See axolotl config

axolotl version: 0.7.0

base_model: unsloth/phi-4

load_in_8bit: false
load_in_4bit: false

bf16: auto
fp16: 
tf32: false

datasets:
  - path: tamewild/y1_sft_split
    split: train
    type: chat_template
    field_messages: conversation

shuffle_merged_datasets: true

test_datasets:
  - path: tamewild/y1_sft_split
    split: validation
    type: chat_template
    field_messages: conversation

dataset_prepared_path: workspace/dataset_prepared

hub_model_id: tamewild/14b_v1_fft

hf_use_auth_token: true

sequence_len: 9000
pad_to_sequence_len: true
sample_packing: true
eval_sample_packing: true # disable if we get errors

# wandb configuration if you're using it
# Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.
wandb_mode: # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb
wandb_project: axolotl # Your wandb project name
wandb_entity: # A wandb Team name if using a Team
wandb_watch:
wandb_name: # Set the name of your wandb run
wandb_run_id: # Set the ID of your wandb run
wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training

output_dir: /workspace/tuned

torch_compile: auto

gradient_accumulation_steps: 2
micro_batch_size: 8
eval_batch_size: 8
num_epochs: 1
warmup_ratio: 0.01
learning_rate: 4.5e-5
logging_steps: 1
eval_steps: # Leave empty to eval at each epoch, integer for every N steps. float for fraction of total steps
evals_per_epoch: 4 # number of times per epoch to run evals, mutually exclusive with eval_steps
save_steps: # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps
saves_per_epoch: 4 # number of times per epoch to save a checkpoint, mutually exclusive with save_steps
save_total_limit: 1 # Checkpoints saved at a time

include_tokens_per_second: true

train_on_inputs: false
group_by_length: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false

lr_scheduler: cosine
lr_scheduler_kwargs:
cosine_min_lr_ratio: 0.025

optimizer: paged_adamw_8bit

weight_decay: 0.01

xformers_attention:
flash_attention: true

seed: 1234

strict: false

14b_v1_fft

This model is a fine-tuned version of unsloth/phi-4 on the tamewild/y1_sft_split dataset. It achieves the following results on the evaluation set:

• Loss: 0.3468

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: 4.5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 1234
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use paged_adamw_8bit 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: 3
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.7528	0.0033	1	0.7171
0.3488	0.2504	77	0.3663
0.3704	0.5008	154	0.3525
0.3338	0.7512	231	0.3468

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+gitf929e0d
• Datasets 3.2.0
• Tokenizers 0.21.0