TypeError: SFTTrainer.__init__() got an unexpected keyword argument 'max_seq_length'

[dwgb93]

I am unable to make distill_logits.py run.

I'm pretty sure this is due to a recent breaking update of TRL, but I'm not quite sure how to fix it.

LogitsTrainer doesn't seem to like the max_seq_length or dataset_text_field arguments.

Looking at the documentation here, I see that I can do something like
my_args = SFTConfig(max_seq_length=config["tokenizer"]["max_length"], dataset_text_field="text", output_dir="./results")

But I'm unclear as to how to include that. Setting args=my_args gets rid of the transformers TrainingArguments specified above.

Further, while attempting to fix these issues, I got another error: TypeError: LogitsTrainer.compute_loss() got an unexpected keyword argument 'num_items_in_batch'

This appears to be related to a different breaking update of transformers.

It's also worth noting that it keeps warning me tokenizer is deprecated and removed starting from version 0.16.0 for LogitsTrainer.__init__. Use processing_class instead.

Any chance you could help here?

[LRHstudy]

trl==0.9.6 accelerate==1.0.1 can run

[dwgb93]

@LRHstudy That didn't seem to work for me.
I'm still getting TypeError: LogitsTrainer.compute_loss() got an unexpected keyword argument 'num_items_in_batch'

I'm including my full pip list for reference, if anybody can spot something strange:
Package Version

---

accelerate 1.0.1
aiohappyeyeballs 2.4.4
aiohttp 3.11.12
aiosignal 1.3.2
annotated-types 0.7.0
async-timeout 5.0.1
attrs 25.1.0
certifi 2025.1.31
charset-normalizer 3.4.1
datasets 3.2.0
deepspeed 0.16.3
dill 0.3.8
distlib 0.3.9
docstring_parser 0.16
einops 0.8.0
filelock 3.16.1
flash_attn 2.7.4.post1
frozenlist 1.5.0
fsspec 2024.9.0
hf_transfer 0.1.9
hjson 3.1.0
huggingface-hub 0.28.1
idna 3.10
Jinja2 3.1.5
markdown-it-py 3.0.0
MarkupSafe 3.0.2
mdurl 0.1.2
meson 1.7.0
mpmath 1.3.0
msgpack 1.1.0
multidict 6.1.0
multiprocess 0.70.16
networkx 3.4.2
ninja 1.11.1.3
numpy 1.26.4
nvidia-cublas-cu12 12.4.5.8
nvidia-cuda-cupti-cu12 12.4.127
nvidia-cuda-nvrtc-cu12 12.4.127
nvidia-cuda-runtime-cu12 12.4.127
nvidia-cudnn-cu12 9.1.0.70
nvidia-cufft-cu12 11.2.1.3
nvidia-curand-cu12 10.3.5.147
nvidia-cusolver-cu12 11.6.1.9
nvidia-cusparse-cu12 12.3.1.170
nvidia-cusparselt-cu12 0.6.2
nvidia-ml-py 12.570.86
nvidia-nccl-cu12 2.21.5
nvidia-nvjitlink-cu12 12.4.127
nvidia-nvtx-cu12 12.4.127
packaging 24.2
pandas 2.2.3
pip 25.0
platformdirs 4.3.6
propcache 0.2.1
psutil 6.1.1
py-cpuinfo 9.0.0
pyarrow 19.0.0
pydantic 2.10.6
pydantic_core 2.27.2
Pygments 2.19.1
python-dateutil 2.9.0.post0
pytz 2025.1
PyYAML 6.0.2
regex 2024.11.6
requests 2.32.3
rich 13.9.4
safetensors 0.5.2
setuptools 75.8.0
shtab 1.7.1
six 1.17.0
sympy 1.13.1
tokenizers 0.21.0
torch 2.6.0
tqdm 4.67.1
transformers 4.48.2
triton 3.2.0
trl 0.9.6
typeguard 4.4.1
typing_extensions 4.12.2
tyro 0.9.14
tzdata 2025.1
urllib3 2.3.0
virtualenv 20.28.0
wheel 0.45.1
xxhash 3.5.0
yarl 1.18.3

Thanks!

[LRHstudy]

def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=None): # add num_items_in_batch=None

[nouranali]

Same error happens to me on latest version

[liuxi0099]

trl==0.9.6 accelerate==1.0.1 can run

trl==0.11.3 can run as well, but 0.18.2 cannot