finetune-gemma3-270m-it-nlp2sql

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

axolotl version: 0.13.0.dev0

base_model: google/gemma-3-270m-it
model_type: GemmaForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

datasets:
  - path: Humenuik/train.jsonl
    type: alpaca
    conversation: "alpaca"

dataset_prepared_path:
  /workspace/output/prepared

output_dir: /workspace/output/finetune-gemma3-270m-it

lora_r: 32 # Set the LoRA rank, influencing the capacity of the adapter layers.
lora_alpha: 16 # Adjust the LoRA alpha, a scaling factor for LoRA updates.
lora_dropout: 0.05 # Apply a dropout rate to LoRA layers for regularization.
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj # Specify the modules within the model where LoRA adapters will be applied (e.g., attention projections, MLP layers).

gradient_checkpointing: false
micro_batch_size: 8 # Define the size of each micro-batch processed by the GPU.
num_epochs: 2 # Set the number of training epochs.
learning_rate: 0.0002 # Specify the learning rate for the optimizer.
lr_scheduler: cosine # Choose a learning rate scheduler (e.g., cosine).
warmup_steps: 10 # Configure warmup steps for the learning rate scheduler.
batch_size: 24
gradient_checkpointing: true # Enable gradient checkpointing to reduce memory consumption during training.

</details><br>

workspace/output/finetune-gemma3-270m-it

This model is a fine-tuned version of google/gemma-3-270m-it on the Humenuik/train.jsonl dataset.

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: 8
• eval_batch_size: 8
• 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: cosine
• lr_scheduler_warmup_steps: 10
• training_steps: 4528

Training results

Framework versions

• Transformers 4.55.2
• Pytorch 2.7.1+cu126
• Datasets 4.0.0
• Tokenizers 0.21.4