The following values were not passed to `accelerate launch` and had defaults used instead:
	`--num_processes` was set to a value of `1`
	`--num_machines` was set to a value of `1`
	`--mixed_precision` was set to a value of `'no'`
	`--dynamo_backend` was set to a value of `'no'`
To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.
Skipping import of cpp extensions due to incompatible torch version. Please upgrade to torch >= 2.11.0 (found 2.10.0+cu128).
__init__.py         :176  2026-09-06 09:54:36,304 Creating authenticated Supabase client
__init__.py         :68   2026-09-06 09:54:36,304 Exchanging API token for JWT
__init__.py         :84   2026-09-06 09:54:36,603 Successfully exchanged API token for JWT
__init__.py         :187  2026-09-06 09:54:37,095 Initialized OpenWeights client for organization: niels.warncke@gmail.com (ID: 8e10fb40-9487-40ae-9b48-f210c6f52a23)
/opt/venv/lib/python3.12/site-packages/unsloth/__init__.py:1543: UserWarning: WARNING: Unsloth should be imported before [transformers, peft] to ensure all optimizations are applied. Your code may run slower or encounter memory issues without these optimizations.

Please restructure your imports with 'import unsloth' at the top of your file.
  from ._gpu_init import *
Training config: {
    "r": 16,
    "beta": 0.1,
    "loss": "dpo",
    "meta": null,
    "seed": 17,
    "model": "Qwen/Qwen3.8-27B",
    "optim": "adamw_torch",
    "epochs": 2,
    "is_peft": true,
    "packing": false,
    "lora_bias": "none",
    "max_steps": 2,
    "test_file": null,
    "lora_alpha": 16,
    "output_dir": "./tmp",
    "save_steps": 16,
    "use_rslora": false,
    "load_in_4bit": false,
    "lora_dropout": 0.0,
    "warmup_steps": 0,
    "weight_decay": 0.0,
    "chat_template": "default",
    "job_id_suffix": null,
    "learning_rate": 0.0001,
    "logging_steps": 1,
    "training_file": "preference:file-050f0fe1c4d1",
    "max_seq_length": 256,
    "target_modules": [
        "q_proj",
        "k_proj",
        "v_proj",
        "o_proj",
        "gate_proj",
        "up_proj",
        "down_proj"
    ],
    "eval_batch_size": 4,
    "push_to_private": true,
    "lr_scheduler_type": "constant",
    "merge_before_push": false,
    "eval_every_n_steps": 5000,
    "finetuned_model_id": "longtermrisk/Qwen3.8-27B-ftjob-7ea5a20e14fc",
    "sampling_callbacks": null,
    "test_file_eval_steps": 16,
    "logp_callback_datasets": {},
    "test_file_eval_strategy": "steps",
    "train_on_responses_only": true,
    "gradient_accumulation_steps": 1,
    "per_device_train_batch_size": 1,
    "model_naming_extra_parameters": null
}
🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.
🦥 Unsloth Zoo will now patch everything to make training faster!
==((====))==  Unsloth 2026.9.2: Fast Qwen3_5 patching. Transformers: 5.5.0.
   \\   /|    NVIDIA H200. Num GPUs = 1. Max memory: 139.811 GB. Platform: Linux.
O^O/ \_/ \    Torch: 2.10.0+cu128. CUDA: 9.0. CUDA Toolkit: 12.8. Triton: 3.6.0
\        /    Bfloat16 = TRUE. FA [Xformers = 0.0.34. FA2 = False]
 "-____-"     Free license: http://github.com/unslothai/unsloth
Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!
/opt/venv/lib/python3.12/site-packages/huggingface_hub/constants.py:299: FutureWarning: The `HF_HUB_ENABLE_HF_TRANSFER` environment variable is deprecated as 'hf_transfer' is not used anymore. Please use `HF_XET_HIGH_PERFORMANCE` instead to enable high performance transfer with Xet. Visit https://huggingface.co/docs/huggingface_hub/package_reference/environment_variables#hfxethighperformance for more details.
  warnings.warn(
Skipping import of cpp extensions due to incompatible torch version. Please upgrade to torch >= 2.11.0 (found 2.10.0+cu128).
__init__.py         :176  2026-09-06 09:54:58,624 Creating authenticated Supabase client
__init__.py         :68   2026-09-06 09:54:58,624 Exchanging API token for JWT
__init__.py         :84   2026-09-06 09:54:58,906 Successfully exchanged API token for JWT
__init__.py         :187  2026-09-06 09:54:59,344 Initialized OpenWeights client for organization: niels.warncke@gmail.com (ID: 8e10fb40-9487-40ae-9b48-f210c6f52a23)
The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
Unsloth: QLoRA and full finetuning all not selected. Switching to 16bit LoRA.

Loading weights:   0%|          | 0/1184 [00:00<?, ?it/s]
Loading weights:  75%|███████▌  | 890/1184 [00:00<00:00, 8858.99it/s]
Loading weights: 100%|██████████| 1184/1184 [00:00<00:00, 9323.32it/s]
NOTE: Unwrapping Qwen3VLProcessor to get underlying tokenizer
Creating new LoRA adapter
Unsloth: Explicit target_modules are constrained by the finetune_(vision|language|attention|mlp) filters; adapters attach only where both select.
[VRAM-probe:unsloth/dpo] use_cache='<no-config>' _attn_implementation='sdpa' is_gradient_checkpointing=True
files.py            :207  2026-09-06 09:56:41,010 Downloading file: preference:file-050f0fe1c4d1
files.py            :210  2026-09-06 09:56:41,783 File downloaded: preference:file-050f0fe1c4d1 (33053 bytes)

Map:   0%|          | 0/128 [00:00<?, ? examples/s]
Map: 100%|██████████| 128/128 [00:00<00:00, 2908.61 examples/s]

Extracting prompt in train dataset (num_proc=5):   0%|          | 0/128 [00:00<?, ? examples/s]
Extracting prompt in train dataset (num_proc=5):  20%|██        | 26/128 [00:00<00:00, 189.05 examples/s]
Extracting prompt in train dataset (num_proc=5): 100%|██████████| 128/128 [00:00<00:00, 476.29 examples/s]

Applying chat template to train dataset (num_proc=5):   0%|          | 0/128 [00:00<?, ? examples/s]
Applying chat template to train dataset (num_proc=5):  20%|██        | 26/128 [00:00<00:03, 28.36 examples/s]
Applying chat template to train dataset (num_proc=5):  41%|████      | 52/128 [00:01<00:01, 55.71 examples/s]
Applying chat template to train dataset (num_proc=5):  61%|██████    | 78/128 [00:01<00:00, 82.21 examples/s]
Applying chat template to train dataset (num_proc=5):  80%|████████  | 103/128 [00:01<00:00, 108.00 examples/s]
Applying chat template to train dataset (num_proc=5): 100%|██████████| 128/128 [00:01<00:00, 133.09 examples/s]
Applying chat template to train dataset (num_proc=5): 100%|██████████| 128/128 [00:01<00:00, 80.66 examples/s]

Tokenizing train dataset:   0%|          | 0/128 [00:00<?, ? examples/s]
Tokenizing train dataset:   0%|          | 0/128 [00:00<?, ? examples/s]
Traceback (most recent call last):
  File "/tmp/tmpwhdbcbxf/training.py", line 232, in <module>
    main(sys.argv[1])
  File "/tmp/tmpwhdbcbxf/training.py", line 228, in main
    train(training_config)
  File "/tmp/tmpwhdbcbxf/training.py", line 137, in train
    trainer = dpo_train(
              ^^^^^^^^^^
  File "/tmp/tmpwhdbcbxf/dpo_ft.py", line 87, in dpo_train
    trainer = DPOTrainer(
              ^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/unsloth/trainer.py", line 860, in new_init
    original_init(self, *args, **kwargs)
  File "/tmp/tmpwhdbcbxf/unsloth_compiled_cache/UnslothDPOTrainer.py", line 3083, in __init__
    super().__init__(
  File "/tmp/tmpwhdbcbxf/unsloth_compiled_cache/UnslothDPOTrainer.py", line 1229, in __init__
    train_dataset = self._prepare_dataset(train_dataset, processing_class, args, "train")
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/tmpwhdbcbxf/unsloth_compiled_cache/UnslothDPOTrainer.py", line 1421, in _prepare_dataset
    dataset = dataset.map(
              ^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/unsloth_zoo/temporary_patches/misc.py", line 1266, in map
    return original_map(self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/datasets/arrow_dataset.py", line 562, in wrapper
    out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/datasets/arrow_dataset.py", line 3332, in map
    for rank, done, content in Dataset._map_single(**unprocessed_kwargs):
  File "/opt/venv/lib/python3.12/site-packages/datasets/arrow_dataset.py", line 3664, in _map_single
    for i, example in iter_outputs(shard_iterable):
  File "/opt/venv/lib/python3.12/site-packages/datasets/arrow_dataset.py", line 3638, in iter_outputs
    yield i, apply_function(example, i, offset=offset)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/datasets/arrow_dataset.py", line 3561, in apply_function
    processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/tmpwhdbcbxf/unsloth_compiled_cache/UnslothDPOTrainer.py", line 435, in dpo_trainer_vision_process_row
    processor, tokenizer = processing_class, processing_class.tokenizer
                                             ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/transformers/tokenization_utils_base.py", line 1293, in __getattr__
    raise AttributeError(f"{self.__class__.__name__} has no attribute {key}")
AttributeError: TokenizersBackend has no attribute tokenizer. Did you mean: '_tokenizer'?
Traceback (most recent call last):
  File "/opt/venv/bin/accelerate", line 6, in <module>
    sys.exit(main())
             ^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/accelerate/commands/accelerate_cli.py", line 50, in main
    args.func(args)
  File "/opt/venv/lib/python3.12/site-packages/accelerate/commands/launch.py", line 1407, in launch_command
    simple_launcher(args)
  File "/opt/venv/lib/python3.12/site-packages/accelerate/commands/launch.py", line 993, in simple_launcher
    raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd)
subprocess.CalledProcessError: Command '['/opt/venv/bin/python', 'training.py', 'ftjob-7ea5a20e14fc']' returned non-zero exit status 1.
