See axolotl config

axolotl version: 0.13.0.dev0

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:

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