This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-news-IFT on the hZzy/train_pairwise dataset. It achieves the following results on the evaluation set:

Loss: 0.3417
Logps: -93.8085
Logits: -1.1885
Objective: 0.3403
Dpo Loss: 0.7083
Regularize: 0.2695
Ranking Simple: 0.5186
Ranking Idealized: 0.5399
Ranking Idealized Expo: 0.5243

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-07
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 3
gradient_accumulation_steps: 8
total_train_batch_size: 96
total_eval_batch_size: 12
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 10
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
0.0798	0.0945	50	0.0807	-98.5040	-1.3072	0.0808	0.6932	0.0115	0.5238	0.5399	0.5243
0.081	0.1889	100	0.0819	-98.4932	-1.3084	0.0819	0.6934	0.0125	0.5238	0.5399	0.5243
0.0839	0.2834	150	0.0823	-98.5417	-1.3079	0.0823	0.6931	0.0130	0.5233	0.5399	0.5243
0.0891	0.3779	200	0.0840	-98.6517	-1.3063	0.0839	0.6927	0.0147	0.5238	0.5399	0.5243
0.1019	0.4724	250	0.0865	-98.6753	-1.3058	0.0864	0.6929	0.0171	0.5233	0.5399	0.5243
0.1094	0.5668	300	0.0928	-98.3448	-1.3087	0.0930	0.6926	0.0238	0.5238	0.5399	0.5243
0.1267	0.6613	350	0.0995	-98.4803	-1.3097	0.1004	0.6942	0.0310	0.5243	0.5399	0.5243
0.1414	0.7558	400	0.1020	-98.6999	-1.3138	0.1027	0.6920	0.0335	0.5248	0.5399	0.5243
0.156	0.8503	450	0.1102	-98.6961	-1.3001	0.1107	0.6917	0.0415	0.5228	0.5399	0.5243
0.1843	0.9447	500	0.1425	-98.3685	-1.2985	0.1416	0.6934	0.0723	0.5217	0.5399	0.5243
0.1954	1.0392	550	0.1388	-98.2336	-1.3154	0.1383	0.6946	0.0689	0.5228	0.5399	0.5243
0.2073	1.1337	600	0.1374	-98.7215	-1.3071	0.1369	0.6936	0.0676	0.5228	0.5399	0.5243
0.2165	1.2282	650	0.1478	-97.9261	-1.2926	0.1487	0.6916	0.0796	0.5233	0.5399	0.5243
0.2333	1.3226	700	0.1470	-97.1071	-1.2930	0.1450	0.6915	0.0758	0.5243	0.5399	0.5243
0.229	1.4171	750	0.1718	-97.0923	-1.2689	0.1725	0.6929	0.1032	0.5238	0.5399	0.5243
0.2565	1.5116	800	0.1817	-97.2621	-1.2540	0.1830	0.6944	0.1136	0.5243	0.5399	0.5243
0.2479	1.6060	850	0.1864	-96.3423	-1.2708	0.1853	0.6946	0.1159	0.5243	0.5399	0.5243
0.2586	1.7005	900	0.1839	-97.2157	-1.2623	0.1825	0.6944	0.1131	0.5223	0.5399	0.5243
0.2347	1.7950	950	0.1995	-94.8402	-1.2678	0.1989	0.6945	0.1295	0.5238	0.5399	0.5243
0.2414	1.8895	1000	0.1895	-95.8793	-1.2579	0.1901	0.6924	0.1209	0.5254	0.5399	0.5243
0.2433	1.9839	1050	0.2097	-95.7970	-1.2552	0.2068	0.6923	0.1376	0.5259	0.5399	0.5243
0.2393	2.0784	1100	0.2156	-96.9313	-1.2422	0.2149	0.6962	0.1452	0.5264	0.5399	0.5243
0.2476	2.1729	1150	0.2195	-95.8618	-1.2485	0.2191	0.6958	0.1495	0.5238	0.5399	0.5243
0.2443	2.2674	1200	0.2318	-97.1362	-1.2241	0.2317	0.6998	0.1617	0.5259	0.5399	0.5243
0.2337	2.3618	1250	0.2494	-96.2629	-1.2313	0.2515	0.6950	0.1820	0.5269	0.5399	0.5243
0.2264	2.4563	1300	0.2473	-94.4504	-1.2535	0.2456	0.6981	0.1758	0.5223	0.5399	0.5243
0.2398	2.5508	1350	0.2467	-96.2065	-1.2349	0.2462	0.7027	0.1760	0.5197	0.5399	0.5243
0.2346	2.6453	1400	0.2565	-94.6591	-1.2562	0.2567	0.7002	0.1867	0.5212	0.5399	0.5243
0.242	2.7397	1450	0.2640	-94.6555	-1.2141	0.2641	0.7015	0.1939	0.5243	0.5399	0.5243
0.2372	2.8342	1500	0.2747	-94.9289	-1.2472	0.2726	0.7027	0.2024	0.5202	0.5399	0.5243
0.2133	2.9287	1550	0.2529	-95.1991	-1.2345	0.2512	0.7006	0.1811	0.5243	0.5399	0.5243
0.2292	3.0231	1600	0.2840	-93.6334	-1.2437	0.2861	0.7038	0.2157	0.5197	0.5399	0.5243
0.2227	3.1176	1650	0.2854	-93.4763	-1.2332	0.2851	0.7025	0.2149	0.5217	0.5399	0.5243
0.2123	3.2121	1700	0.2752	-95.6906	-1.2311	0.2756	0.7008	0.2055	0.5233	0.5399	0.5243
0.218	3.3066	1750	0.2800	-95.9042	-1.2167	0.2783	0.7037	0.2079	0.5238	0.5399	0.5243
0.2086	3.4010	1800	0.2945	-95.6983	-1.2183	0.2932	0.7027	0.2230	0.5233	0.5399	0.5243
0.216	3.4955	1850	0.2895	-93.0784	-1.2235	0.2873	0.7028	0.2171	0.5212	0.5399	0.5243
0.2182	3.5900	1900	0.2973	-95.2384	-1.2138	0.2977	0.7019	0.2275	0.5207	0.5399	0.5243
0.2097	3.6845	1950	0.3023	-93.4940	-1.2111	0.3000	0.7046	0.2295	0.5217	0.5399	0.5243
0.2076	3.7789	2000	0.3084	-93.0939	-1.2337	0.3067	0.7034	0.2364	0.5243	0.5399	0.5243
0.2099	3.8734	2050	0.2962	-93.1727	-1.2280	0.2954	0.7044	0.2249	0.5212	0.5399	0.5243
0.2001	3.9679	2100	0.3139	-93.9210	-1.2079	0.3123	0.7063	0.2417	0.5186	0.5399	0.5243
0.2082	4.0624	2150	0.3119	-93.6768	-1.2148	0.3124	0.7037	0.2420	0.5217	0.5399	0.5243
0.1914	4.1568	2200	0.3139	-94.5737	-1.2179	0.3138	0.7032	0.2434	0.5197	0.5399	0.5243
0.2026	4.2513	2250	0.3179	-93.2220	-1.2044	0.3177	0.7035	0.2473	0.5202	0.5399	0.5243
0.1908	4.3458	2300	0.3067	-94.3151	-1.2117	0.3085	0.7022	0.2383	0.5233	0.5399	0.5243
0.1931	4.4402	2350	0.3241	-93.4124	-1.2066	0.3236	0.7058	0.2530	0.5223	0.5399	0.5243
0.195	4.5347	2400	0.3111	-94.2419	-1.2062	0.3113	0.7035	0.2410	0.5217	0.5399	0.5243
0.1947	4.6292	2450	0.3312	-93.6715	-1.1956	0.3317	0.7067	0.2610	0.5228	0.5399	0.5243
0.1837	4.7237	2500	0.3289	-93.6179	-1.2041	0.3304	0.7077	0.2596	0.5223	0.5399	0.5243
0.1751	4.8181	2550	0.3254	-93.4709	-1.1993	0.3247	0.7060	0.2541	0.5212	0.5399	0.5243
0.1717	4.9126	2600	0.3287	-94.2886	-1.2078	0.3292	0.7050	0.2587	0.5207	0.5399	0.5243
0.1761	5.0071	2650	0.3257	-93.6210	-1.2055	0.3239	0.7061	0.2533	0.5217	0.5399	0.5243
0.1692	5.1016	2700	0.3396	-93.0109	-1.2063	0.3378	0.7072	0.2670	0.5223	0.5399	0.5243
0.1676	5.1960	2750	0.3402	-93.9591	-1.1978	0.3384	0.7084	0.2675	0.5202	0.5399	0.5243
0.1743	5.2905	2800	0.3371	-93.9100	-1.1972	0.3351	0.7076	0.2643	0.5217	0.5399	0.5243
0.1715	5.3850	2850	0.3408	-93.6808	-1.1939	0.3405	0.7084	0.2696	0.5212	0.5399	0.5243
0.1643	5.4795	2900	0.3434	-93.0381	-1.1941	0.3434	0.7095	0.2724	0.5192	0.5399	0.5243
0.1569	5.5739	2950	0.3403	-94.4489	-1.1993	0.3406	0.7083	0.2698	0.5192	0.5399	0.5243
0.16	5.6684	3000	0.3337	-94.1339	-1.1952	0.3332	0.7068	0.2625	0.5233	0.5399	0.5243
0.1556	5.7629	3050	0.3379	-93.7011	-1.1943	0.3366	0.7075	0.2658	0.5197	0.5399	0.5243
0.1544	5.8573	3100	0.3407	-93.8059	-1.1896	0.3385	0.7082	0.2677	0.5212	0.5399	0.5243
0.1539	5.9518	3150	0.3377	-93.3647	-1.2013	0.3358	0.7079	0.2650	0.5207	0.5399	0.5243
0.1448	6.0463	3200	0.3418	-93.0674	-1.1912	0.3402	0.7086	0.2693	0.5181	0.5399	0.5243
0.1479	6.1408	3250	0.3437	-93.1651	-1.1883	0.3423	0.7079	0.2715	0.5217	0.5399	0.5243
0.1408	6.2352	3300	0.3427	-93.4029	-1.1821	0.3405	0.7074	0.2698	0.5197	0.5399	0.5243
0.1475	6.3297	3350	0.3401	-93.6032	-1.1856	0.3383	0.7078	0.2675	0.5192	0.5399	0.5243
0.1339	6.4242	3400	0.3415	-93.5229	-1.1891	0.3402	0.7082	0.2693	0.5212	0.5399	0.5243
0.1394	6.5187	3450	0.3398	-94.0518	-1.1959	0.3379	0.7083	0.2671	0.5186	0.5399	0.5243
0.1324	6.6131	3500	0.3401	-93.9466	-1.1836	0.3389	0.7075	0.2682	0.5192	0.5399	0.5243
0.1385	6.7076	3550	0.3449	-93.6245	-1.1866	0.3437	0.7080	0.2729	0.5202	0.5399	0.5243
0.1289	6.8021	3600	0.3433	-93.8482	-1.1858	0.3412	0.7088	0.2703	0.5192	0.5399	0.5243
0.1272	6.8966	3650	0.3431	-93.9371	-1.1979	0.3417	0.7080	0.2709	0.5202	0.5399	0.5243
0.125	6.9910	3700	0.3436	-93.9666	-1.1952	0.3425	0.7079	0.2717	0.5202	0.5399	0.5243
0.1227	7.0855	3750	0.3404	-93.8781	-1.2022	0.3382	0.7086	0.2674	0.5197	0.5399	0.5243
0.1142	7.1800	3800	0.3426	-93.8234	-1.1874	0.3420	0.7083	0.2712	0.5207	0.5399	0.5243
0.1142	7.2744	3850	0.3454	-93.6895	-1.1775	0.3442	0.7090	0.2733	0.5202	0.5399	0.5243
0.1128	7.3689	3900	0.3417	-94.0521	-1.1838	0.3406	0.7083	0.2698	0.5197	0.5399	0.5243
0.1158	7.4634	3950	0.3434	-93.9208	-1.1875	0.3423	0.7086	0.2714	0.5197	0.5399	0.5243
0.113	7.5579	4000	0.3428	-93.6866	-1.1850	0.3411	0.7087	0.2702	0.5197	0.5399	0.5243
0.1113	7.6523	4050	0.3434	-93.6171	-1.1837	0.3425	0.7087	0.2716	0.5202	0.5399	0.5243
0.1082	7.7468	4100	0.3411	-94.0013	-1.1852	0.3403	0.7081	0.2695	0.5192	0.5399	0.5243
0.1051	7.8413	4150	0.3425	-93.8552	-1.1848	0.3417	0.7083	0.2709	0.5197	0.5399	0.5243
0.1047	7.9358	4200	0.3422	-93.6696	-1.1872	0.3411	0.7085	0.2702	0.5197	0.5399	0.5243
0.0985	8.0302	4250	0.3416	-93.6924	-1.1844	0.3403	0.7083	0.2695	0.5197	0.5399	0.5243
0.0964	8.1247	4300	0.3422	-93.5025	-1.1871	0.3409	0.7082	0.2701	0.5202	0.5399	0.5243
0.0997	8.2192	4350	0.3423	-93.8074	-1.1866	0.3408	0.7081	0.2700	0.5186	0.5399	0.5243
0.0963	8.3137	4400	0.3434	-93.6885	-1.1861	0.3419	0.7084	0.2711	0.5202	0.5399	0.5243
0.0966	8.4081	4450	0.3434	-93.7312	-1.1875	0.3419	0.7084	0.2711	0.5186	0.5399	0.5243
0.0956	8.5026	4500	0.3431	-93.8431	-1.1866	0.3416	0.7081	0.2708	0.5186	0.5399	0.5243
0.0928	8.5971	4550	0.3428	-93.8243	-1.1859	0.3414	0.7084	0.2706	0.5186	0.5399	0.5243
0.0924	8.6915	4600	0.3418	-93.7706	-1.1871	0.3406	0.7082	0.2698	0.5186	0.5399	0.5243
0.0908	8.7860	4650	0.3415	-93.7405	-1.1872	0.3403	0.7079	0.2695	0.5202	0.5399	0.5243
0.0922	8.8805	4700	0.3419	-93.7126	-1.1888	0.3405	0.7078	0.2698	0.5202	0.5399	0.5243
0.0895	8.9750	4750	0.3417	-93.7926	-1.1886	0.3402	0.7080	0.2694	0.5202	0.5399	0.5243
0.0877	9.0694	4800	0.3425	-93.7523	-1.1891	0.3415	0.7083	0.2706	0.5197	0.5399	0.5243
0.0862	9.1639	4850	0.3423	-93.8492	-1.1894	0.3406	0.7082	0.2698	0.5207	0.5399	0.5243
0.0856	9.2584	4900	0.3417	-93.8453	-1.1883	0.3404	0.7081	0.2696	0.5197	0.5399	0.5243
0.0883	9.3529	4950	0.3414	-93.8773	-1.1886	0.3401	0.7080	0.2693	0.5202	0.5399	0.5243
0.0866	9.4473	5000	0.3414	-93.8593	-1.1880	0.3402	0.7081	0.2694	0.5197	0.5399	0.5243
0.0843	9.5418	5050	0.3417	-93.8241	-1.1880	0.3405	0.7081	0.2697	0.5207	0.5399	0.5243
0.0862	9.6363	5100	0.3419	-93.8268	-1.1884	0.3404	0.7081	0.2696	0.5197	0.5399	0.5243
0.0851	9.7308	5150	0.3418	-93.8247	-1.1881	0.3405	0.7082	0.2697	0.5192	0.5399	0.5243
0.0852	9.8252	5200	0.3415	-93.8257	-1.1886	0.3402	0.7081	0.2694	0.5197	0.5399	0.5243
0.0873	9.9197	5250	0.3418	-93.8220	-1.1885	0.3404	0.7083	0.2696	0.5197	0.5399	0.5243

Framework versions

Transformers 4.42.0
Pytorch 2.3.0+cu121
Datasets 2.19.1
Tokenizers 0.19.1

hZzy/qwen2.5-0.5b-expo-EXDPO-WEIGHT-BETA0.2

qwen2.5-0.5b-expo-EXDPO-WEIGHT-BETA0.2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

hZzy/
qwen2.5-0.5b-expo-EXDPO-WEIGHT-BETA0.2