mistral-7b-expo-7b-L2EXPO-25-last-1

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: 0.4397
• Objective: 0.4403
• Reward Accuracy: 0.6644
• Logp Accuracy: 0.6488
• Log Diff Policy: 15.5945
• Chosen Logps: -164.8095
• Rejected Logps: -180.4040
• Chosen Rewards: -0.7013
• Rejected Rewards: -0.8534
• Logits: -2.0380

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: 0.1
• num_epochs: 1

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
0.5865	0.0758	50	0.5115	0.5085	0.5296	0.5176	0.4580	-92.9565	-93.4145	0.0173	0.0165	-2.2007
0.6073	0.1517	100	0.5074	0.5043	0.5671	0.5271	0.9476	-93.0270	-93.9747	0.0166	0.0109	-2.2026
0.627	0.2275	150	0.4960	0.4934	0.5808	0.5506	2.8875	-94.9425	-97.8300	-0.0026	-0.0277	-2.0708
0.5403	0.3033	200	0.4882	0.4891	0.5920	0.5909	7.7676	-112.2413	-120.0089	-0.1756	-0.2494	-1.9125
0.5335	0.3792	250	0.4786	0.4788	0.6093	0.6035	9.3911	-115.1676	-124.5586	-0.2049	-0.2949	-1.9601
0.5282	0.4550	300	0.4665	0.4686	0.6376	0.6306	12.4430	-143.1194	-155.5625	-0.4844	-0.6050	-1.9915
0.4989	0.5308	350	0.4669	0.4697	0.6370	0.6323	14.0382	-141.6395	-155.6777	-0.4696	-0.6061	-2.1328
0.4806	0.6067	400	0.4599	0.4608	0.6493	0.6370	14.2207	-129.4153	-143.6360	-0.3473	-0.4857	-2.1904
0.4752	0.6825	450	0.4531	0.4555	0.6546	0.6454	15.4063	-148.1931	-163.5994	-0.5351	-0.6853	-2.1390
0.4943	0.7583	500	0.4544	0.4545	0.6572	0.6535	17.1657	-148.8972	-166.0628	-0.5421	-0.7100	-2.1458
0.4309	0.8342	550	0.4602	0.4618	0.6577	0.6463	17.9948	-158.6875	-176.6823	-0.6401	-0.8162	-2.1403
0.4533	0.9100	600	0.4523	0.4541	0.6636	0.6521	17.8286	-180.8165	-198.6450	-0.8613	-1.0358	-2.1089
0.4929	0.9858	650	0.4402	0.4397	0.6709	0.6530	15.6406	-138.0694	-153.7100	-0.4339	-0.5865	-2.0532

Framework versions

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