--- library_name: transformers license: gemma base_model: google/gemma-3-27b-it tags: - generated_from_trainer datasets: - shisa-ai/shisa-v2-best-of-n-athenev2-tulu70b-llama33-only-no-sysprompt - shisa-ai/shisa-v2-roleplaying-sft - shisa-ai/translation_set_april_6 - shisa-ai/rewild-set-deepseek-subset - shisa-ai/magpie-ultra-set - shisa-ai/magpie-advanced-questions-set - shisa-ai/japan-magpie-set - shisa-ai/shisa-v2-instruction-following-sft model-index: - name: outputs/ablation-196-finalsft2-shisa-v2-gemma3-27b results: [] --- [Built with Axolotl](https://github.com/axolotl-ai-cloud/axolotl)
See axolotl config axolotl version: `0.8.0.dev0` ```yaml # train w/ shisa-ai/shisa-v1-athenev2-reannotated-filtered base_model: google/gemma-3-27b-it load_in_8bit: false load_in_4bit: false strict: false datasets: - path: shisa-ai/shisa-v2-best-of-n-athenev2-tulu70b-llama33-only-no-sysprompt type: chat_template field_messages: conversations message_field_role: from message_field_content: value - path: shisa-ai/shisa-v2-roleplaying-sft type: chat_template field_messages: conversations message_property_mappings: role: role content: content roles: system: - system assistant: - gpt - model - assistant user: - human - user roles_to_train: ["assistant"] - path: shisa-ai/translation_set_april_6 split: train[:25%] type: chat_template field_messages: conversations message_property_mappings: role: role content: content roles: system: - system assistant: - gpt - model - assistant user: - human - user roles_to_train: ["assistant"] - path: shisa-ai/rewild-set-deepseek-subset split: train[:25%] type: chat_template field_messages: conversations message_property_mappings: role: role content: content roles: system: - system assistant: - gpt - model - assistant user: - human - user roles_to_train: ["assistant"] - path: shisa-ai/magpie-ultra-set split: train[:8%] type: chat_template field_messages: conversations message_property_mappings: role: role content: content roles: system: - system assistant: - gpt - model - assistant user: - human - user roles_to_train: ["assistant"] - path: shisa-ai/magpie-advanced-questions-set split: train[:8%] type: chat_template field_messages: conversations message_property_mappings: role: role content: content roles: system: - system assistant: - gpt - model - assistant user: - human - user roles_to_train: ["assistant"] - path: shisa-ai/japan-magpie-set split: train type: chat_template field_messages: conversations message_property_mappings: role: role content: content roles: system: - system assistant: - gpt - model - assistant user: - human - user roles_to_train: ["assistant"] - path: shisa-ai/shisa-v2-instruction-following-sft split: train[:50%] type: chat_template field_messages: conversations message_property_mappings: role: role content: content roles: system: - system assistant: - gpt - model - assistant user: - human - user roles_to_train: ["assistant"] dataset_prepared_path: last_run_prepared val_set_size: 0.05 output_dir: ./outputs/ablation-196-finalsft2-shisa-v2-gemma3-27b sequence_len: 8192 sample_packing: true pad_to_sequence_len: true # marginal difference neftune_noise_alpha: 5 use_wandb: true wandb_project: shisa-v2 wandb_entity: augmxnt wandb_name: ablation-196-finalsft2-shisa-v2-gemma3-27b gradient_accumulation_steps: 2 micro_batch_size: 1 num_epochs: 3 optimizer: paged_adamw_8bit lr_scheduler: linear learning_rate: 5.4e-6 train_on_inputs: false group_by_length: false bf16: auto fp16: tf32: false gradient_checkpointing: true gradient_checkpointing_kwargs: use_reentrant: false early_stopping_patience: resume_from_checkpoint: logging_steps: 1 xformers_attention: flash_attention: true warmup_steps: 100 evals_per_epoch: 2 eval_table_size: saves_per_epoch: 0 save_total_limit: 1 # Only store a single checkpoint debug: deepspeed: zero3_bf16.json weight_decay: 1e-4 fsdp: fsdp_config: special_tokens: ```

# outputs/ablation-196-finalsft2-shisa-v2-gemma3-27b This model is a fine-tuned version of [google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it) on the shisa-ai/shisa-v2-best-of-n-athenev2-tulu70b-llama33-only-no-sysprompt, the shisa-ai/shisa-v2-roleplaying-sft, the shisa-ai/translation_set_april_6, the shisa-ai/rewild-set-deepseek-subset, the shisa-ai/magpie-ultra-set, the shisa-ai/magpie-advanced-questions-set, the shisa-ai/japan-magpie-set and the shisa-ai/shisa-v2-instruction-following-sft datasets. It achieves the following results on the evaluation set: - Loss: 0.5417 ## 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: 5.4e-06 - train_batch_size: 1 - eval_batch_size: 1 - seed: 42 - distributed_type: multi-GPU - num_devices: 64 - gradient_accumulation_steps: 2 - total_train_batch_size: 128 - total_eval_batch_size: 64 - optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments - lr_scheduler_type: linear - lr_scheduler_warmup_steps: 100 - num_epochs: 3.0 ### Training results | Training Loss | Epoch | Step | Validation Loss | |:-------------:|:------:|:----:|:---------------:| | 11.3287 | 0.0027 | 1 | 5.4644 | | 1.1409 | 0.4993 | 182 | 0.5644 | | 1.0588 | 0.9986 | 364 | 0.5344 | | 0.9207 | 1.4966 | 546 | 0.5322 | | 0.8979 | 1.9959 | 728 | 0.5245 | | 0.7673 | 2.4938 | 910 | 0.5432 | | 0.7521 | 2.9931 | 1092 | 0.5417 | ### Framework versions - Transformers 4.50.0 - Pytorch 2.6.0+cu124 - Datasets 3.4.1 - Tokenizers 0.21.1