mistral-7b-expo-7b-L2EXPO-25-3

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: 0.4241
• Objective: 0.4184
• Reward Accuracy: 0.6924
• Logp Accuracy: 0.6784
• Log Diff Policy: 19.6421
• Chosen Logps: -184.7314
• Rejected Logps: -204.3735
• Chosen Rewards: -0.9006
• Rejected Rewards: -1.0934
• Logits: -2.4579

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: 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: 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
0.5314	0.1011	50	0.4821	0.4763	0.6228	0.6093	10.3431	-109.8486	-120.1917	-0.1518	-0.2516	-1.9219
0.4837	0.2022	100	0.4592	0.4504	0.6767	0.6675	19.7753	-191.6833	-211.4587	-0.9701	-1.1643	-2.0366
0.439	0.3033	150	0.4358	0.4278	0.6860	0.6625	15.8163	-178.8420	-194.6583	-0.8417	-0.9963	-2.4778
0.4082	0.4044	200	0.4291	0.4206	0.6969	0.6809	18.5694	-196.0356	-214.6050	-1.0136	-1.1957	-2.4561
0.3852	0.5056	250	0.4314	0.4233	0.6874	0.6697	19.4767	-198.4446	-217.9214	-1.0377	-1.2289	-2.3852
0.3499	0.6067	300	0.4265	0.4197	0.6879	0.6706	19.3593	-169.0813	-188.4406	-0.7441	-0.9341	-2.4955
0.3632	0.7078	350	0.4240	0.4172	0.6902	0.6734	19.4575	-185.9196	-205.3771	-0.9125	-1.1034	-2.6119
0.3285	0.8089	400	0.4251	0.4171	0.6876	0.6555	15.3121	-220.3820	-235.6940	-1.2571	-1.4066	-2.5469
0.3226	0.9100	450	0.4211	0.4121	0.6952	0.6739	18.1247	-182.7602	-200.8849	-0.8809	-1.0585	-2.6342

Framework versions

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