README.md

metadata

license: apache-2.0
base_model: hZzy/qwen2.5-0.5b-sft-news-IFT
tags:
  - alignment-handbook
  - ndcg
  - trl
  - expo
  - generated_from_trainer
  - trl
  - expo
  - generated_from_trainer
datasets:
  - hZzy/train_pairwise
model-index:
  - name: qwen2.5-0.5b-expo-L2EXPO-EXPERIMENT-0.1-1e6
    results: []

qwen2.5-0.5b-expo-L2EXPO-EXPERIMENT-0.1-1e6

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: 0.4062
• Logps: -90.0446
• Logits: -1.4303
• Objective: 0.4077
• Dpo Loss: 0.6829
• Regularize: 0.4077
• Ranking Simple: 0.5248
• Ranking Idealized: 0.5888
• Ranking Idealized Expo: 0.5103

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: 6
• gradient_accumulation_steps: 12
• total_train_batch_size: 288
• total_eval_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
0.3854	0.2834	50	0.4056	-91.4065	-1.4801	0.4076	0.6886	0.4076	0.5124	0.5888	0.5103
0.3126	0.5668	100	0.4022	-91.3166	-1.4526	0.4009	0.6817	0.4009	0.5207	0.5888	0.5103
0.2481	0.8503	150	0.4118	-93.3285	-1.4781	0.4156	0.6853	0.4156	0.5186	0.5888	0.5103
0.1986	1.1337	200	0.4053	-90.6332	-1.4691	0.4089	0.6828	0.4089	0.5207	0.5888	0.5103
0.1805	1.4171	250	0.4086	-90.2497	-1.4648	0.4084	0.6831	0.4084	0.5248	0.5888	0.5103
0.1668	1.7005	300	0.4080	-89.8657	-1.4761	0.4114	0.6842	0.4114	0.5207	0.5888	0.5103
0.1476	1.9839	350	0.4086	-89.6008	-1.4348	0.4084	0.6835	0.4084	0.5217	0.5888	0.5103
0.1232	2.2674	400	0.4064	-89.9367	-1.4142	0.4060	0.6825	0.4060	0.5238	0.5888	0.5103
0.1085	2.5508	450	0.4057	-90.6112	-1.4381	0.4068	0.6829	0.4068	0.5238	0.5888	0.5103
0.099	2.8342	500	0.4075	-89.7867	-1.4538	0.4090	0.6837	0.4090	0.5248	0.5888	0.5103
0.0841	3.1176	550	0.4074	-89.1923	-1.4288	0.4091	0.6836	0.4091	0.5269	0.5888	0.5103
0.0673	3.4010	600	0.4056	-89.8307	-1.4326	0.4069	0.6824	0.4069	0.5238	0.5888	0.5103
0.0589	3.6845	650	0.4060	-89.4758	-1.4302	0.4077	0.6829	0.4077	0.5248	0.5888	0.5103
0.0551	3.9679	700	0.4065	-90.0660	-1.4301	0.4080	0.6831	0.4080	0.5238	0.5888	0.5103
0.042	4.2513	750	0.4064	-90.0447	-1.4307	0.4078	0.6830	0.4078	0.5248	0.5888	0.5103
0.0411	4.5347	800	0.4062	-90.1140	-1.4310	0.4078	0.6830	0.4078	0.5238	0.5888	0.5103
0.0355	4.8181	850	0.4062	-90.0432	-1.4302	0.4077	0.6829	0.4077	0.5248	0.5888	0.5103

Framework versions

• Transformers 4.42.0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1