See axolotl config

axolotl version: 0.13.0.dev0

## model
base_model: shb777/Llama-3.3-8B-Instruct-128K

## qlora COPE!!!
load_in_8bit: false
load_in_4bit: false
strict: false

# === Data Configuration ===
datasets:
  - path: WokeAI/polititune-tankie-warmup
    type: chat_template
    split: train

chat_template: llama3

shuffle_merged_datasets: true
dataset_prepared_path: dataset_prepareds
val_set_size: 0.0
output_dir: ./output

# === LoRA Configuration ===
adapter: lora
lora_r: 64
lora_alpha: 16
lora_dropout: 0.35
lora_target_modules:
lora_target_linear: true
peft_use_rslora: true
max_grad_norm: 0.1

## Liger + CCE
plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: false
cut_cross_entropy: true

## CTX settings
sequence_len: 4096
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

## WandB
wandb_project: newyear
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

## hoe params
gradient_accumulation_steps: 2 # ???
micro_batch_size: 2
num_epochs: 2
lr_scheduler: rex
learning_rate: 1e-5
optimizer: adamw_torch_8bit # Options: "paged_ademamix_8bit", "adamw_bnb_8bit", "paged_adamw_8bit"

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

special_tokens:
  pad_token: <|reserved_special_token_2|>

gradient_checkpointing: offload
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

output

This model is a fine-tuned version of shb777/Llama-3.3-8B-Instruct-128K on the WokeAI/polititune-tankie-warmup dataset.

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: 1e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Use OptimizerNames.ADAMW_TORCH_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: 2
• training_steps: 84

Training results

Framework versions

• PEFT 0.18.0
• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.4.2
• Tokenizers 0.22.1