clevrpwn/gemma-3-270m-codealpaca-finetune

base_model: google/gemma-3-270m-it
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

ddp_find_unused_parameters: true

load_in_8bit: false
load_in_4bit: false

chat_template: gemma3
eot_tokens:
  - "<end_of_turn>"

datasets:
  - path: HuggingFaceH4/CodeAlpaca_20K
    type:
      field_instruction: prompt
      field_input: input
      field_output: output
      format: |
        Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

        ### Instruction:
        {instruction}

        ### Input:
        {input}

        ### Response:
      no_input_format: |
        Below is an instruction that describes a task. Write a response that appropriately completes the request.

        ### Instruction:
        {instruction}

        ### Response:

val_set_size: 0.05 # Use 5% of the data for validation
output_dir: ./outputs/gemma-3-270m-codealpaca-finetune

sequence_len: 2048
sample_packing: true
eval_sample_packing: false

gradient_accumulation_steps: 4
micro_batch_size: 1
num_epochs: 3
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.00002

bf16: true
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false

resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_ratio: 0.1
evals_per_epoch: 1
saves_per_epoch: 1
weight_decay: 0.0
special_tokens: