base_model: google/gemma-3-12b-it
hub_model_id: lmagiera/gemma-3-12b-it-tuned-lum5

load_in_8bit: false
load_in_4bit: true
strict: false

# datasets:
#   - path: lmagiera/fake-football-cup # <-- Use your new dataset path
#     type: dolly                     # <-- 'dolly' is a simpler way to map the fields
# val_set_size: 0.05                  # <-- Keep this to create a validation set
# databricks/databricks-dolly-15k
# lmagiera/fake-football-cup

# datasets:
#   - path: lmagiera/fake-football-cup
#     type:
#       field_instruction: instruction       
#       field_input: context
#       field_output: response
# val_set_size: 0.05

datasets:
  - path: /mnt/disks/gcs/training/datasets/manual/prepared_dataset # <-- Point to the correct subfolder
    ds_type: arrow
    # Be explicit about the columns to avoid other errors
    type:
      field_instruction: instruction
      field_input: context
      field_output: response
val_set_size: 0.05
 

sequence_len: 2048
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: true

adapter: qlora
lora_model_dir:
lora_r: 32
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 4
num_epochs: 1
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: false

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
weight_decay: 0.0
debug:
deepspeed: deepspeed_configs/zero2.json
fsdp:
fsdp_config:
special_tokens:
output_dir: "/mnt/disks/gcs/training/jobs/google--gemma-3-12b-it-20250602-144550/out/"
dataset_prepared_path: "/mnt/disks/gcs/training/datasets"

gemma-3-12b-it-tuned-lum5

This model is a fine-tuned version of google/gemma-3-12b-it on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 3.3421

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: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 4
total_train_batch_size: 32
total_eval_batch_size: 8
optimizer: Use OptimizerNames.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
training_steps: 3

Training results

Training Loss	Epoch	Step	Validation Loss
2.874	0.3333	1	3.9189
2.6864	0.6667	2	3.8282
2.9838	1.0	3	3.3421

Framework versions

PEFT 0.15.2
Transformers 4.51.3
Pytorch 2.7.0+cu126
Datasets 3.5.1
Tokenizers 0.21.1