See axolotl config

axolotl version: 0.10.0.dev0

base_model: google/gemma-3-12b-it

model_config:
  attn_implementation: eager

overrides_of_model_kwargs:
  attn_implementation: eager

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: thesantatitan/pixelprose-sample-5k
    type: chat_template
    split: train
    chat_template: tokenizer_default
    field_messages: messages
    roles_to_train: ["assistant"]

dataset_prepared_path: text2svg-prepared-pixelprose
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 4096
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: false

adapter: lora
lora_model_dir:
lora_r: 128
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_modules_to_save: # required when adding new tokens to LLaMA/Mistral
  - embed_tokens
  - lm_head

wandb_project: svg-sft-gemma-12b-saved
wandb_entity:
wandb_watch:
wandb_run_id: sexyrun1

gradient_accumulation_steps: 32
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.0001

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

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

warmup_steps: 10
save_steps: 20
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json
weight_decay: 0.0
fsdp:
fsdp_config:

hub_strategy: every_save
hub_model_id: thesantatitan/gemma-svg-sft

gemma-svg-sft

This model is a fine-tuned version of google/gemma-3-12b-it on the thesantatitan/pixelprose-sample-5k dataset. It achieves the following results on the evaluation set:

• Loss: 0.7442

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 128
• total_eval_batch_size: 4
• 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
• num_epochs: 4.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.7293	0.9832	33	0.7807
0.6371	1.9832	66	0.7512
0.6369	2.9832	99	0.7448
0.6108	3.9832	132	0.7442

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.1
• Tokenizers 0.21.1