See axolotl config

axolotl version: 0.6.0

# 29 but using llamafied unsloth/phi-4 vs microsoft/phi-4 

base_model: unsloth/phi-4
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

# User Liger
plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: true

chat_template: llama3
datasets:
  - path: shisa-ai/shisa-v1-athenev2-reannotated-filtered
    # type: sharegpt deprecated
    type: chat_template
    field_messages: conversations
    message_field_role: from
    message_field_content: value
dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./outputs/ablation-33-rafathenev2.unphi4-shisa-v2-unphi-4-14b

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

# marginal difference
neftune_noise_alpha: 5

use_wandb: true
wandb_project: shisa-v2
wandb_entity: augmxnt
wandb_name: ablation-33-rafathenev2.unphi4-shisa-v2-unphi-4-14b

gradient_accumulation_steps: 2
micro_batch_size: 4
num_epochs: 3
optimizer: paged_adamw_8bit
lr_scheduler: linear
learning_rate: 8e-6

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

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_ratio: 0.05
evals_per_epoch: 2
eval_table_size:
saves_per_epoch: 0
save_total_limit: 1 # Only store a single checkpoint
debug:
deepspeed: zero3_bf16.json
weight_decay: 0.00
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

outputs/ablation-33-rafathenev2.unphi4-shisa-v2-unphi-4-14b

This model is a fine-tuned version of unsloth/phi-4 on the shisa-ai/shisa-v1-athenev2-reannotated-filtered dataset. It achieves the following results on the evaluation set:

• Loss: 0.2875

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: 8e-06
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 34
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.4883	0.0043	1	0.3333
0.3985	0.5021	117	0.2838
0.3448	1.0043	234	0.2770
0.2902	1.5064	351	0.2787
0.2914	2.0086	468	0.2781
0.2927	2.5107	585	0.2875

Framework versions

• Transformers 4.48.3
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0