Error while importing "Trainer"

🤗Transformers
1

I am trying to run two commands one after another

  1. !pip install transformers torch gradio datasets --upgrade

  2. import torch

from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments

from datasets import load_dataset

import gradio as gr

While running the command 2, I am getting below error message.

The error message ModuleNotFoundError: Could not import module 'Trainer' suggests that a crucial component of the transformers library, specifically the Trainer class, cannot be found or loaded correctly. This usually points to a problem with the installation of the transformers package itself. Even though you ran pip install --upgrade, sometimes a full force-reinstall is necessary to fix these kinds of persistent import errors, especially if there were any corrupted files or conflicting dependencies from previous installations. The plan to force reinstall the libraries should address this.

I have followed above guideline to force-reinstall the package, (!pip install transformers torch gradio datasets --upgrade –force-reinstall) and still after restarting the runtime, issue doesn’t resolve. Any help much appreciated.

1 Like
2

If you’re using an older Python version (3.9), you might run into that kind of error…

Likely root cause in your specific setup

In most environments, Trainer is not “missing”. The error usually happens because importing Trainer triggers a lazy import of transformers.trainer, and something inside that import chain fails (missing dependency, incompatible versions, or a conflicting package). The wrapper then shows up as: “Could not import module ‘Trainer’ …” (GitHub)

For your exact install line:

pip install transformers torch gradio datasets --upgrade

the two most common problems are:

  1. accelerate is missing / incompatible
    Hugging Face’s Trainer docs explicitly note Trainer is powered by Accelerate and even start with installing/upgrading it. (Hugging Face)
    A canonical Transformers issue shows this exact class of failure and the recommended fix: install transformers[torch] or upgrade accelerate. (GitHub)

  2. Your Python / Transformers version combination is incompatible
    As of Transformers 5.1.0 (released Feb 5, 2026), PyPI metadata says it requires Python >= 3.10 and recommends installing with pip install "transformers[torch]". (PyPI)
    There was also a reported case where a v4 release “declared” Python 3.9 compatibility but failed at runtime when importing Trainer on Python 3.9 due to 3.10-only syntax. (GitHub)

A third “gotcha” I would not ignore: in your message you typed –force-reinstall (that looks like an en dash, not two ASCII hyphens). An en dash is a different Unicode character and may not be parsed as an option correctly. (Stack Overflow)

What I would do (works in Colab/Jupyter/Kaggle as well)

0) Confirm Python and where pip installs

Run:

import sys
print(sys.version)
print(sys.executable)
!{sys.executable} -m pip --version

If Python is < 3.10, you should upgrade Python (or pin Transformers to an older version that truly supports your Python). Transformers 5.x requires Python >= 3.10. (PyPI)

1) Install the “Trainer-correct” dependency set

Do not rely on pip install transformers ... alone. Use the extra that Transformers itself recommends:

import sys
!{sys.executable} -m pip install -U --upgrade-strategy eager --no-cache-dir "transformers[torch]" accelerate datasets gradio huggingface_hub

Why this exact approach:

  • PyPI explicitly recommends pip install "transformers[torch]". (PyPI)
  • The Trainer docs call out installing/upgrading accelerate. (Hugging Face)
  • Transformers issues repeatedly point to accelerate / extras as the fix when Trainer import fails. (GitHub)

Then restart the runtime/kernel (important in notebooks; otherwise old modules remain loaded).

2) Verify you’re importing the packages you think you installed

After restart:

import transformers, accelerate, datasets, huggingface_hub
print("transformers", transformers.__version__, transformers.__file__)
print("accelerate", accelerate.__version__, accelerate.__file__)
print("datasets", datasets.__version__, datasets.__file__)
print("huggingface_hub", huggingface_hub.__version__, huggingface_hub.__file__)

This catches two important failure modes:

  • “pip installed into a different environment than the kernel”
  • importing a conflicting package named datasets (see below)

3) Import Trainer

from transformers import Trainer, TrainingArguments

If it works now, you are done.

If it still fails: extract the real underlying exception

The wrapper message is not enough. Run:

import importlib, traceback
try:
    importlib.import_module("transformers.trainer")
except Exception:
    traceback.print_exc()

Then apply the matching fix:

A) Error mentions accelerate>=... or PartialState

That’s a Transformers ↔ Accelerate mismatch. The official guidance is to install transformers[torch] or upgrade accelerate. (GitHub)
Fix:

pip install -U accelerate "transformers[torch]"

B) Error mentions TypedDict from huggingface_hub

That’s a transformers ↔ huggingface_hub mismatch; it can break the Trainer import chain. (GitHub)
Fix:

pip install -U huggingface_hub transformers

C) Error shows it’s importing keras/datasets/... or something unexpected for datasets

There is a real-world case where transformers.trainer does import datasets and the environment resolves it incorrectly, leading to Trainer import failure. (GitHub)
Fix: ensure import datasets; print(datasets.__file__) points to Hugging Face datasets, not a different module; remove/upgrade the conflicting package or rename any local datasets.py.

D) Error mentions multiprocess conflict

There is a reported case where a third-party multiprocess package caused Trainer import to fail; uninstalling it fixed the issue. (Stack Overflow)
Fix:

pip uninstall -y multiprocess

One more correction: make sure you used the correct --force-reinstall

If you try force reinstall again, ensure it’s two normal hyphen-minus characters:

pip install -U --force-reinstall transformers torch gradio datasets

Not –force-reinstall (en dash). (Stack Overflow)

The fastest “most likely to work” recipe

If you want one sequence to run in a fresh session:

  1. Install:
import sys
!{sys.executable} -m pip install -U --no-cache-dir --upgrade-strategy eager "transformers[torch]" accelerate datasets huggingface_hub gradio

  1. Restart runtime/kernel

  2. Test:

from transformers import Trainer, TrainingArguments

If it still fails after that, the importlib.import_module("transformers.trainer") traceback will identify which branch (Accelerate vs Hub vs datasets conflict vs multiprocess vs Python version) you’re in.