# === Model Configuration ===
base_model: Qwen/Qwen3-14B-Base
load_in_8bit: false
load_in_4bit: true

# === HF Configuration === 
hub_model_id: ToastyPigeon/q3-14b-completion-lora
hub_strategy: "every_save"

# === Training Setup ===
num_epochs: 2
micro_batch_size: 1
gradient_accumulation_steps: 8
sequence_len: 16384
sample_packing: true
pad_to_sequence_len: false

# === Evaluation ===
#val_set_size: 250
#evals_per_epoch: 5
#eval_steps: 20
#max_steps: 60
#eval_table_size:
#eval_max_new_tokens: 256
#eval_sample_packing: true
eval_strategy: "no"

# === Experimental QAT ===
qat:
  activation_dtype: int4
  weight_dtype: int4
  group_size: 32
  fake_quant_after_n_steps: 10


# === LoRA Configuration ===
adapter: qlora
lora_model_dir:
lora_r: 16
lora_alpha: 32
lora_dropout: 0.25
lora_target_linear: 
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

#lora_mlp_kernel: true
#lora_qkv_kernel: true
#lora_o_kernel: true

# === Hyperparameter Configuration ===
#optimizer: apollo_adamw_layerwise
warmup_steps: 20
optimizer: adamw_torch_fused
#optimizer: paged_adamw_8bit
#optim_args:
#  enable_stochastic_rounding: true
#  enable_cautious: true
#  enable_8bit: true
# Apollo-mini configuration:
#optim_args: "proj=random,rank=128,scale=128.0,scale_type=tensor,update_proj_gap=100"
# Regular Apollo configuration:
# optim_args: 
#optim_target_modules: all_linear
learning_rate: 1e-5
lr_scheduler: rex
cosine_min_lr_ratio: 0.2
#lr_scheduler: cosine_with_min_lr
#lr_scheduler_kwargs:
#  cosine_min_lr: 1e-6
weight_decay: 0.01
max_grad_norm: 1.0
#warmup_steps: 0
#warmup_ratio: 0.025


# === Data Configuration ===
#chat_template: jinja
#chat_template_jinja: "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{{ '<start_of_turn>' + role + '\n' + message['content'] | trim + '<end_of_turn>\n' }}{% endfor %}{% if add_generation_prompt %}{{'<start_of_turn>model\n'}}{% endif %}"
special_tokens:
  eos_token: "<|im_end|>"
#chat_template: jinja
#chat_template_jinja: chat_template.jinja
shuffle_merged_datasets: true
datasets:
  - path: ToastyPigeon/new-story-dataset
    type: completion
    data_files: new-story-dataset-v2.json
  - path: ToastyPigeon/new-story-dataset
    type: completion
    data_files: ehl_samples.json
  - path: ToastyPigeon/some-erotica
    type: completion
    split: train[:50%]
  - path: ToastyPigeon/skein-text-adventures
    type: completion
    split: train[:50%]
  - path: ToastyPigeon/SpringDragon
    type: completion
    split: train
  - path: ToastyPigeon/disco-chat
    type: completion
    split: train
dataset_prepared_path: last_run_prepared


# === Plugins ===
plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

# === Hardware Optimization ===
#gradient_checkpointing: offload
#gradient_checkpointing_kwargs:
#  use_reentrant: false
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
cut_cross_entropy: true

#deepspeed: axolotl/deepspeed_configs/zero3_bf16.json

# === FSDP Config === 
fsdp:
  - full_shard
  - auto_wrap
fsdp_config:
  fsdp_limit_all_gathers: true
  fsdp_sync_module_states: true
  fsdp_offload_params: true
  fsdp_activation_checkpointing: true
  fsdp_use_orig_params: false
  fsdp_cpu_ram_efficient_loading: true
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_transformer_layer_cls_to_wrap: Qwen3DecoderLayer
  fsdp_state_dict_type: FULL_STATE_DICT
  fsdp_sharding_strategy: FULL_SHARD


# === Wandb Tracking ===
wandb_project: Qwen3
# wandb_entity: [WANDB_ENTITY]
# wandb_name: [WANDB_RUN_NAME]

# === Checkpointing ===
saves_per_epoch: 20
save_total_limit: 1

# === Advanced Settings ===
output_dir: /workspace/aibox-standalone-pool/axolotl/qwen14b-ckpts
bf16: auto
flash_attention: true
train_on_inputs: false
group_by_length: false
save_safetensors: true
logging_steps: 1
gc_steps: 10
seed: 69

q3-14b-completion-lora

This model is a fine-tuned version of Qwen/Qwen3-14B-Base on the ToastyPigeon/new-story-dataset, the ToastyPigeon/new-story-dataset, the ToastyPigeon/some-erotica, the ToastyPigeon/skein-text-adventures, the ToastyPigeon/SpringDragon and the ToastyPigeon/disco-chat datasets.

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: 1
eval_batch_size: 1
seed: 69
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 8
total_train_batch_size: 16
total_eval_batch_size: 2
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 20
num_epochs: 2.0

Training results

Framework versions

PEFT 0.15.2
Transformers 4.51.3
Pytorch 2.7.1+cu126
Datasets 3.5.0
Tokenizers 0.21.1