[2026-07-18 18:16:26,193] [WARNING] [axolotl.utils.config.normalize_config:282] [PID:18257] Gemma4 requires use_reentrant=False for gradient checkpointing in distributed training. Setting use_reentrant=False. [2026-07-18 18:16:26,194] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:18257] baseline 0.000GB () [2026-07-18 18:16:26,195] [INFO] [axolotl.cli.config.load_cfg:333] [PID:18257] config: { "activation_offloading": true, "adapter": "qlora", "attn_implementation": "sdpa", "attn_needs_dtype_cast": false, "attn_supports_packing": false, "attn_uses_flash_lib": false, "axolotl_config_path": "DeltaDPO.yaml", "base_model": "google/gemma-4-31B", "base_model_config": "google/gemma-4-31B", "batch_size": 16, "bf16": true, "bnb_config_kwargs": { "bnb_4bit_use_double_quant": true }, "capabilities": { "bf16": true, "compute_capability": "sm_80", "fp8": false, "n_gpu": 1, "n_node": 1, "tf32": true }, "chat_template": "gemma4", "context_parallel_size": 1, "cut_cross_entropy": true, "dataloader_num_workers": 1, "dataloader_pin_memory": true, "dataloader_prefetch_factor": 256, "dataset_num_proc": 31, "datasets": [ { "path": "ConicCat/Charcards_Delta_Qwen3_5", "split": "train", "type": "chat_template.default" }, { "path": "ConicCat/Lamp_P_Preference", "split": "train", "type": "chat_template.default" } ], "ddp": false, "device": "cuda:0", "dion_rank_fraction": 1.0, "dion_rank_multiple_of": 1, "dpo_use_weighting": true, "eaft_alpha": 1.0, "eaft_k": 20, "env_capabilities": { "torch_version": "2.10.0" }, "eot_tokens": [ "" ], "eval_batch_size": 1, "eval_causal_lm_metrics": [ "sacrebleu", "comet", "ter", "chrf" ], "eval_max_new_tokens": 128, "eval_table_size": 0, "experimental_skip_move_to_device": true, "fp16": false, "gemma4_hybrid_attn_impl": false, "generate_samples": false, "generation_do_sample": true, "generation_max_new_tokens": 50, "generation_prompt_ratio": 0.5, "generation_temperature": 0.7, "gradient_accumulation_steps": 16, "gradient_checkpointing": true, "gradient_checkpointing_kwargs": { "use_reentrant": false }, "include_tkps": true, "is_multimodal": true, "layer_offloading": false, "learning_rate": 1.66e-06, "liger_glu_activation": true, "liger_layer_norm": true, "liger_rms_norm": true, "liger_rms_norm_gated": true, "liger_rope": true, "lisa_layers_attribute": "model.layers", "load_best_model_at_end": false, "load_in_4bit": true, "load_in_8bit": false, "local_rank": 0, "logging_steps": 1, "lora_alpha": 128, "lora_dropout": 0.0, "lora_r": 64, "lora_target_modules": "model.language_model.layers.[\\d]+.(_checkpoint_wrapped_module.)?(mlp|self_attn).(up|down|gate|q|k|v|o)_proj", "loraplus_lr_embedding": 1e-06, "loraplus_lr_ratio": 16.0, "lr_scheduler": "cosine", "max_grad_norm": 1.0, "mean_resizing_embeddings": false, "merge_method": "memory_efficient", "micro_batch_size": 1, "model_config_type": "gemma4", "model_config_type_text": "gemma4_text", "num_epochs": 1.0, "num_generation_samples": 3, "optimizer": "adamw_torch_8bit", "otel_metrics_host": "localhost", "otel_metrics_port": 8000, "output_dir": "./Gemma-DPO", "plugins": [ "axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin", "axolotl.integrations.liger.LigerPlugin" ], "pretrain_multipack_attn": true, "processor_config": "google/gemma-4-31B", "profiler_steps_start": 0, "qgalore_cos_threshold": 0.4, "qgalore_gamma_proj": 2, "qgalore_proj_bits": 4, "qgalore_proj_group_size": 256, "qgalore_proj_quant": true, "qgalore_proj_type": "std", "qgalore_queue_size": 5, "qgalore_rank": 256, "qgalore_scale": 0.25, "qgalore_update_proj_gap": 200, "qlora_sharded_model_loading": false, "quantize_moe_experts": false, "ray_num_workers": 1, "relora_prune_method": "magnitude", "resources_per_worker": { "GPU": 1 }, "rl": "dpo", "rl_beta": 0.01, "sample_packing": false, "sample_packing_bin_size": 200, "sample_packing_group_size": 100000, "save_only_model": false, "save_safetensors": true, "save_strategy": "epoch", "seed": 42, "sequence_len": 4096, "shuffle_before_merging_datasets": false, "shuffle_merged_datasets": true, "skip_prepare_dataset": false, "streaming_multipack_buffer_size": 10000, "strict": false, "tensor_parallel_size": 1, "tf32": true, "tiled_mlp_use_original_mlp": true, "tokenizer_config": "google/gemma-4-31B", "tokenizer_save_jinja_files": true, "torch_dtype": "torch.bfloat16", "train_on_inputs": false, "trl": { "async_prefetch": false, "log_completions": false, "mask_truncated_completions": false, "ref_model_mixup_alpha": 0.9, "ref_model_sync_steps": 64, "replay_buffer_size": 0, "replay_recompute_logps": true, "reroll_max_groups": 1, "reroll_start_fraction": 1.0, "reward_num_workers": 1, "scale_rewards": true, "skip_zero_advantage_batches": true, "sync_ref_model": false, "use_data_producer": false, "use_vllm": false, "vllm_lora_sync": false, "vllm_server_host": "0.0.0.0", "vllm_server_port": 8000 }, "use_otel_metrics": false, "use_ray": false, "use_tensorboard": true, "val_set_size": 0.0, "vllm": { "device": "auto", "dtype": "auto", "gpu_memory_utilization": 0.9, "host": "0.0.0.0", "port": 8000 }, "warmup_ratio": 0.05, "weight_decay": 0.0, "world_size": 1 } [2026-07-18 18:16:26,198] [WARNING] [axolotl.cli.checks.check_user_token:48] [PID:18257] Error verifying HuggingFace token. Remember to log in using `hf auth login` and get your access token from https://huggingface.co/settings/tokens if you want to use gated models or datasets. [2026-07-18 18:16:29,183] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:18257] EOS: 1 / [2026-07-18 18:16:29,183] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:18257] BOS: 2 / [2026-07-18 18:16:29,183] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:18257] PAD: 0 / [2026-07-18 18:16:29,183] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:18257] UNK: 3 / [2026-07-18 18:16:29,185] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:482] [PID:18257] Unable to find prepared dataset in last_run_prepared/e4e860592df519f55193e4bc6fa2dae0 Downloading (incomplete total...): 0.00B [00:00, ?B/s] Fetching 0 files: 0it [00:00, ?it/s] Fetching 0 files: 0it [00:00, ?it/s] Download complete: : 0.00B [00:00, ?B/s] Download complete: : 0.00B [00:00, ?B/s] Downloading (incomplete total...): 0.00B [00:00, ?B/s] Fetching 0 files: 0it [00:00, ?it/s] Fetching 0 files: 0it [00:00, ?it/s] Download complete: : 0.00B [00:00, ?B/s] Download complete: : 0.00B [00:00, ?B/s] [2026-07-18 18:16:33,155] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:18257] EOS: 1 / [2026-07-18 18:16:33,155] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:18257] BOS: 2 / [2026-07-18 18:16:33,155] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:18257] PAD: 0 / [2026-07-18 18:16:33,155] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:18257] UNK: 3 / [2026-07-18 18:16:41,068] [WARNING] [axolotl.utils.data.rl._load_split:434] [PID:18257] Dropped 8 long samples from dataset index 0 [2026-07-18 18:16:48,901] [INFO] [axolotl.utils.data.shared.merge_datasets:566] [PID:18257] Merging datasets... [2026-07-18 18:16:48,911] [DEBUG] [axolotl.utils.data.shared.merge_datasets:570] [PID:18257] Shuffling merged datasets... Saving the dataset (0/14 shards): 0%| | 0/3684 [00:00 [2026-07-18 18:17:06,220] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:18257] BOS: 2 / [2026-07-18 18:17:06,220] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:18257] PAD: 0 / [2026-07-18 18:17:06,220] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:18257] UNK: 3 / [2026-07-18 18:17:12,089] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:18257] Loading model [2026-07-18 18:17:12,141] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:18257] Patched OptimState8bit for torch.compile compatibility [2026-07-18 18:17:12,141] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:18257] Patched OptimState4bit for torch.compile compatibility [2026-07-18 18:17:12,141] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:18257] Patched OptimStateFp8 for torch.compile compatibility [2026-07-18 18:17:12,147] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:18257] Patched Trainer.evaluation_loop with nanmean loss calculation [2026-07-18 18:17:12,148] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:18257] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation [2026-07-18 18:17:12,257] [INFO] [axolotl.monkeypatch.models.gemma4.fused_attn.patch_gemma4_fused_attn:207] [PID:18257] Patched Gemma4TextAttention.forward with fused RMSNorm+RoPE Triton kernels [2026-07-18 18:17:12,257] [INFO] [axolotl.monkeypatch.models.gemma4.fused_attn.patch_gemma4_fused_attn:211] [PID:18257] Installed Gemma4 shared_kv_states side channel (PR #3611) [2026-07-18 18:17:12,263] [INFO] [axolotl.loaders.patch_manager._apply_trl_trainer_utils_patches:1022] [PID:18257] Patched trl.trainer.utils with Triton selective_log_softmax and entropy_from_logits [2026-07-18 18:17:12,291] [INFO] [axolotl.integrations.cut_cross_entropy.pre_model_load:94] [PID:18257] Applying Cut Cross Entropy to model type: gemma4 [2026-07-18 18:17:13,126] [WARNING] [axolotl.integrations.liger.plugin.pre_model_load:277] [PID:18257] Liger RoPE is not compatible with Gemma4 (separate q/k application). Skipping. [2026-07-18 18:17:13,126] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:290] [PID:18257] Applied Liger kernels for gemma4: rms_norm=True, glu=True, rope=False (incompatible), layer_norm=True Loading weights: 0%| | 0/1188 [00:00