hardlyworking/Aura-12B-FT

See axolotl config

axolotl version: 0.8.0.dev0

base_model: hardlyworking/Aura-12B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: hardlyworking/LighterRPSet
    type: chat_template
    chat_template: chatml
    roles_to_train: ["gpt"]
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    train_on_eos: turn
  - path: jeiku/Writing
    type: completion
    field: text

shuffle_merged_datasets: true
dataset_prepared_path:
val_set_size: 0.01
output_dir: ./output/out

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: true

hub_model_id: hardlyworking/Test12B
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

wandb_project: Aura-12B
wandb_entity:
wandb_watch:
wandb_name: Aura-12B
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 5e-6

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: true

max_grad_norm: 0.01

warmup_ratio: 0.1
evals_per_epoch: 2
eval_table_size:
eval_max_new_tokens:
saves_per_epoch: 1
debug:
deepspeed: 
weight_decay: 0.05
fsdp:
fsdp_config:
special_tokens:

Test12B

This model is a fine-tuned version of hardlyworking/Aura-12B on the hardlyworking/LighterRPSet and the jeiku/Writing datasets. It achieves the following results on the evaluation set:

• Loss: 2.2116

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: 5e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 41
• num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss
2.2984	0.0048	1	2.5678
1.8777	0.5009	104	2.2361
2.5154	1.0048	208	2.2155
1.8757	1.5057	312	2.2116

Framework versions

• Transformers 4.50.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1