Unable to load ORPO dataset in a *.json file

SicariusSicariiStuff commented 3 months ago

Please check that this issue hasn't been reported before.

[X] I searched previous Bug Reports didn't find any similar reports.

Expected Behavior

Local dataset to work the same as with loading a dataset from HF hub

Current behaviour

FileNotFoundError: Couldn't find a dataset script at

Steps to reproduce

If you have:

rl: orpo
orpo_alpha: 0.1
chat_template: chatml
datasets:
  - path: HF_username/Dataset_name
    type: chat_template.argilla
    chat_template: chatml

Replace it with the same file locally (parquet\json doesn't matter) And you'll get :

FileNotFoundError: Couldn't find a dataset script at

Config yaml

base_model: SicariusSicariiStuff/2B-ad
output_dir: /home/sicarius/test/
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
save_safetensors: true

num_epochs: 2
saves_per_epoch: 1
saves_per_epoch: 1
save_total_limit: 2

learning_rate: 4e-6
lora_r: 16
lora_alpha: 32

sequence_len: 1024

lora_target_modules:

rl: orpo
orpo_alpha: 0.1
chat_template: chatml
datasets:
  - path: /home/sicarius/test/orpo1.json
    type: chat_template.argilla
    chat_template: chatml

remove_unused_columns: false
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: false

val_set_size: 0.0

adapter: qlora
lora_dropout: 0
lora_target_linear: true
load_in_8bit: false
load_in_4bit: true
strict: false

gradient_accumulation_steps: 1
micro_batch_size: 1

#optimizer: adamw_torch
optimizer: adamw_bnb_8bit
lr_scheduler: cosine

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 0
#warmup_ratio: 0.1
evals_per_epoch:
eval_table_size:
eval_max_new_tokens: 128

debug:
#deepspeed: deepspeed_configs/zero3_bf16_cpuoffload_all.json
weight_decay: 0.0
fsdp:
fsdp_config:
lora_modules_to_save: [embed_tokens, lm_head]
special_tokens:
  eos_token: "<|im_end|>"
  pad_token: "<|end_of_text|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

Possible solution

Using a similar processing logic as in a loaded dataset from the hub

Which Operating Systems are you using?

[X] Linux
[ ] macOS
[ ] Windows

Python Version

3.10

axolotl branch-commit

latest release

Acknowledgements

[X] My issue title is concise, descriptive, and in title casing.
[X] I have searched the existing issues to make sure this bug has not been reported yet.
[X] I am using the latest version of axolotl.
[X] I have provided enough information for the maintainers to reproduce and diagnose the issue.

RishabhMaheshwary commented 2 months ago

I am also facing the same issue with the following trace:

Traceback (most recent call last):
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/mnt/rishabh/axolotl/src/axolotl/cli/preprocess.py", line 103, in <module>
    fire.Fire(do_cli)
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/fire/core.py", line 143, in Fire
    component_trace = _Fire(component, args, parsed_flag_args, context, name)
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/fire/core.py", line 477, in _Fire
    component, remaining_args = _CallAndUpdateTrace(
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/fire/core.py", line 693, in _CallAndUpdateTrace
    component = fn(*varargs, **kwargs)
  File "/mnt/rishabh/axolotl/src/axolotl/cli/preprocess.py", line 74, in do_cli
    load_rl_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
  File "/mnt/rishabh/axolotl/src/axolotl/cli/__init__.py", line 445, in load_rl_datasets
    train_dataset, eval_dataset = load_prepare_dpo_datasets(cfg)
  File "/mnt/rishabh/axolotl/src/axolotl/utils/data/rl.py", line 131, in load_prepare_dpo_datasets
    train_dataset = load_split(cfg.datasets, cfg)
  File "/mnt/rishabh/axolotl/src/axolotl/utils/data/rl.py", line 110, in load_split
    split_datasets[i] = map_dataset(
  File "/mnt/rishabh/axolotl/src/axolotl/utils/data/rl.py", line 67, in map_dataset
    data_set = data_set.map(
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 602, in wrapper
    out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 567, in wrapper
    out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3156, in map
    for rank, done, content in Dataset._map_single(**dataset_kwargs):
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3517, in _map_single
    example = apply_function_on_filtered_inputs(example, i, offset=offset)
  File "/mnt/rishabh/anaconda3/envs/axolotl1/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3416, in apply_function_on_filtered_inputs
    processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
  File "/mnt/rishabh/axolotl/src/axolotl/prompt_strategies/orpo/chat_template.py", line 272, in transform_fn
    [msg.model_dump() for msg in dataset_parser.get_prompt(sample).messages],
  File "/mnt/rishabh/axolotl/src/axolotl/prompt_strategies/orpo/chat_template.py", line 111, in get_prompt
    content=prompt["chosen"][i * 2 + 1]["content"],

SicariusSicariiStuff commented 1 month ago

any update?

bursteratom commented 1 week ago

Hi @SicariusSicariiStuff thank you for following up! Any chance you can provide us the dataset you are using so we can do deeper testing on it?

axolotl-ai-cloud / axolotl