[Bug] Runtime error in google colab

[borisalmonacid]

What happened?

This file https://github.com/oumi-ai/oumi/blob/main/notebooks/Oumi%20-%20Finetuning%20Tutorial.ipynb

In google colab get the following error message:

---

ValueError Traceback (most recent call last)
in <cell line: 0>()
----> 1 from oumi.builders import build_tokenizer
2 from oumi.core.configs import ModelParams
3 from oumi.datasets import PromptResponseDataset
4
5 # Initialize the dataset

10 frames
/usr/local/lib/python3.11/dist-packages/numpy/random/_pickle.py in
----> 1 from .mtrand import RandomState
2 from ._philox import Philox
3 from ._pcg64 import PCG64, PCG64DXSM
4 from ._sfc64 import SFC64
5

numpy/random/mtrand.pyx in init numpy.random.mtrand()

ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

Steps to reproduce the bug

1.- open in google colab this file https://github.com/oumi-ai/oumi/blob/main/notebooks/Oumi%20-%20Finetuning%20Tutorial.ipynb

2.- run the project

System Info

N/A

[taenin]

Thanks for the report, @borisalmonacid ! Verified that I can reproduce this in a google collab. Taking a look now

[taenin]

The result of oumi env in google collab:

@@@@@@@@@@@@@@@@@@@
@                 @
@   @@@@@  @  @   @
@   @   @  @  @   @
@   @@@@@  @@@@   @
@                 @
@   @@@@@@@   @   @
@   @  @  @   @   @
@   @  @  @   @   @
@                 @
@@@@@@@@@@@@@@@@@@@

----------Oumi environment information:----------

Oumi version: 0.1.4
Python version: 3.11.11
Platform: Linux-6.1.85+-x86_64-with-glibc2.35

Installed dependencies:
PACKAGE              VERSION
accelerate           1.2.1
aiohttp              3.11.13
bitsandbytes         0.45.3
datasets             3.2.0
diffusers            0.32.2
einops               0.8.1
jsonlines            4.0.0
liger-kernel         0.5.5
llama-cpp-python     <not installed>
lm-eval              0.4.8
numpy                1.26.4
nvidia-ml-py         12.560.30
omegaconf            2.3.0
open_clip_torch      <not installed>
pandas               2.2.2
peft                 0.14.0
pexpect              4.8.0
pillow               10.3.0
pydantic             2.9.2
responses            0.25.7
skypilot             0.7.0
tensorboard          2.18.0
timm                 1.0.15
torch                2.5.1
torchdata            0.9.0
torchvision          0.20.1
tqdm                 4.67.1
transformers         4.45.2
trl                  0.11.4
typer                0.15.2
vllm                 0.7.0
wandb                0.19.8

Environment variables:
VARIABLE                            VALUE
ACCELERATE_DYNAMO_BACKEND           <not set>
ACCELERATE_DYNAMO_MODE              <not set>
ACCELERATE_DYNAMO_USE_DYNAMIC       <not set>
ACCELERATE_DYNAMO_USE_FULLGRAPH     <not set>
ACCELERATE_USE_FSDP                 <not set>
LOCAL_RANK                          <not set>
LOCAL_WORLD_SIZE                    <not set>
OUMI_EXTRA_DEPS_FILE                <not set>
RANK                                <not set>
WORLD_SIZE                          <not set>

PyTorch information:
CUDA available: False

[taenin]

Before installing oumi, the installed version of numpy is:

import numpy
print(numpy.__version__)

2.0.2

[borisalmonacid]

@@@@@@@@@@@@@@@@@@@
@ @
@ @@@@@ @ @ @
@ @ @ @ @ @
@ @@@@@ @@@@ @
@ @
@ @@@@@@@ @ @
@ @ @ @ @ @
@ @ @ @ @ @
@ @
@@@@@@@@@@@@@@@@@@@

----------Oumi environment information:----------

Oumi version: 0.1.4
Python version: 3.11.11
Platform: Linux-6.1.85+-x86_64-with-glibc2.35

Installed dependencies:
PACKAGE VERSION
accelerate 1.2.1
aiohttp 3.11.13
bitsandbytes 0.45.3
datasets 3.2.0
diffusers 0.32.2
einops 0.8.1
jsonlines 4.0.0
liger-kernel 0.5.5
llama-cpp-python
lm-eval 0.4.8
numpy 1.26.4
nvidia-ml-py 12.560.30
omegaconf 2.3.0
open_clip_torch
pandas 2.2.2
peft 0.14.0
pexpect 4.8.0
pillow 10.3.0
pydantic 2.9.2
responses 0.25.7
skypilot 0.7.0
tensorboard 2.18.0
timm 1.0.15
torch 2.5.1
torchdata 0.9.0
torchvision 0.20.1
tqdm 4.67.1
transformers 4.45.2
trl 0.11.4
typer 0.15.2
vllm 0.7.0
wandb 0.19.8

Environment variables:
VARIABLE VALUE
ACCELERATE_DYNAMO_BACKEND
ACCELERATE_DYNAMO_MODE
ACCELERATE_DYNAMO_USE_DYNAMIC
ACCELERATE_DYNAMO_USE_FULLGRAPH
ACCELERATE_USE_FSDP
LOCAL_RANK
LOCAL_WORLD_SIZE
OUMI_EXTRA_DEPS_FILE
RANK
WORLD_SIZE

PyTorch information:
CUDA available: True
CUDA version: 12.4
cuDNN version: 90.1.0
Number of GPUs: 1
GPU type: Tesla T4
GPU memory: 14.7GB

[taenin]

Looks like for some reason colab isn't install the latest version of oumi

It should be 0.1.8 instead of 0.1.4...

[borisalmonacid]

Perhaps Google Collab has an outdated Pip or isn't bringing the latest published version of the package.
I think you could try forcing the required Oumi version into Pip, and at the same time eliminate a potential external problem.

[taenin]

After testing several iterations in colab, it looks like uv isn't installing the proper oumi version. To get unblocked quickly, you can swap out the cell:

%pip install uv -q
!uv pip install oumi[gpu] vllm --no-progress --system

with the following:

%pip install oumi[gpu] vllm

That should get you unblocked, albeit with a slower install. I'm going to dig a bit further as to why uv isn't picking up the correct version here.

[borisalmonacid]

works :)