cgato/Nemo12b-TheSyntheticOne

Trained using https://huggingface.co/datasets/cgato/TheSmarts for demonstration purposes. Probably a fairly competent assistant model. May do KTO overtop to sand down the edges and improve performance later.

Roles: system, user, assistant

Training Loss	Epoch	Step	Validation Loss
0.9338	0.0003	1	0.9347
0.8271	0.0328	100	0.7933
0.9541	0.0656	200	0.8407
0.7497	0.0984	300	0.7934
0.8786	0.1311	400	0.7724
0.8257	0.1639	500	0.7627
0.8258	0.1967	600	0.7679
0.7207	0.2295	700	0.7497
0.9439	0.2623	800	0.7576
0.852	0.2951	900	0.7361
0.7852	0.3279	1000	0.7375
0.7	0.3607	1100	0.7298
0.7865	0.3934	1200	0.7202
0.6182	0.4262	1300	0.7146
0.6885	0.4590	1400	0.7131
0.7154	0.4918	1500	0.7083
0.7187	0.5246	1600	0.7016
0.6877	0.5574	1700	0.6976
0.7908	0.5902	1800	0.6946
0.7664	0.6230	1900	0.6894
0.7214	0.6557	2000	0.6857
0.6971	0.6885	2100	0.6837
0.6527	0.7213	2200	0.6804
0.6815	0.7541	2300	0.6781
0.6359	0.7869	2400	0.6759
0.6874	0.8197	2500	0.6742
0.5999	0.8525	2600	0.6728
0.7391	0.8852	2700	0.6719
0.6509	0.9180	2800	0.6710
0.6346	0.9508	2900	0.6702
0.7023	0.9836	3000	0.6696