🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning. ⚙️ Running in WANDB offline mode 🦥 Unsloth Zoo will now patch everything to make training faster! ==((====))== Unsloth 2026.9.2: Fast Olmo3 patching. Transformers: 4.57.6. \\ /| NVIDIA H100 NVL. Num GPUs = 1. Max memory: 93.085 GB. Platform: Linux. O^O/ \_/ \ Torch: 2.11.0+cu128. CUDA: 9.0. CUDA Toolkit: 12.8. Triton: 3.6.0 \ / Bfloat16 = TRUE. FA [Xformers = 0.0.35. FA2 = False] "-____-" Free license: http://github.com/unslothai/unsloth Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored! Unsloth: Dropout = 0 is supported for fast patching. You are using dropout = 0.05. Unsloth will patch all other layers, except LoRA matrices, causing a performance hit. Unsloth: QLoRA and full finetuning all not selected. Switching to 16bit LoRA. Unsloth: Explicit target_modules are constrained by the finetune_(vision|language|attention|mlp) filters; adapters attach only where both select. Unsloth: Allowing gradients for `base_model.model.model.embed_tokens` since it's in `modules_to_save`. Unsloth: Allowing gradients for `base_model.model.lm_head` since it's in `modules_to_save`. Generating train split: 0 examples [00:00, ? examples/s] Generating train split: 5000 examples [00:00, 8091.85 examples/s] Generating train split: 5000 examples [00:00, 7864.64 examples/s] tokenizing (num_proc=8): 0%| | 0/23500 [00:00 15339 kept; by source: {'apigen': 4824, 'xlam': 5000, 'recovery': 3750, 'replay:urban_12738_cleaned': 1265, 'efficiency': 500} [manifest] git=fedc86c2f311602b2ebdb86e609383ceee4786c0 data_sha256={'train_olmo-urban_replay50.jsonl': '5efef55a9b7c', 'xlam5000_max1call.jsonl': 'a8aae6cf5c36'} wandb: Tracking run with wandb version 0.30.0 wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing. wandb: Run data is saved locally in /workspace/anchor/olmo-urban/wandb/wandb/offline-run-20261012_205325-hbpaapsf wandb: View this run in the terminal with `wandb leet` wandb: Detected [huggingface_hub.inference] in use. wandb: Use W&B Weave for improved LLM call tracing. Install Weave with `pip install weave` then add `import weave` to the top of your script. wandb: For more information, check out the docs at: https://weave-docs.wandb.ai ==((====))== Unsloth - 2x faster free finetuning | Num GPUs used = 1 \\ /| Num examples = 14,955 | Num Epochs = 1 | Total steps = 935 O^O/ \_/ \ Batch size per device = 2 | Gradient accumulation steps = 8 \ / Data Parallel GPUs = 1 | Total batch size (2 x 8 x 1) = 16 "-____-" Trainable parameters = 981,385,216 of 8,279,396,352 (11.85% trained) Unsloth: Setting lr = 2.00e-05 instead of 2.00e-04 for embed_tokens. Unsloth: Setting lr = 2.00e-05 instead of 2.00e-04 for lm_head. 0%| | 0/935 [00:00