See axolotl config

axolotl version: 0.13.0.dev0

base_model: syvai/jacob-kan-tale-v2

hub_model_id: syvai/jacob-kan-tale-v3

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

datasets:
  - path: syvai/jacob-tts-zac
    type:  # leave empty to load pre-tokenized
dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ./outputs/out

sequence_len: 4092
sample_packing: true


wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 4
num_epochs: 2
optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 2e-5

bf16: auto
tf32: false

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

warmup_ratio: 0.01
evals_per_epoch: 1
saves_per_epoch: 1
weight_decay: 0.05

special_tokens:
  pad_token: <custom_token_7>

# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

jacob-kan-tale-v3

This model is a fine-tuned version of syvai/jacob-kan-tale-v2 on the syvai/jacob-tts-zac dataset. It achieves the following results on the evaluation set:

• Loss: 4.9049
• Memory/max Active (gib): 24.76
• Memory/max Allocated (gib): 24.76
• Memory/device Reserved (gib): 29.81

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: 2e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• 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
• training_steps: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	5.0922	12.44	12.44	12.48
5.1069	0.8	1	4.9320	24.76	24.76	29.81
4.9685	1.0	2	4.9049	24.76	24.76	29.81

Framework versions

• Transformers 4.55.3
• Pytorch 2.7.1+cu126
• Datasets 2.19.2
• Tokenizers 0.21.4