See axolotl config

axolotl version: 0.13.0.dev0

base_model: google/gemma-3-270m-it
# optionally might have model_type or tokenizer_type
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
hub_model_id: skomadinajs/gemma-3-270m-it-emoji

# gemma3 doesn't seem to play nice with ddp
ddp_find_unused_parameters: true

load_in_8bit: false
load_in_4bit: true

# huggingface repo
chat_template: gemma3
eot_tokens:
  - <end_of_turn>
datasets:
  - path: kr15t3n/text2emoji
    type:
      system_prompt: "Translate this text to emoji:"
      field_instruction: text
      field_output: emoji

val_set_size: 0.1
output_dir: /workspace-data/output

adapter: qlora
lora_r: 16
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true

sequence_len: 256
sample_packing: false
eval_sample_packing: false


wandb_project: gemma-3-270m-qlora-002
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:


gradient_accumulation_steps: 1
micro_batch_size: 4
num_epochs: 3
optimizer: adamw_torch_fused
lr_scheduler: constant
learning_rate: 0.00005

bf16: auto
tf32: true

gradient_checkpointing: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_ratio: 0.0
evals_per_epoch: 1
saves_per_epoch: 1
weight_decay: 0.01
special_tokens:

gemma-3-270m-it-emoji

This model is a fine-tuned version of google/gemma-3-270m-it on the kr15t3n/text2emoji dataset. It achieves the following results on the evaluation set:

• Loss: 2.3954
• Memory/max Active (gib): 1.05
• Memory/max Allocated (gib): 1.05
• Memory/device Reserved (gib): 1.47

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: 4
• eval_batch_size: 4
• seed: 42
• 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: constant
• lr_scheduler_warmup_steps: 51
• training_steps: 1707

Training results

Training Loss	Epoch	Step	Validation Loss	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	8.3493	1.02	1.02	1.57
2.9462	1.0	569	2.5616	1.05	1.05	1.53
2.2067	2.0	1138	2.4661	1.05	1.05	1.47
1.9333	3.0	1707	2.3954	1.05	1.05	1.47

Framework versions

• PEFT 0.17.1
• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.1