1	3.2	1.5
DFT 2026.05		2.7	2.7	3	1.9	1.1
DFT 2026.05		2.6	-0.6	5.2	2.3	2.4
TALR 2026.05		1.3	1	4.5	5.2	2.5
STM 2026.05		1.2	1.1	1.4	0.9	0.6
LoRA 2026.05		0.9	0.6	15.7	3.8	5.3
SFT (small lr) 2026.05		0.7	0.5	6.2	0.9	1.4
FINCH 2026.05		0.6	1.4	0.5	0.8	0.1
STM 2026.05		0.4	0	12.3	0.7	3.4
TALR 2026.05		0.3	0.4	-3.5	-0.8	-0.9
SFT (small lr) 2026.05		0.1	-2.4	-0.1	-0.4	-0.7
Base 2026.05		0	0	0	0	0
Base 2026.05		0	0	0	0	0
Base 2026.05		0	0	0	0	0
Base 2026.05		0	0	0	0	0
STM 2026.05		-0.8	0.2	-4.1	0.8	-1
FINCH 2026.05		-1.2	-0.6	0	0	-0.5
FINCH 2026.05		-1.6	-2.7	0	-5.1	-2.4
SFT (small lr) 2026.05		-3.6	0.2	-4.1	0.4	-1.8
STM 2026.05		-4.2	-1.4	-8.3	-2	-4
LoRA 2026.05		-4.6	-2.2	-16.7	-1	-6.1
DFT 2026.05		-5.1	-3.4	-20.8	-5	-8.6
TALR 2026.05		-6.4	-0.4	-8.3	0	-3.8
SFT 2026.05		-10	-11.6	-25.1	-10.4	-14.3
LoRA 2026.05		-10.8	5.1	-45.6	-2.1	-13.4
SFT 2026.05		-12.2	-0.8	-16.7	-4.2	-8.5
L2 Reg 2026.05		-12.2	-1.8	-16.7	-4	-8.7
L2 Reg 2026.05		-12.4	-11.7	-39.2	-11.2	-18.6
WiSE-FT 2026.05		-12.6	-2	-16.7	-5	-9.1
FLOW 2026.05		-15.5	-8.7	-15.6	-9.4	-12.3
WiSE-FT 2026.05		-16.3	-9.6	-26.6	-7.8	-15.1
FLOW 2026.05		-31.8	-5.2	-20.8	-13.2	-17.8