llama3-chat_1M

This model is a fine-tuned version of unsloth/llama-3-8b-Instruct-bnb-4bit on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.3835, 39.5 bleu on PhoMT test en-vi, 34.4 on IWSLT15

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: 0.0002
• train_batch_size: 16
• eval_batch_size: 16
• seed: 3407
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 5
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.6092	0.032	500	1.4727
1.539	0.064	1000	1.4609
1.5211	0.096	1500	1.4528
1.5228	0.128	2000	1.4453
1.5106	0.16	2500	1.4431
1.5023	0.192	3000	1.4393
1.506	0.224	3500	1.4377
1.4887	0.256	4000	1.4342
1.4942	0.288	4500	1.4334
1.4826	0.32	5000	1.4307
1.4895	0.352	5500	1.4269
1.4854	0.384	6000	1.4249
1.4799	0.416	6500	1.4246
1.4837	0.448	7000	1.4227
1.4766	0.48	7500	1.4223
1.4799	0.512	8000	1.4206
1.4728	0.544	8500	1.4177
1.4753	0.576	9000	1.4173
1.4705	0.608	9500	1.4153
1.4679	0.64	10000	1.4159
1.4646	0.672	10500	1.4163
1.4601	0.704	11000	1.4135
1.4648	0.736	11500	1.4113
1.4618	0.768	12000	1.4109
1.4644	0.8	12500	1.4096
1.4593	0.832	13000	1.4084
1.4629	0.864	13500	1.4080
1.4565	0.896	14000	1.4079
1.4502	0.928	14500	1.4043
1.4558	0.96	15000	1.4024
1.45	0.992	15500	1.4040
1.3885	1.024	16000	1.4058
1.3681	1.056	16500	1.4071
1.3719	1.088	17000	1.4074
1.3687	1.12	17500	1.4063
1.3736	1.152	18000	1.4067
1.3767	1.184	18500	1.4061
1.3764	1.216	19000	1.4036
1.3751	1.248	19500	1.4031
1.3698	1.28	20000	1.4031
1.3764	1.312	20500	1.4024
1.379	1.3440	21000	1.4012
1.3758	1.376	21500	1.3990
1.3764	1.408	22000	1.3996
1.3715	1.44	22500	1.3982
1.3775	1.472	23000	1.3976
1.3719	1.504	23500	1.3974
1.3745	1.536	24000	1.3973
1.3704	1.568	24500	1.3961
1.3659	1.6	25000	1.3950
1.3665	1.6320	25500	1.3947
1.3628	1.6640	26000	1.3923
1.367	1.696	26500	1.3915
1.3616	1.728	27000	1.3899
1.3671	1.76	27500	1.3891
1.3651	1.792	28000	1.3884
1.3609	1.8240	28500	1.3872
1.3647	1.8560	29000	1.3871
1.3595	1.888	29500	1.3852
1.3579	1.92	30000	1.3845
1.3575	1.952	30500	1.3837
1.3576	1.984	31000	1.3835
1.3102	2.016	31500	1.3964
1.2595	2.048	32000	1.3966
1.2622	2.08	32500	1.3978
1.2606	2.112	33000	1.3967
1.2665	2.144	33500	1.3982
1.2658	2.176	34000	1.3974
1.2574	2.208	34500	1.3971
1.2584	2.24	35000	1.3963
1.2635	2.2720	35500	1.3970
1.2579	2.304	36000	1.3956
1.2633	2.336	36500	1.3956
1.2602	2.368	37000	1.3952
1.2597	2.4	37500	1.3953
1.2635	2.432	38000	1.3948
1.2646	2.464	38500	1.3947
1.2609	2.496	39000	1.3946
1.2562	2.528	39500	1.3941
1.2586	2.56	40000	1.3943
1.2604	2.592	40500	1.3940
1.2636	2.624	41000	1.3940
1.2635	2.656	41500	1.3940
1.2587	2.6880	42000	1.3938
1.2603	2.7200	42500	1.3939
1.2592	2.752	43000	1.3937
1.2568	2.784	43500	1.3934
1.2595	2.816	44000	1.3936
1.2565	2.848	44500	1.3935
1.2585	2.88	45000	1.3936
1.2624	2.912	45500	1.3933
1.2581	2.944	46000	1.3934
1.2571	2.976	46500	1.3934

Framework versions

• PEFT 0.10.0
• Transformers 4.40.2
• Pytorch 2.3.0
• Datasets 2.19.1
• Tokenizers 0.19.1