cockroach54/opensolar-qlora-dpo

just for practice

• trained by transformers, peft, bitsandbytes
• evaluated by https://github.com/davidkim205/kollm_evaluation

base model

• beomi/OPEN-SOLAR-KO-10.7B

Tasks	Version	Filter	n-shot	Metric	Value		Stderr
kobest_wic	1	none	0	acc	0.4873	±	0.0141
		none	0	f1	0.3276	±	N/A
kobest_sentineg	1	none	0	acc	0.6499	±	0.0240
		none	0	f1	0.6134	±	N/A
kobest_hellaswag	1	none	0	acc	0.4980	±	0.0224
		none	0	f1	0.4952	±	N/A
		none	0	acc_norm	0.6280	±	0.0216
kobest_copa	1	none	0	acc	0.8030	±	0.0126
		none	0	f1	0.8026	±	N/A
kobest_boolq	1	none	0	acc	0.8711	±	0.0089
		none	0	f1	0.8710	±	N/A
ko_truthfulqa	2	none	0	acc	0.2717	±	0.0156
ko_hellaswag	1	none	0	acc	0.4109	±	0.0049
		none	0	acc_norm	0.5503	±	0.0050
ko_common_gen	1	none	0	acc	0.9041	±	0.0075
		none	0	acc_norm	0.9041	±	0.0075
ko_arc_easy	1	none	0	acc	0.3490	±	0.0139
		none	0	acc_norm	0.4326	±	0.0145