riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 02:02:21 +06:00
Code Issues Actions 23 Packages Projects Releases Wiki Activity

Deprecate unused OpenLlama architecture (#24922)

Browse Source 
* Resolve typo in check_repo.py

* Specify encoding when opening modeling files

* Deprecate the OpenLlama architecture

* Add disclaimer pointing to Llama

I'm open to different wordings here

* Match the capitalisation of LLaMA
This commit is contained in:
Tom Aarsen 2023-07-20 13:03:24 +02:00 committed by GitHub
parent 8fd8c8e49e
commit 79444f370f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
12 changed files with 64 additions and 419 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
docs/source/en/model_doc/open-llama.md
View File

@ -16,6 +16,21 @@ rendered properly in your Markdown viewer.
# Open-Llama
<Tip warning={true}>
This model is in maintenance mode only, so we won't accept any new PRs changing its code.
If you run into any issues running this model, please reinstall the last version that supported this model: v4.31.0.
You can do so by running the following command: `pip install -U transformers==4.31.0`.
</Tip>
<Tip warning={true}>
This model differs from the [OpenLLaMA models](https://huggingface.co/models?search=openllama) on the Hugging Face Hub, which primarily use the [LLaMA](llama) architecture.
</Tip>
## Overview
The Open-Llama model was proposed in [Open-Llama project](https://github.com/s-JoL/Open-Llama) by community developer s-JoL.

22
src/transformers/__init__.py
View File

@ -278,6 +278,7 @@ _import_structure = {
"MCTCTProcessor",
],
"models.deprecated.mmbt": ["MMBTConfig"],
"models.deprecated.open_llama": ["OPEN_LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP", "OpenLlamaConfig"],
"models.deprecated.retribert": [
"RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP",
"RetriBertConfig",
@ -445,7 +446,6 @@ _import_structure = {
"NystromformerConfig",
],
"models.oneformer": ["ONEFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "OneFormerConfig", "OneFormerProcessor"],
"models.open_llama": ["OPEN_LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP", "OpenLlamaConfig"],
"models.openai": ["OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP", "OpenAIGPTConfig", "OpenAIGPTTokenizer"],
"models.opt": ["OPTConfig"],
"models.owlvit": [
@ -1536,6 +1536,9 @@ else:
]
)
_import_structure["models.deprecated.mmbt"].extend(["MMBTForClassification", "MMBTModel", "ModalEmbeddings"])
_import_structure["models.deprecated.open_llama"].extend(
["OpenLlamaForCausalLM", "OpenLlamaForSequenceClassification", "OpenLlamaModel", "OpenLlamaPreTrainedModel"]
)
_import_structure["models.deprecated.retribert"].extend(
["RETRIBERT_PRETRAINED_MODEL_ARCHIVE_LIST", "RetriBertModel", "RetriBertPreTrainedModel"]
)
@ -2300,9 +2303,6 @@ else:
"OneFormerPreTrainedModel",
]
)
_import_structure["models.open_llama"].extend(
["OpenLlamaForCausalLM", "OpenLlamaForSequenceClassification", "OpenLlamaModel", "OpenLlamaPreTrainedModel"]
)
_import_structure["models.openai"].extend(
[
"OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST",
@ -4239,6 +4239,7 @@ if TYPE_CHECKING:
MCTCTProcessor,
)
from .models.deprecated.mmbt import MMBTConfig
from .models.deprecated.open_llama import OPEN_LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenLlamaConfig
from .models.deprecated.retribert import (
RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
RetriBertConfig,
@ -4390,7 +4391,6 @@ if TYPE_CHECKING:
from .models.nllb_moe import NLLB_MOE_PRETRAINED_CONFIG_ARCHIVE_MAP, NllbMoeConfig
from .models.nystromformer import NYSTROMFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, NystromformerConfig
from .models.oneformer import ONEFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, OneFormerConfig, OneFormerProcessor
from .models.open_llama import OPEN_LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenLlamaConfig
from .models.openai import OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenAIGPTConfig, OpenAIGPTTokenizer
from .models.opt import OPTConfig
from .models.owlvit import (
@ -5334,6 +5334,12 @@ if TYPE_CHECKING:
MCTCTPreTrainedModel,
)
from .models.deprecated.mmbt import MMBTForClassification, MMBTModel, ModalEmbeddings
from .models.deprecated.open_llama import (
OpenLlamaForCausalLM,
OpenLlamaForSequenceClassification,
OpenLlamaModel,
OpenLlamaPreTrainedModel,
)
from .models.deprecated.retribert import (
RETRIBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
RetriBertModel,
@ -5954,12 +5960,6 @@ if TYPE_CHECKING:
OneFormerModel,
OneFormerPreTrainedModel,
)
from .models.open_llama import (
OpenLlamaForCausalLM,
OpenLlamaForSequenceClassification,
OpenLlamaModel,
OpenLlamaPreTrainedModel,
)
from .models.openai import (
OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST,
OpenAIGPTDoubleHeadsModel,

1
src/transformers/models/__init__.py
View File

@ -145,7 +145,6 @@ from . import (
nllb_moe,
nystromformer,
oneformer,
open_llama,
openai,
opt,
owlvit,

1
src/transformers/models/auto/configuration_auto.py
View File

@ -652,6 +652,7 @@ DEPRECATED_MODELS = [
"bort",
"mctct",
"mmbt",
"open_llama",
"retribert",
"tapex",
"trajectory_transformer",

2
src/transformers/models/open_llama/__init__.py → src/transformers/models/deprecated/open_llama/__init__.py
View File

@ -13,7 +13,7 @@
# limitations under the License.
from typing import TYPE_CHECKING
from ...utils import (
from ....utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,

4
src/transformers/models/open_llama/configuration_open_llama.py → src/transformers/models/deprecated/open_llama/configuration_open_llama.py
View File

@ -19,8 +19,8 @@
# limitations under the License.
""" Open-Llama model configuration"""
from ...configuration_utils import PretrainedConfig
from ...utils import logging
from ....configuration_utils import PretrainedConfig
from ....utils import logging
logger = logging.get_logger(__name__)

8
src/transformers/models/open_llama/modeling_open_llama.py → src/transformers/models/deprecated/open_llama/modeling_open_llama.py
View File

@ -26,10 +26,10 @@ import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
from ...activations import ACT2FN
from ...modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast, SequenceClassifierOutputWithPast
from ...modeling_utils import PreTrainedModel
from ...utils import add_start_docstrings, add_start_docstrings_to_model_forward, logging, replace_return_docstrings
from ....activations import ACT2FN
from ....modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast, SequenceClassifierOutputWithPast
from ....modeling_utils import PreTrainedModel
from ....utils import add_start_docstrings, add_start_docstrings_to_model_forward, logging, replace_return_docstrings
from .configuration_open_llama import OpenLlamaConfig

56
src/transformers/utils/dummy_pt_objects.py
View File

@ -2396,6 +2396,34 @@ class ModalEmbeddings(metaclass=DummyObject):
requires_backends(self, ["torch"])
class OpenLlamaForCausalLM(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
class OpenLlamaForSequenceClassification(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
class OpenLlamaModel(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
class OpenLlamaPreTrainedModel(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
RETRIBERT_PRETRAINED_MODEL_ARCHIVE_LIST = None
@ -5461,34 +5489,6 @@ class OneFormerPreTrainedModel(metaclass=DummyObject):
requires_backends(self, ["torch"])
class OpenLlamaForCausalLM(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
class OpenLlamaForSequenceClassification(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
class OpenLlamaModel(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
class OpenLlamaPreTrainedModel(metaclass=DummyObject):
_backends = ["torch"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["torch"])
OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST = None

0
tests/models/open_llama/__init__.py
View File

370
tests/models/open_llama/test_modeling_open_llama.py
View File

@ -1,370 +0,0 @@
# coding=utf-8
# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Testing suite for the PyTorch Open-Llama model. """
import unittest
from parameterized import parameterized
from transformers import OpenLlamaConfig, is_torch_available, set_seed
from transformers.testing_utils import require_torch, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import OpenLlamaForCausalLM, OpenLlamaForSequenceClassification, OpenLlamaModel
class OpenLlamaModelTester:
def __init__(
self,
parent,
batch_size=13,
seq_length=7,
is_training=True,
use_input_mask=True,
use_token_type_ids=False,
use_labels=True,
vocab_size=99,
hidden_size=32,
num_hidden_layers=5,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
type_sequence_label_size=2,
initializer_range=0.02,
num_labels=3,
num_choices=4,
scope=None,
):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.use_labels = use_labels
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.num_labels = num_labels
self.num_choices = num_choices
self.scope = scope
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = random_attention_mask([self.batch_size, self.seq_length])
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
sequence_labels = None
token_labels = None
choice_labels = None
if self.use_labels:
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = self.get_config()
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
def get_config(self):
return OpenLlamaConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
is_decoder=False,
initializer_range=self.initializer_range,
use_stable_embedding=False,
)
def create_and_check_model(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = OpenLlamaModel(config=config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask)
result = model(input_ids)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
def create_and_check_model_as_decoder(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
encoder_hidden_states,
encoder_attention_mask,
):
config.add_cross_attention = True
model = OpenLlamaModel(config)
model.to(torch_device)
model.eval()
result = model(
input_ids,
attention_mask=input_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
)
result = model(
input_ids,
attention_mask=input_mask,
encoder_hidden_states=encoder_hidden_states,
)
result = model(input_ids, attention_mask=input_mask)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
def create_and_check_for_causal_lm(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
encoder_hidden_states,
encoder_attention_mask,
):
model = OpenLlamaForCausalLM(config=config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask, labels=token_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
def create_and_check_decoder_model_past_large_inputs(
self,
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
encoder_hidden_states,
encoder_attention_mask,
):
config.is_decoder = True
config.add_cross_attention = True
model = OpenLlamaForCausalLM(config=config)
model.to(torch_device)
model.eval()
# first forward pass
outputs = model(
input_ids,
attention_mask=input_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
use_cache=True,
)
past_key_values = outputs.past_key_values
# create hypothetical multiple next token and extent to next_input_ids
next_tokens = ids_tensor((self.batch_size, 3), config.vocab_size)
next_mask = ids_tensor((self.batch_size, 3), vocab_size=2)
# append to next input_ids and
next_input_ids = torch.cat([input_ids, next_tokens], dim=-1)
next_attention_mask = torch.cat([input_mask, next_mask], dim=-1)
output_from_no_past = model(
next_input_ids,
attention_mask=next_attention_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
output_hidden_states=True,
)["hidden_states"][0]
output_from_past = model(
next_tokens,
attention_mask=next_attention_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
past_key_values=past_key_values,
output_hidden_states=True,
)["hidden_states"][0]
# select random slice
random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item()
output_from_no_past_slice = output_from_no_past[:, -3:, random_slice_idx].detach()
output_from_past_slice = output_from_past[:, :, random_slice_idx].detach()
self.parent.assertTrue(output_from_past_slice.shape[1] == next_tokens.shape[1])
# test that outputs are equal for slice
self.parent.assertTrue(torch.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3))
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
) = config_and_inputs
inputs_dict = {"input_ids": input_ids, "attention_mask": input_mask}
return config, inputs_dict
@require_torch
class OpenLlamaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
all_model_classes = (
(OpenLlamaModel, OpenLlamaForCausalLM, OpenLlamaForSequenceClassification) if is_torch_available() else ()
)
all_generative_model_classes = (OpenLlamaForCausalLM,) if is_torch_available() else ()
pipeline_model_mapping = (
{
"feature-extraction": OpenLlamaModel,
"text-classification": OpenLlamaForSequenceClassification,
"text-generation": OpenLlamaForCausalLM,
"zero-shot": OpenLlamaForSequenceClassification,
}
if is_torch_available()
else {}
)
test_headmasking = False
test_pruning = False
def setUp(self):
self.model_tester = OpenLlamaModelTester(self)
self.config_tester = ConfigTester(self, config_class=OpenLlamaConfig, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
def test_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
self.model_tester.create_and_check_model(*config_and_inputs)
def test_open_llama_sequence_classification_model(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.num_labels = 3
input_ids = input_dict["input_ids"]
attention_mask = input_ids.ne(1).to(torch_device)
sequence_labels = ids_tensor([self.model_tester.batch_size], self.model_tester.type_sequence_label_size)
model = OpenLlamaForSequenceClassification(config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=attention_mask, labels=sequence_labels)
self.assertEqual(result.logits.shape, (self.model_tester.batch_size, self.model_tester.num_labels))
def test_open_llama_sequence_classification_model_for_single_label(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.num_labels = 3
config.problem_type = "single_label_classification"
input_ids = input_dict["input_ids"]
attention_mask = input_ids.ne(1).to(torch_device)
sequence_labels = ids_tensor([self.model_tester.batch_size], self.model_tester.type_sequence_label_size)
model = OpenLlamaForSequenceClassification(config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=attention_mask, labels=sequence_labels)
self.assertEqual(result.logits.shape, (self.model_tester.batch_size, self.model_tester.num_labels))
def test_open_llama_sequence_classification_model_for_multi_label(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.num_labels = 3
config.problem_type = "multi_label_classification"
input_ids = input_dict["input_ids"]
attention_mask = input_ids.ne(1).to(torch_device)
sequence_labels = ids_tensor(
[self.model_tester.batch_size, config.num_labels], self.model_tester.type_sequence_label_size
).to(torch.float)
model = OpenLlamaForSequenceClassification(config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=attention_mask, labels=sequence_labels)
self.assertEqual(result.logits.shape, (self.model_tester.batch_size, self.model_tester.num_labels))
@unittest.skip("Open-Llama buffers include complex numbers, which breaks this test")
def test_save_load_fast_init_from_base(self):
pass
@parameterized.expand([("linear",), ("dynamic",)])
def test_model_rope_scaling(self, scaling_type):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
short_input = ids_tensor([1, 10], config.vocab_size)
long_input = ids_tensor([1, int(config.max_position_embeddings * 1.5)], config.vocab_size)
set_seed(42) # Fixed seed at init time so the two models get the same random weights
original_model = OpenLlamaModel(config)
original_model.to(torch_device)
original_model.eval()
original_short_output = original_model(short_input).last_hidden_state
original_long_output = original_model(long_input).last_hidden_state
set_seed(42) # Fixed seed at init time so the two models get the same random weights
config.rope_scaling = {"type": scaling_type, "factor": 10.0}
scaled_model = OpenLlamaModel(config)
scaled_model.to(torch_device)
scaled_model.eval()
scaled_short_output = scaled_model(short_input).last_hidden_state
scaled_long_output = scaled_model(long_input).last_hidden_state
# Dynamic scaling does not change the RoPE embeddings until it receives an input longer than the original
# maximum sequence length, so the outputs for the short input should match.
if scaling_type == "dynamic":
self.assertTrue(torch.allclose(original_short_output, scaled_short_output, atol=1e-5))
else:
self.assertFalse(torch.allclose(original_short_output, scaled_short_output, atol=1e-5))
# The output should be different for long inputs
self.assertFalse(torch.allclose(original_long_output, scaled_long_output, atol=1e-5))

2
utils/check_config_attributes.py
View File

@ -238,7 +238,7 @@ def check_config_attributes_being_used(config_class):
modeling_sources = []
for path in modeling_paths:
if os.path.isfile(path):
with open(path) as fp:
with open(path, encoding="utf8") as fp:
modeling_sources.append(fp.read())
unused_attributes = []

2
utils/check_repo.py
View File

@ -1093,7 +1093,7 @@ def check_deprecated_constant_is_up_to_date():
if len(missing_models) != 0:
missing_models = ", ".join(missing_models)
message.append(
"The following models are in the deprecated folder, make sur to add them to `DEPRECATED_MODELS` in "
"The following models are in the deprecated folder, make sure to add them to `DEPRECATED_MODELS` in "
f"`models/auto/configuration_auto.py`: {missing_models}."
)