base_model: Qwen/Qwen3-14B
# Automatically upload checkpoint and final model to HF
hub_model_id: JustQuiteMadMax/Qwen3-14B-ZNO

plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
strict: false

chat_template: qwen3
datasets:
  - path: JustQuiteMadMax/ZNO_Test_Conversations
    type: chat_template
    split: train
    field_messages: messages
    message_property_mappings:
      role: from
      content: value
val_set_size: 0.2
output_dir: ./outputs/out
dataset_prepared_path: last_run_prepared

sequence_len: 3072
sample_packing: true
eval_sample_packing: true
pad_to_sequence_len: true

load_in_4bit: true
adapter: qlora
lora_r: 16
lora_alpha: 32
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - down_proj
  - up_proj
lora_mlp_kernel: true
lora_qkv_kernel: true
lora_o_kernel: true

use_wandb: true
wandb_project: genai-hw3-zno-train
wandb_entity: m-rudko-pn-ukrainian-catholic-university
wandb_watch:
wandb_name:
wandb_log_model: checkpoint

gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_torch_4bit
lr_scheduler: cosine
learning_rate: 0.0002

bf16: auto
tf32: true

gradient_checkpointing: offload
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
weight_decay: 0.0
special_tokens:

Visualize in Weights & Biases

Qwen3-14B-ZNO

This model is a fine-tuned version of Qwen/Qwen3-14B on the JustQuiteMadMax/ZNO_Test_Conversations dataset. It achieves the following results on the evaluation set:

Loss: 0.5582
Memory/max Active (gib): 10.36
Memory/max Allocated (gib): 10.36
Memory/device Reserved (gib): 11.98

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: 0.0002
train_batch_size: 1
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 2
optimizer: Use OptimizerNames.ADAMW_TORCH_4BIT 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: 10
training_steps: 171

Training results

Training Loss	Epoch	Step	Validation Loss	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	2.0783	10.3	10.3	15.87
0.6365	0.2507	43	0.5787	10.36	10.36	11.98
0.5913	0.5015	86	0.5604	10.36	10.36	11.98
0.5179	0.7522	129	0.5582	10.36	10.36	11.98

Framework versions

PEFT 0.18.1
Transformers 4.57.1
Pytorch 2.9.1+cu128
Datasets 4.4.1
Tokenizers 0.22.2