# TRL importation error on Kaggle **URL:** \ **Category:** Beginners\ **Created:** [December 23, 2023, 1:49pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817 "2023-12-23T13:49:04Z")\ **Posts on this page:** 11\ **Page:** 1
**Author:** ![DrPalmiere](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@DrPalmiere](https://discuss.huggingface.co/u/DrPalmiere)\ **Post date:** [December 23, 2023, 1:49pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/1 "2023-12-23T13:49:04Z")
Hello. Please does anyone know if there is an issue with TRL library or a way to fix it? I am trying to fine tune Mistral 7B inside Kaggle, I already did it once with the same notebook, but since yesterday I am receiving an error while importing TRL. I already restarted the kernel and did factory reset, but the error still happens. from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig,HfArgumentParser,TrainingArguments,pipeline, logging from peft import LoraConfig, PeftModel, prepare\_model\_for\_kbit\_training, get\_peft\_model import os,torch, wandb from datasets import load\_dataset from trl import SFTTrainer This is the error: File /opt/conda/lib/python3.10/site-packages/trl/trainer/ **init**.py:44 42 from .ppo\_trainer import PPOTrainer 43 from .reward\_trainer import RewardTrainer, compute\_accuracy —\> 44 from .sft\_trainer import SFTTrainer 45 from .training\_configs import RewardConfig File /opt/conda/lib/python3.10/site-packages/trl/trainer/sft\_trainer.py:23 21 import torch.nn as nn 22 from datasets import Dataset —\> 23 from datasets.arrow\_writer import SchemaInferenceError 24 from datasets.builder import DatasetGenerationError 25 from transformers import ( 26 AutoModelForCausalLM, 27 AutoTokenizer, (…) 33 TrainingArguments, 34 ) ImportError: cannot import name ‘SchemaInferenceError’ from ‘datasets.arrow\_writer’ (/opt/conda/lib/python3.10/site-packages/datasets/arrow\_writer.py) ---
**Author:** ![alexgpetrov](https://avatars.discourse-cdn.com/v4/letter/a/bc8723/32.png) [@alexgpetrov](https://discuss.huggingface.co/u/alexgpetrov)\ **Post date:** [December 24, 2023, 1:53pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/2 "2023-12-24T13:53:31Z")
+1 have the sane issue in Kaggle for Mistral-7B fine tuning ---
**Author:** ![alexgpetrov](https://avatars.discourse-cdn.com/v4/letter/a/bc8723/32.png) [@alexgpetrov](https://discuss.huggingface.co/u/alexgpetrov)\ **Post date:** [December 25, 2023, 12:12pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/3 "2023-12-25T12:12:58Z")
@DrPalmiere I fixed it in Kaggle notebook installing latest version of datasets package. Seems, that by default older version of datasets package is used without SchemaInferenceError definition. Use following %pip install datasets==2.16.0 ---
**Author:** ![DrPalmiere](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@DrPalmiere](https://discuss.huggingface.co/u/DrPalmiere)\ **Post date:** [December 26, 2023, 9:26pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/4 "2023-12-26T21:26:40Z")
Thanks, I am using this notebook too (Mistral-7B fine tuning). I tried to update the datasets library but I got the same error. I fix the problem by creating a fresh notebook and copying all the code from the old notebook to the new one, and worked. ---
**Author:** ![mar-yam1497](https://avatars.discourse-cdn.com/v4/letter/m/90ced4/32.png) [@mar-yam1497](https://discuss.huggingface.co/u/mar-yam1497)\ **Post date:** [December 30, 2023, 7:46pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/5 "2023-12-30T19:46:51Z")
I tried that but I’m still getting the same error: ImportError: cannot import name ‘SchemaInferenceError’ from ‘datasets.arrow\_writer’ (/opt/conda/lib/python3.10/site-packages/datasets/arrow\_writer.py) ---
**Author:** ![DrPalmiere](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@DrPalmiere](https://discuss.huggingface.co/u/DrPalmiere)\ **Post date:** [December 30, 2023, 11:56pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/6 "2023-12-30T23:56:58Z")
Did you try to create an empty notebook and copy cell by cell from the seed notebook? I did it and it is working now? ---
**Author:** ![kaamraan](https://avatars.discourse-cdn.com/v4/letter/k/b487fb/32.png) [@kaamraan](https://discuss.huggingface.co/u/kaamraan)\ **Post date:** [January 2, 2024, 10:35pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/7 "2024-01-02T22:35:52Z")
I have done that, I also upgraded the datasets to 2.16.1, But the problem remained. ---
**Author:** ![vuluu](https://avatars.discourse-cdn.com/v4/letter/v/f6c823/32.png) [@vuluu](https://discuss.huggingface.co/u/vuluu)\ **Post date:** [January 10, 2024, 3:32pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/8 "2024-01-10T15:32:12Z")
You guys should try this, it works for me The tip is from: > > > \### Describe the bug > > While importing from packages getting the error > Code: > … > \`\`\` > import os > import torch > from datasets import load\_dataset, Dataset > from transformers import ( > AutoModelForCausalLM, > AutoTokenizer, > BitsAndBytesConfig, > HfArgumentParser, > TrainingArguments, > pipeline, > logging > ) > from peft import LoraConfig, PeftModel > from trl import SFTTrainer > from huggingface\_hub import login > import pandas as pd > \`\`\` > > Error: > \`\`\`\` > \--------------------------------------------------------------------------- > ImportError Traceback (most recent call last) > Cell In\[5\], line 14 > 4 from transformers import ( > 5 AutoModelForCausalLM, > 6 AutoTokenizer, > (...) > 11 logging > 12 ) > 13 from peft import LoraConfig, PeftModel > \---\> 14 from trl import SFTTrainer > 15 from huggingface\_hub import login > 16 import pandas as pd > > File /opt/conda/lib/python3.10/site-packages/trl/\_\_init\_\_.py:21 > 8 from .import\_utils import ( > 9 is\_diffusers\_available, > 10 is\_npu\_available, > (...) > 13 is\_xpu\_available, > 14 ) > 15 from .models import ( > 16 AutoModelForCausalLMWithValueHead, > 17 AutoModelForSeq2SeqLMWithValueHead, > 18 PreTrainedModelWrapper, > 19 create\_reference\_model, > 20 ) > \---\> 21 from .trainer import ( > 22 DataCollatorForCompletionOnlyLM, > 23 DPOTrainer, > 24 IterativeSFTTrainer, > 25 PPOConfig, > 26 PPOTrainer, > 27 RewardConfig, > 28 RewardTrainer, > 29 SFTTrainer, > 30 ) > 33 if is\_diffusers\_available(): > 34 from .models import ( > 35 DDPOPipelineOutput, > 36 DDPOSchedulerOutput, > 37 DDPOStableDiffusionPipeline, > 38 DefaultDDPOStableDiffusionPipeline, > 39 ) > > File /opt/conda/lib/python3.10/site-packages/trl/trainer/\_\_init\_\_.py:44 > 42 from .ppo\_trainer import PPOTrainer > 43 from .reward\_trainer import RewardTrainer, compute\_accuracy > \---\> 44 from .sft\_trainer import SFTTrainer > 45 from .training\_configs import RewardConfig > > File /opt/conda/lib/python3.10/site-packages/trl/trainer/sft\_trainer.py:23 > 21 import torch.nn as nn > 22 from datasets import Dataset > \---\> 23 from datasets.arrow\_writer import SchemaInferenceError > 24 from datasets.builder import DatasetGenerationError > 25 from transformers import ( > 26 AutoModelForCausalLM, > 27 AutoTokenizer, > (...) > 33 TrainingArguments, > 34 ) > > ImportError: cannot import name 'SchemaInferenceError' from 'datasets.arrow\_writer' (/opt/conda/lib/python3.10/site-packages/datasets/arrow\_writer.py > > \`\`\`\` > > transformers version: 4.36.2 > python version: 3.10.12 > datasets version: 2.16.1 > > \### Steps to reproduce the bug > > 1. Install packages > \`\`\` > !pip install -U datasets trl accelerate peft bitsandbytes transformers trl huggingface\_hub > \`\`\` > > 2. import packages > \`\`\` > import os > import torch > from datasets import load\_dataset, Dataset > from transformers import ( > AutoModelForCausalLM, > AutoTokenizer, > BitsAndBytesConfig, > HfArgumentParser, > TrainingArguments, > pipeline, > logging > ) > from peft import LoraConfig, PeftModel > from trl import SFTTrainer > from huggingface\_hub import login > import pandas as pd > \`\`\` > > \### Expected behavior > > No error while importing > > \### Environment info > > \- \`datasets\` version: 2.16.0 > \- Platform: Linux-5.15.133+-x86\_64-with-glibc2.35 > \- Python version: 3.10.12 > \- \`huggingface\_hub\` version: 0.20.1 > \- PyArrow version: 11.0.0 > \- Pandas version: 2.1.4 > \- \`fsspec\` version: 2023.10.0 "In kaggle I used: - `%pip install -U datasets` and then restarted runtime and then everything works fine." ---
**Author:** ![DrPalmiere](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@DrPalmiere](https://discuss.huggingface.co/u/DrPalmiere)\ **Post date:** [January 10, 2024, 9:17pm UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/9 "2024-01-10T21:17:07Z")
I tried it, but didn’ t work. I only fixed it by starting a fresh notebook and doing copy and paste from old notebook to the new one. ---
**Author:** ![RadioactiveSquirrel](https://avatars.discourse-cdn.com/v4/letter/r/f07891/32.png) [@RadioactiveSquirrel](https://discuss.huggingface.co/u/RadioactiveSquirrel)\ **Post date:** [January 16, 2024, 10:56am UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/10 "2024-01-16T10:56:22Z")
This worked for me 🙂 ---
**Author:** ![MH0386](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/mh0386/32/52392_2.png) [@MH0386](https://discuss.huggingface.co/u/MH0386)\ **Post date:** [March 25, 2024, 2:21am UTC](https://discuss.huggingface.co/t/trl-importation-error-on-kaggle/66817/11 "2024-03-25T02:21:22Z")
`!pip install datasets==2.16.0 -q` worked for me