This model is a fine-tuned version of slm-research-vn/Qwen2-7B-Instruct-SPPO-Function-call-v2.5 on the slm-research-vn/dpo-format-function-calling-v4, the slm-research-vn/dpo-format-glaive-code-assistant-v3-with-mistral-large-slm-iter4 and the argilla/dpo-mix-7k datasets. It achieves the following results on the evaluation set:

Loss: 0.3005
Rewards/chosen: 1.6737
Rewards/rejected: -0.4932
Rewards/accuracies: 0.8699
Rewards/margins: 2.1670
Logps/rejected: -276.8380
Logps/chosen: -200.9362
Logits/rejected: -0.6568
Logits/chosen: -0.6408

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: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 4
total_train_batch_size: 32
total_eval_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6437	0.0916	100	0.6128	0.3050	0.0739	0.7254	0.2311	-265.4963	-228.3116	-0.7319	-0.7206
0.5175	0.1832	200	0.4987	1.1265	0.2914	0.8237	0.8351	-261.1460	-211.8815	-0.7134	-0.7068
0.3903	0.2749	300	0.4279	1.7297	0.4889	0.8468	1.2408	-257.1960	-199.8173	-0.6700	-0.6642
0.3712	0.3665	400	0.3781	1.7272	0.2255	0.8468	1.5017	-262.4645	-199.8672	-0.6756	-0.6691
0.3064	0.4581	500	0.3477	1.7220	-0.0183	0.8613	1.7403	-267.3389	-199.9704	-0.6642	-0.6488
0.3054	0.5497	600	0.3271	1.6469	-0.1977	0.8671	1.8447	-270.9281	-201.4723	-0.6576	-0.6407
0.2919	0.6413	700	0.3144	1.7376	-0.3034	0.8642	2.0410	-273.0414	-199.6590	-0.6753	-0.6672
0.314	0.7329	800	0.3056	1.7037	-0.4229	0.8671	2.1266	-275.4323	-200.3379	-0.6685	-0.6574
0.3014	0.8246	900	0.3020	1.6807	-0.4632	0.8699	2.1439	-276.2374	-200.7971	-0.6702	-0.6641
0.268	0.9162	1000	0.2999	1.6798	-0.4929	0.8844	2.1726	-276.8312	-200.8157	-0.6690	-0.6635

Framework versions

PEFT 0.12.0
Transformers 4.44.0
Pytorch 2.3.1+cu121
Datasets 2.20.0
Tokenizers 0.19.1

khongtrunght/Qwen2-7B-Instruct-SPPO-Function-call-v2.6

Qwen2-7B-Instruct-SPPO-Function-call-v2.6

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

khongtrunght/
Qwen2-7B-Instruct-SPPO-Function-call-v2.6