config and model missing in Trainer

[jetaudio]

When using deepspeed backend, Error occur: Trainer init missing 2 positional config and model. Fixing in line 321 of ludwig/backend/base.py: add missing variable.
This is the code I used:

qlora_fine_tuning_config = yaml.safe_load(
"""
model_type: llm
base_model: alexsherstinsky/Mistral-7B-v0.1-sharded
input_features:

• name: instruction
  type: text
  output_features:
• name: output
  type: text
  prompt:
  template: >-
  Below is an instruction that describes a task, paired with an input
  that provides further context. Write a response that appropriately
  completes the request.
  Instruction: {instruction}
  Input: {input}
  Response:
  generation:
  temperature: 0.1
  max_new_tokens: 512
  adapter:
  type: qlora
  quantization:
  bits: 4
  preprocessing:
  global_max_sequence_length: 512
  split:
  type: random
  probabilities:
  • 0.9
  • 0.05
  • 0.05
    trainer:
    type: finetune
    epochs: 1
    batch_size: 1
    eval_batch_size: 2
    gradient_accumulation_steps: 16
    learning_rate: 0.0004
    learning_rate_scheduler:
    warmup_fraction: 0.03
    backend:
    type: deepspeed
    zero_optimization:
    stage: 3
    offload_optimizer:
    device: nvme
    nvme_path: ./offload
    pin_memory: true
    fast
    """
    )
    model = LudwigModel(config=qlora_fine_tuning_config, logging_level=logging.INFO,backend='deepspeed')
    results, _ , _ = model.train(dataset=df)`

[justinxzhao]

Hi @jetaudio,

Thanks for proposing a fix with #3750.

The PR looks fine to me, but it would be good to get a quick look from @tgaddair and @arnavgarg1 who worked on the original deepspeed integration.

[tgaddair]

Thanks for the PR, @jetaudio! Closing this issue as #3750 has been landed :)