See axolotl config

axolotl version: 0.4.0

# use google/gemma-7b if you have access
base_model: unsloth/gemma-2b
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

# huggingface repo
datasets:
  - path: OdiaGenAIdata/culturax-gemma-data
    type: completion
val_set_size: 0.1
output_dir: ./gemma-odia-2b-pretrain-unsloth
hub_model_id: sam2ai/gemma_odia_2b_unsloth

adapter: qlora
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

wandb_project: gemma-completion-2b-odia-unsloth
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:


gradient_accumulation_steps: 3
micro_batch_size: 2
num_epochs: 10
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: false

warmup_ratio: 0.1
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

gemma_odia_2b_unsloth

This model is a fine-tuned version of unsloth/gemma-2b on the None dataset. It achieves the following results on the evaluation set:

• Loss: 3.4007

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: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 3
• total_train_batch_size: 48
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 87
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
48.3499	0.0	1	48.2901
22.3743	0.25	449	22.4176
22.3342	0.5	898	22.3606
9.2934	0.75	1347	9.2611
3.8237	1.0	1796	3.5233
4.7071	1.24	2245	4.3919
5.0601	1.49	2694	4.8608
3.966	1.74	3143	3.7664
3.7972	1.99	3592	3.6383
3.802	2.22	4041	3.4831
3.7412	2.47	4490	3.4955
3.6174	2.72	4939	3.4462
3.6126	2.97	5388	3.3908
3.5759	3.2	5837	3.3827
3.4854	3.45	6286	3.3748
3.4987	3.7	6735	3.2868
14.4221	3.95	7184	14.0660
16.2072	4.19	7633	15.8277
3.5762	4.44	8082	3.3616
15.155	4.69	8531	15.1050
3.7657	4.94	8980	3.6526
5.0469	5.17	9429	4.8438
4.0484	5.42	9878	3.8946
4.0601	5.67	10327	3.8040
3.7711	5.92	10776	3.5799
3.6364	6.16	11225	3.4930
3.5855	6.41	11674	3.4586
3.5484	6.66	12123	3.4197
3.8341	6.91	12572	3.6314
3.5392	7.14	13021	3.4121
3.6463	7.39	13470	3.3959
3.6237	7.64	13919	3.4071
3.542	7.89	14368	3.4076
3.5737	8.13	14817	3.4041
3.6167	8.38	15266	3.4153
3.6356	8.63	15715	3.4068
3.5233	8.88	16164	3.4054
3.5382	9.11	16613	3.4019
3.5788	9.36	17062	3.4008
3.7003	9.61	17511	3.4007

Framework versions

• PEFT 0.9.0
• Transformers 4.40.0.dev0
• Pytorch 2.4.0.dev20240326+rocm6.0
• Datasets 2.18.0
• Tokenizers 0.15.0