See axolotl config

axolotl version: 0.9.2

base_model: Qwen/Qwen3-0.6B-Base
# 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_M3_mcqa_dataset # timarni/MNLP_intstruction_tuning
    name: stem_instruction_tuning_balanced_mini
    type: alpaca
    split: train

shuffle_merged_datasets: true

val_set_size: 0.1
output_dir: ./outputs/base_it_bal
dataset_prepared_path: last_run_prepared

sequence_len: 2048 #2048
sample_packing: true # was true -> need to check if it actually learns on the samples or not (better understand te hyperparam and event. install axolotl to debug)
eval_sample_packing: true
pad_to_sequence_len: true
train_on_inputs: false # NEW
# group_by_length: false NEW?

# 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: base_it_bal
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 1 # 2
num_epochs: 6
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 5e-6 # 0.00005
cosine_min_lr_ratio: 0.1

warmup_ratio: 0.05
weight_decay: 0.01

bf16: auto
tf32: true

gradient_checkpointing: offload
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
gradient_clipping: 1.0 # or max_grad_norm?
flash_attention: true

evals_per_epoch: 4
saves_per_epoch: 2
save_total_limit: 25
special_tokens:

outputs/base_it_bal

This model is a fine-tuned version of Qwen/Qwen3-0.6B-Base on the timarni/MNLP_M3_mcqa_dataset dataset. It achieves the following results on the evaluation set:

• Loss: 4.7220

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-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• total_eval_batch_size: 2
• 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: 217
• num_epochs: 6.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.8047	0.0014	1	6.2910
0.1479	0.2509	182	5.3610
0.1379	0.5017	364	4.9984
0.1154	0.7526	546	4.9730
0.1503	1.0028	728	4.8248
0.1373	1.2536	910	4.7810
0.1169	1.5045	1092	4.7103
0.1194	1.7553	1274	4.7154
0.1506	2.0055	1456	4.7224
0.1454	2.2564	1638	4.7103
0.1481	2.5072	1820	4.6918
0.141	2.7581	2002	4.6967
0.1495	3.0083	2184	4.6989
0.0994	3.2591	2366	4.7124
0.1369	3.5100	2548	4.7364
0.1266	3.7609	2730	4.7268
0.151	4.0110	2912	4.7217
0.088	4.2619	3094	4.6797
0.1203	4.5127	3276	4.7181
0.1598	4.7636	3458	4.7157
0.1399	5.0138	3640	4.6857
0.1258	5.2646	3822	4.7303
0.1296	5.5155	4004	4.7357
0.1174	5.7664	4186	4.7220

Framework versions

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