See axolotl config

axolotl version: 0.4.0

base_model: meta-llama/Meta-Llama-3-8B
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: ./llama_3_8b_pretrain_v2
hub_model_id: sam2ai/llama3_8b_odia_v2

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

lora_r: 64
lora_alpha: 128
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
#lora_modules_to_save:
#  - embed_tokens
#  - lm_head
lora_fan_in_fan_out:

wandb_project: llama-3-8b-pretrain-odia-plain
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 2
num_epochs: 4
optimizer: paged_adamw_32bit
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_steps: 10
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: "<|end_of_text|>"
save_safetensors: True

llama3_8b_odia_v2

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B on the None dataset. It achieves the following results on the evaluation set:

• Loss: nan

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: 10
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
13.7841	0.0007	1	nan
0.0	0.25	384	nan
0.0	0.5	768	nan
0.0	0.75	1152	nan
0.0	1.0	1536	nan
0.0	1.2362	1920	nan
0.0	1.4862	2304	nan
0.0	1.7362	2688	nan
0.0	1.9862	3072	nan
0.0	2.2220	3456	nan
0.0	2.4720	3840	nan
0.0	2.7220	4224	nan
0.0	2.9720	4608	nan
0.0	3.2078	4992	nan
0.0	3.4578	5376	nan
0.0	3.7078	5760	nan
0.0	3.9578	6144	nan

Framework versions

• PEFT 0.9.0
• Transformers 4.40.0
• Pytorch 2.4.0.dev20240326+rocm6.0
• Datasets 2.15.0
• Tokenizers 0.19.1