See axolotl config

axolotl version: 0.4.0

base_model: google/gemma-2b
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: OdiaGenAIdata/culturax-odia
    type: completion
    field: text
dataset_prepared_path:
val_set_size: 0.1
output_dir: ./gemma-odia-2b-pretrain-v1
hub_model_id: sam2ai/gemma_odia_2b_v1

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

wandb_project: gemma-completion-2b-odia-v1
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

lora_r: 64
lora_alpha: 128
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
  - gate_proj
  - down_proj
  - up_proj
lora_modules_to_save:
  - embed_tokens
  - lm_head
lora_target_linear: true
lora_fan_in_fan_out:

gradient_accumulation_steps: 8
micro_batch_size: 2
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: false
fp16: false
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
eval_sample_packing: False
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

save_safetensors: True

gemma_odia_2b_v1

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

• Loss: 3.2357

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: 8
• total_train_batch_size: 128
• 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: 6
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
48.3861	0.0	1	48.2747
3.3986	0.25	169	3.2901
3.3659	0.5	338	3.2334
3.1731	0.75	507	3.0614
3.1942	1.0	676	3.0977
3.3983	1.24	845	3.3234
3.3853	1.49	1014	3.2983
3.3254	1.74	1183	3.2357

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