Training Software Runner is broken

[btarifi10]

• Runner for training does not support vision models and has some bugs in the evaluation of the loss
• See errors below:

File "/home/bkt123/dev/advanced-deep-learning-systems/mase-2/machop/chop/actions/search/strategies/runners/software/__init__.py", line 11, in get_sw_runner
    return SW_RUNNERS[name](model_info, task, dataset_info, accelerator, config)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/bkt123/dev/advanced-deep-learning-systems/mase-2/machop/chop/actions/search/strategies/runners/software/base.py", line 22, in __init__
    self._post_init_setup()
  File "/home/bkt123/dev/advanced-deep-learning-systems/mase-2/machop/chop/actions/search/strategies/runners/software/train.py", line 60, in _post_init_setup
    self._setup_metric()
  File "/home/bkt123/dev/advanced-deep-learning-systems/mase-2/machop/chop/actions/search/strategies/runners/software/train.py", line 82, in _setup_metric
    raise ValueError(f"model type {self.model_info} is not supported.")
ValueError: model type MaseModelInfo(name='jsc-tiny', model_source=<ModelSource.PHYSICAL: 'physical'>, task_type=<ModelTaskType.PHYSICAL: 'physical'>, image_classification=False, physical_data_point_classification=True, sequence_classification=False, seq2seqLM=False, causal_LM=False, is_quantized=False, is_lora=False, is_sparse=False, is_fx_traceable=True) is not supported.

---

To reproduce:

• Add the following runner in any search config toml, for example configs/examples/jsc_toy_by_type.toml:

  [search.strategy.sw_runner.basic_train]
  name = "accuracy"
  data_loader = "train_dataloader"
  num_samples = 1000000
  max_epochs = 2
  lr_scheduler = "linear"
  optimizer = "adam"
  learning_rate = 1e-4
  num_warmup_steps = 0

• Run the search:

  ./ch search --config configs/examples/jsc_toy_by_type.toml