Elgato_dark/chronos-forecasting
1
0
Fork 0
mirror of https://github.com/amazon-science/chronos-forecasting synced 2026-05-23 17:48:23 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix issue with new caching mechanism in transformers and bump versions (#313)

Browse source 
*Issue #, if available:* Fixes #310 and closes #302

*Description of changes:* This PR fixes an issue related to the new
caching mechanism for T5 introduced in `transformers==4.54`. [Prior
versions
set](https://github.com/huggingface/transformers/blob/v4.53.3/src/transformers/models/t5/modeling_t5.py#L1328)
`encoder_config.is_encoder_decoder = False` when initializing encoder
and decoder. Following transformers, we also initialized Chronos-Bolt in
the same way. However, in v4.54 this line [has been
removed](3fd456b200/src/transformers/models/t5/modeling_t5.py (L1301))
and [new logic has been
added](3fd456b200/src/transformers/models/t5/modeling_t5.py (L494))
which relies on `is_encoder_decoder` [being
True](3fd456b200/src/transformers/models/t5/modeling_t5.py (L1007)).
This causes Chronos-Bolt to break as described in #310. This PR removes
`is_encoder_decoder = False` for both encoder and decoder which fixes
the issue. I re-ran our mini eval in the CI and got the same results for
v4.54 and v4.48 (our current lower bound).

This PR also bumps package versions. 


By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.
This commit is contained in:
Abdul Fatir 2025-08-05 10:46:12 +02:00 committed by GitHub
parent 6a9c8dadac
commit fcd09fe8b6
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 42 additions and 24 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

50
pyproject.toml
View file

@ -1,25 +1,21 @@
[project]
name = "chronos-forecasting"
version = "1.5.2"
version = "1.5.3"
authors = [
{ name="Abdul Fatir Ansari", email="ansarnd@amazon.com" },
{ name="Lorenzo Stella", email="stellalo@amazon.com" },
{ name="Caner Turkmen", email="atturkm@amazon.com" },
{ name = "Abdul Fatir Ansari", email = "ansarnd@amazon.com" },
{ name = "Lorenzo Stella", email = "stellalo@amazon.com" },
{ name = "Caner Turkmen", email = "atturkm@amazon.com" },
]
description = "Chronos: Pretrained models for time series forecasting"
readme = "README.md"
license = { file = "LICENSE" }
requires-python = ">=3.9"
dependencies = [
"torch>=2.0,<3", # package was tested on 2.2
"transformers>=4.48,<5",
"accelerate>=0.32,<2",
]
dependencies = ["torch>=2.0,<3", "transformers>=4.48,<5", "accelerate>=0.32,<2"]
classifiers = [
"Programming Language :: Python :: 3",
"License :: OSI Approved :: Apache Software License",
"Operating System :: OS Independent",
"Topic :: Scientific/Engineering :: Artificial Intelligence",
"Programming Language :: Python :: 3",
"License :: OSI Approved :: Apache Software License",
"Operating System :: OS Independent",
"Topic :: Scientific/Engineering :: Artificial Intelligence",
]
[build-system]
@ -30,10 +26,24 @@ build-backend = "hatchling.build"
packages = ["src/chronos"]
[project.optional-dependencies]
test = ["pytest~=8.0", "numpy~=1.21"]
test = ["pytest~=8.0", "numpy>=1.21,<3"]
typecheck = ["mypy~=1.9"]
training = ["gluonts[pro]~=0.15", "numpy~=1.21", "datasets~=2.18", "typer", "typer-config", "joblib", "scikit-learn", "tensorboard"]
evaluation = ["gluonts[pro]~=0.15", "numpy~=1.21", "datasets~=2.18", "typer"]
training = [
"gluonts[pro]~=0.15",
"numpy>=1.21,<3",
"datasets>=2.18,<4",
"typer",
"typer-config",
"joblib",
"scikit-learn",
"tensorboard",
]
evaluation = [
"gluonts[pro]~=0.15",
"numpy>=1.21,<3",
"datasets>=2.18,<4",
"typer",
]
[project.urls]
Homepage = "https://github.com/amazon-science/chronos-forecasting"
@ -42,3 +52,11 @@ Paper = "https://arxiv.org/abs/2403.07815"
[tool.mypy]
ignore_missing_imports = true
[tool.ruff]
line-length = 88
lint.ignore = [
"E501", # Line too long
"E731", # Do not assign a `lambda` expression, use a `def`
"E722", # Do not use bare `except`
]

14
src/chronos/chronos.py
View file

@ -302,10 +302,12 @@ class ChronosModel(nn.Module):
A tensor of encoder embeddings with shape
(batch_size, sequence_length, d_model).
"""
assert (
self.config.model_type == "seq2seq"
), "Encoder embeddings are only supported for encoder-decoder models"
assert hasattr(self.model, "encoder")
assert self.config.model_type == "seq2seq", (
"Encoder embeddings are only supported for encoder-decoder models"
)
assert hasattr(self.model, "encoder") and isinstance(
self.model.encoder, nn.Module
)
return self.model.encoder(
input_ids=input_ids, attention_mask=attention_mask
@ -346,7 +348,7 @@ class ChronosModel(nn.Module):
if top_p is None:
top_p = self.config.top_p
assert hasattr(self.model, "generate")
assert callable(getattr(self.model, "generate", None))
preds = self.model.generate(
input_ids=input_ids,
@ -362,7 +364,7 @@ class ChronosModel(nn.Module):
top_k=top_k,
top_p=top_p,
),
)
) # type: ignore
if self.config.model_type == "seq2seq":
preds = preds[..., 1:] # remove the decoder start token

2
src/chronos/chronos_bolt.py
View file

@ -179,7 +179,6 @@ class ChronosBoltModelForForecasting(T5PreTrainedModel):
encoder_config = copy.deepcopy(config)
encoder_config.is_decoder = False
encoder_config.use_cache = False
encoder_config.is_encoder_decoder = False
self.encoder = T5Stack(encoder_config, self.shared)
self._init_decoder(config)
@ -381,7 +380,6 @@ class ChronosBoltModelForForecasting(T5PreTrainedModel):
def _init_decoder(self, config):
decoder_config = copy.deepcopy(config)
decoder_config.is_decoder = True
decoder_config.is_encoder_decoder = False
decoder_config.num_layers = config.num_decoder_layers
self.decoder = T5Stack(decoder_config, self.shared)