AttributeError: module 'timesfm' has no attribute 'TimesFM_2p5_200M_torch'

[veridelisi]

Hello,

I tried running the example from the README in Google Colab:

import numpy as np
import timesfm

model = timesfm.TimesFM_2p5_200M_torch()
model.load_checkpoint()
model.compile(
timesfm.ForecastConfig(
max_context=1024,
max_horizon=256,
normalize_inputs=True,
use_continuous_quantile_head=True,
force_flip_invariance=True,
infer_is_positive=True,
fix_quantile_crossing=True,
)
)

But I get the following error:

AttributeError: module 'timesfm' has no attribute 'TimesFM_2p5_200M_torch'

Steps I followed:

Fresh Colab runtime

Installed with:

git clone https://github.com/google-research/timesfm.git
cd timesfm
pip install -e .

Then ran the code above.

Expected behavior: The class TimesFM_2p5_200M_torch should be accessible directly from import timesfm as shown in the README example.

Actual behavior: The attribute is missing, leading to an AttributeError.

Could you confirm whether the class has been renamed, moved to another submodule, or not yet exported in init.py?

Thank you!

[veridelisi]

TimesFM attributes: []
❌ Error: module 'timesfm' has no attribute 'TimesFM_2p5_200M_torch'

[derfugu1]

Right now, the example posted in the README is broken.

[cccdenhart]

You have to cd into src (or add it to sys path)

[axios-shosan]

if you are using torch here is a workaround

`
import torch
import numpy as np
import timesfm
import timesfm.timesfm_2p5.timesfm_2p5_torch as timesfm_2p5_torch

torch.set_float32_matmul_precision("high")
model = timesfm_2p5_torch.TimesFM_2p5_200M_torch.from_pretrained("google/timesfm-2.5-200m-pytorch")

model.compile(
forecast_config = timesfm.ForecastConfig(
max_context=1024,
max_horizon=256,
normalize_inputs=True,
use_continuous_quantile_head=True,
force_flip_invariance=True,
infer_is_positive=True,
fix_quantile_crossing=True,
)
)
point_forecast, quantile_forecast = model.forecast(
horizon=12,
inputs=[
np.linspace(0, 1, 100),
np.sin(np.linspace(0, 20, 67)),
], # Two dummy inputs
)
point_forecast.shape # (2, 12)
quantile_forecast.shape # (2, 12, 10): mean, then 10th to 90th quantiles.
`