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

Loss: 213.8410
Logps: -86.8482
Logits: -1.2868
Objective: 206.3083
Dpo Loss: 0.6894
Regularize: 0.4694
Ranking Simple: 0.5280
Wo Beta: 15.8446

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: 1e-06
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 3
gradient_accumulation_steps: 12
total_train_batch_size: 144
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: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Wo Beta
175.6822	0.1417	50	183.1265	-91.3250	-1.4199	181.6852	0.6866	0.4119	0.5274	16.2457
158.0048	0.2834	100	187.5626	-92.8634	-1.4295	185.1372	0.6803	0.4198	0.5295	16.2784
170.2156	0.4251	150	198.8161	-90.2801	-1.3391	195.4767	0.6839	0.4448	0.5336	16.0768
177.2075	0.5668	200	198.3241	-88.0323	-1.3280	192.5403	0.6833	0.4402	0.5321	15.9895
166.0305	0.7085	250	202.1109	-87.1848	-1.3573	195.7011	0.6871	0.4491	0.5311	16.0690
159.4572	0.8503	300	208.5400	-89.4338	-1.2967	204.3580	0.6901	0.4665	0.5305	16.1355
139.7142	0.9920	350	208.0794	-88.7720	-1.2946	202.5639	0.6873	0.4618	0.5290	16.0553
132.3562	1.1337	400	209.3769	-86.7283	-1.2669	202.8542	0.6858	0.4624	0.5295	15.8495
113.8641	1.2754	450	213.1838	-88.9803	-1.2472	206.6228	0.6906	0.4722	0.5280	15.9318
116.7569	1.4171	500	215.5911	-88.2623	-1.2901	207.2833	0.6870	0.4720	0.5290	15.7777
101.1159	1.5588	550	215.9760	-87.7554	-1.3051	207.8397	0.6893	0.4738	0.5290	15.8730
92.5438	1.7005	600	217.6953	-86.1461	-1.2952	210.4230	0.6916	0.4795	0.5295	15.8750
93.1056	1.8422	650	217.1667	-86.9562	-1.2989	208.9618	0.6900	0.4758	0.5290	15.8184
81.9079	1.9839	700	214.1251	-86.8994	-1.2788	205.7279	0.6888	0.4685	0.5264	15.8115
63.0917	2.1256	750	215.1674	-86.9307	-1.2831	208.1704	0.6905	0.4733	0.5274	15.8272
54.8988	2.2674	800	213.3371	-86.7932	-1.2914	205.5635	0.6889	0.4683	0.5269	15.8558
47.8826	2.4091	850	214.3945	-86.7633	-1.2883	206.9195	0.6901	0.4707	0.5280	15.8444
46.1635	2.5508	900	214.1872	-86.7292	-1.2930	206.3981	0.6892	0.4695	0.5274	15.8404
43.9185	2.6925	950	214.2727	-86.8524	-1.2873	206.7197	0.6895	0.4704	0.5285	15.8399
41.5921	2.8342	1000	213.8429	-86.8651	-1.2871	206.3308	0.6894	0.4694	0.5280	15.8452
42.8137	2.9759	1050	213.8409	-86.8482	-1.2868	206.3081	0.6894	0.4694	0.5280	15.8446

Framework versions

Transformers 4.42.0
Pytorch 2.3.0+cu121
Datasets 3.2.0
Tokenizers 0.19.1

hZzy/qwen2.5-0.5b-expo-L2EXPO-W0-noES6-0.2

qwen2.5-0.5b-expo-L2EXPO-W0-noES6-0.2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

hZzy/
qwen2.5-0.5b-expo-L2EXPO-W0-noES6-0.2