timarni/qwen3_pre_books_mmlu

See axolotl config

axolotl version: 0.9.2

base_model: timarni/qwen3_pretrain_textbooks
# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name

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

chat_template: qwen3
datasets:
  - path: timarni/MNLP_dataset_mmlu_train
    type: alpaca
    split: train

val_set_size: 0.15
output_dir: ./outputs/qwen3_pretrain_textbooks+mmlu
dataset_prepared_path: last_run_prepared

sequence_len: 4096 #2048
sample_packing: true
eval_sample_packing: true
pad_to_sequence_len: true

# To be sure that no LORA is done
adapter: null
lora: false
merge_lora: false

wandb_project: mnlp_project
wandb_entity: tim-arni
wandb_watch:
wandb_name: qwen3-0.6B-pretrained-textbooks+mmlu
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.00005 # 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:

outputs/qwen3_pretrain_textbooks+mmlu

This model is a fine-tuned version of timarni/qwen3_pretrain_textbooks on the timarni/MNLP_dataset_mmlu_train dataset. It achieves the following results on the evaluation set:

• Loss: 0.0613

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: 5e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• total_eval_batch_size: 4
• optimizer: Use adamw_torch 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
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.2791	0.0009	1	0.4097
0.069	0.2502	275	0.0736
0.0552	0.5005	550	0.0646
0.0633	0.7507	825	0.0613

Framework versions

• Transformers 4.51.3
• Pytorch 2.5.1+cu121
• Datasets 3.5.1
• Tokenizers 0.21.1