mistral-7b-expo-7b-L2EXPO-25-final-2

This model is a fine-tuned version of hZzy/mistral-7b-sft-25-1 on the hZzy/direction_right2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4637
• Objective: 0.4623
• Reward Accuracy: 0.6510
• Logp Accuracy: 0.5705
• Log Diff Policy: 3.5252
• Chosen Logps: -91.4483
• Rejected Logps: -94.9735
• Chosen Rewards: 0.1617
• Rejected Rewards: 0.0045
• Logits: -2.0668

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-06
• train_batch_size: 3
• eval_batch_size: 3
• seed: 42
• distributed_type: multi-GPU
• num_devices: 3
• gradient_accumulation_steps: 12
• total_train_batch_size: 108
• total_eval_batch_size: 9
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.2
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Objective	Reward Accuracy	Logp Accuracy	Log Diff Policy	Chosen Logps	Rejected Logps	Chosen Rewards	Rejected Rewards	Logits
0.6143	0.1213	80	0.5109	0.5074	0.5414	0.5154	0.4308	-93.9032	-94.3340	0.0390	0.0365	-2.2003
0.5624	0.2427	160	0.5040	0.5009	0.5713	0.5210	0.7341	-92.2105	-92.9446	0.1236	0.1060	-2.1577
0.5296	0.3640	240	0.4863	0.4843	0.6079	0.5316	1.4806	-93.1361	-94.6168	0.0773	0.0224	-2.1544
0.5062	0.4853	320	0.4769	0.4758	0.6295	0.5467	2.3729	-88.3420	-90.7150	0.3170	0.2175	-2.2017
0.489	0.6067	400	0.4707	0.4703	0.6418	0.5576	2.7726	-89.4151	-92.1878	0.2634	0.1438	-2.2280
0.4965	0.7280	480	0.4679	0.4692	0.6331	0.5610	2.8915	-89.0811	-91.9726	0.2801	0.1546	-2.2266
0.4905	0.8493	560	0.4693	0.4712	0.6390	0.5607	3.1056	-91.1217	-94.2273	0.1780	0.0419	-2.2026
0.4547	0.9707	640	0.4653	0.4671	0.6353	0.5624	3.1047	-92.8997	-96.0044	0.0891	-0.0470	-2.1916
0.4497	1.0920	720	0.4672	0.4683	0.6404	0.5660	3.3740	-84.0528	-87.4268	0.5315	0.3819	-2.1664
0.4358	1.2133	800	0.4617	0.4629	0.6398	0.5629	3.2392	-85.6140	-88.8532	0.4534	0.3106	-2.1284
0.4572	1.3347	880	0.4665	0.4689	0.6398	0.5682	3.3859	-89.4141	-92.8000	0.2634	0.1132	-2.1583
0.4362	1.4560	960	0.4647	0.4657	0.6395	0.5702	3.5918	-86.3147	-89.9066	0.4184	0.2579	-2.1314
0.3976	1.5774	1040	0.4635	0.4660	0.6365	0.5685	3.5224	-88.0452	-91.5676	0.3319	0.1748	-2.1032
0.4082	1.6987	1120	0.4628	0.4651	0.6429	0.5707	3.4228	-85.0947	-88.5175	0.4794	0.3273	-2.0848
0.4037	1.8200	1200	0.4621	0.4621	0.6449	0.5775	3.6505	-93.7257	-97.3762	0.0478	-0.1156	-2.0348
0.3961	1.9414	1280	0.4607	0.4613	0.6418	0.5744	3.4066	-90.0586	-93.4653	0.2312	0.0800	-2.0810

Framework versions

• PEFT 0.11.1
• Transformers 4.42.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.19.1