From c23a1c19322152dbc89cfaa5f22310bd0afb92d0 Mon Sep 17 00:00:00 2001 From: Arthur <48595927+ArthurZucker@users.noreply.github.com> Date: Mon, 13 Jan 2025 18:41:15 +0100 Subject: [PATCH] Add-helium (#35669) * Add the helium model. * Add a missing helium. * And add another missing helium. * Use float for the rmsnorm mul. * Add the Helium tokenizer converter. * Add the pad token as suggested by Arthur. * Update the RMSNorm + some other tweaks. * Fix more rebase issues. * fix copies and style * fixes and add helium.md * add missing tests * udpate the backlink * oups * style * update init, and expected results * small fixes * match test outputs * style fixup, fix doc builder * add dummies and we should be good to go!z * update sdpa and fa2 documentation --------- Co-authored-by: laurent --- docs/source/en/_toctree.yml | 2 + docs/source/en/index.md | 1 + docs/source/en/model_doc/helium.md | 158 +++ docs/source/en/perf_infer_gpu_one.md | 2 + src/transformers/__init__.py | 18 + src/transformers/convert_slow_tokenizer.py | 89 ++ src/transformers/models/__init__.py | 1 + .../models/auto/configuration_auto.py | 2 + src/transformers/models/auto/modeling_auto.py | 4 + .../models/auto/tokenization_auto.py | 1 + src/transformers/models/helium/__init__.py | 27 + .../models/helium/configuration_helium.py | 140 +++ .../models/helium/modeling_helium.py | 1065 +++++++++++++++++ .../models/helium/modular_helium.py | 171 +++ src/transformers/utils/dummy_pt_objects.py | 35 + tests/models/helium/__init__.py | 0 tests/models/helium/test_modeling_helium.py | 110 ++ 17 files changed, 1826 insertions(+) create mode 100644 docs/source/en/model_doc/helium.md create mode 100644 src/transformers/models/helium/__init__.py create mode 100644 src/transformers/models/helium/configuration_helium.py create mode 100644 src/transformers/models/helium/modeling_helium.py create mode 100644 src/transformers/models/helium/modular_helium.py create mode 100644 tests/models/helium/__init__.py create mode 100644 tests/models/helium/test_modeling_helium.py diff --git a/docs/source/en/_toctree.yml b/docs/source/en/_toctree.yml index 529b113cf1..40780d24d5 100644 --- a/docs/source/en/_toctree.yml +++ b/docs/source/en/_toctree.yml @@ -452,6 +452,8 @@ title: Granite - local: model_doc/granitemoe title: GraniteMoe + - local: model_doc/helium + title: Helium - local: model_doc/herbert title: HerBERT - local: model_doc/ibert diff --git a/docs/source/en/index.md b/docs/source/en/index.md index d66f4b031a..1c90fd71b3 100644 --- a/docs/source/en/index.md +++ b/docs/source/en/index.md @@ -173,6 +173,7 @@ Flax), PyTorch, and/or TensorFlow. | [Graphormer](model_doc/graphormer) | ✅ | ❌ | ❌ | | [Grounding DINO](model_doc/grounding-dino) | ✅ | ❌ | ❌ | | [GroupViT](model_doc/groupvit) | ✅ | ✅ | ❌ | +| [Helium](model_doc/helium) | ✅ | ❌ | ❌ | | [HerBERT](model_doc/herbert) | ✅ | ✅ | ✅ | | [Hiera](model_doc/hiera) | ✅ | ❌ | ❌ | | [Hubert](model_doc/hubert) | ✅ | ✅ | ❌ | diff --git a/docs/source/en/model_doc/helium.md b/docs/source/en/model_doc/helium.md new file mode 100644 index 0000000000..df5927544d --- /dev/null +++ b/docs/source/en/model_doc/helium.md @@ -0,0 +1,158 @@ + + +# Helium + + +## Overview + +Helium was proposed in [Announcing Helium-1 Preview](https://kyutai.org/2025/01/13/helium.html) by the Kyutai Team. + + +Helium-1 preview is a lightweight language model with 2B parameters, targeting edge and mobile devices. +It supports the following languages: English, French, German, Italian, Portuguese, Spanish. + +- **Developed by:** Kyutai +- **Model type:** Large Language Model +- **Language(s) (NLP):** English, French, German, Italian, Portuguese, Spanish +- **License:** CC-BY 4.0 + + + + +## Evaluation + + + +#### Testing Data + + + +The model was evaluated on MMLU, TriviaQA, NaturalQuestions, ARC Easy & Challenge, Open Book QA, Common Sense QA, +Physical Interaction QA, Social Interaction QA, HellaSwag, WinoGrande, Multilingual Knowledge QA, FLORES 200. + +#### Metrics + + + +We report accuracy on MMLU, ARC, OBQA, CSQA, PIQA, SIQA, HellaSwag, WinoGrande. +We report exact match on TriviaQA, NQ and MKQA. +We report BLEU on FLORES. + +### English Results + +| Benchmark | Helium-1 Preview | HF SmolLM2 (1.7B) | Gemma-2 (2.6B) | Llama-3.2 (3B) | Qwen2.5 (1.5B) | +|--------------|--------|--------|--------|--------|--------| +| | | | | | | +| MMLU | 51.2 | 50.4 | 53.1 | 56.6 | 61.0 | +| NQ | 17.3 | 15.1 | 17.7 | 22.0 | 13.1 | +| TQA | 47.9 | 45.4 | 49.9 | 53.6 | 35.9 | +| ARC E | 80.9 | 81.8 | 81.1 | 84.6 | 89.7 | +| ARC C | 62.7 | 64.7 | 66.0 | 69.0 | 77.2 | +| OBQA | 63.8 | 61.4 | 64.6 | 68.4 | 73.8 | +| CSQA | 65.6 | 59.0 | 64.4 | 65.4 | 72.4 | +| PIQA | 77.4 | 77.7 | 79.8 | 78.9 | 76.0 | +| SIQA | 64.4 | 57.5 | 61.9 | 63.8 | 68.7 | +| HS | 69.7 | 73.2 | 74.7 | 76.9 | 67.5 | +| WG | 66.5 | 65.6 | 71.2 | 72.0 | 64.8 | +| | | | | | | +| Average | 60.7 | 59.3 | 62.2 | 64.7 | 63.6 | + +#### Multilingual Results + +| Language | Benchmark | Helium-1 Preview | HF SmolLM2 (1.7B) | Gemma-2 (2.6B) | Llama-3.2 (3B) | Qwen2.5 (1.5B) | +|-----|--------------|--------|--------|--------|--------|--------| +| | | | | | | | +|German| MMLU | 45.6 | 35.3 | 45.0 | 47.5 | 49.5 | +|| ARC C | 56.7 | 38.4 | 54.7 | 58.3 | 60.2 | +|| HS | 53.5 | 33.9 | 53.4 | 53.7 | 42.8 | +|| MKQA | 16.1 | 7.1 | 18.9 | 20.2 | 10.4 | +| | | | | | | | +|Spanish| MMLU | 46.5 | 38.9 | 46.2 | 49.6 | 52.8 | +|| ARC C | 58.3 | 43.2 | 58.8 | 60.0 | 68.1 | +|| HS | 58.6 | 40.8 | 60.5 | 61.1 | 51.4 | +|| MKQA | 16.0 | 7.9 | 18.5 | 20.6 | 10.6 | + + +## Technical Specifications + +### Model Architecture and Objective + +| Hyperparameter | Value | +|--------------|--------| +| Layers | 24 | +| Heads | 20 | +| Model dimension | 2560 | +| MLP dimension | 7040 | +| Context size | 4096 | +| Theta RoPE | 100,000 | + +Tips: + +- This model was contributed by [Laurent Mazare](https://huggingface.co/lmz) + + +## Usage tips + +`Helium` can be found on the [Huggingface Hub](https://huggingface.co/collections/kyutai/helium-1-preview) + +In the following, we demonstrate how to use `helium-1-preview` for the inference. + +```python +>>> from transformers import AutoModelForCausalLM, AutoTokenizer +>>> device = "cuda" # the device to load the model onto + +>>> model = AutoModelForCausalLM.from_pretrained("helium-1-preview", device_map="auto") +>>> tokenizer = AutoTokenizer.from_pretrained("helium-1-preview") + +>>> prompt = "Give me a short introduction to large language model." + +>>> messages = [{"role": "user", "content": prompt}] + +>>> text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) + +>>> model_inputs = tokenizer([text], return_tensors="pt").to(device) + +>>> generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=512, do_sample=True) + +>>> generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)] + +>>> response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] +``` + +## HeliumConfig + +[[autodoc]] HeliumConfig + +## HeliumModel + +[[autodoc]] HeliumModel + - forward + +## HeliumForCausalLM + +[[autodoc]] HeliumForCausalLM + - forward + +## HeliumForSequenceClassification + +[[autodoc]] HeliumForSequenceClassification + - forward + +## HeliumForTokenClassification + +[[autodoc]] HeliumForTokenClassification + - forward diff --git a/docs/source/en/perf_infer_gpu_one.md b/docs/source/en/perf_infer_gpu_one.md index 3e6d764617..d9bdf6f6e4 100644 --- a/docs/source/en/perf_infer_gpu_one.md +++ b/docs/source/en/perf_infer_gpu_one.md @@ -109,6 +109,7 @@ FlashAttention-2 is currently supported for the following architectures: * [SigLIP](https://huggingface.co/docs/transformers/model_doc/siglip) * [UniSpeech](https://huggingface.co/docs/transformers/v4.39.3/en/model_doc/unispeech#transformers.UniSpeechModel) * [unispeech_sat](https://huggingface.co/docs/transformers/v4.39.3/en/model_doc/unispeech-sat#transformers.UniSpeechSatModel) +* [helium](https://huggingface.co/docs/transformers/main/en/model_doc/heliumtransformers.HeliumModel) You can request to add FlashAttention-2 support for another model by opening a GitHub Issue or Pull Request. @@ -324,6 +325,7 @@ For now, Transformers supports SDPA inference and training for the following arc * [XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta#transformers.XLMRobertaModel) * [XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl#transformers.XLMRobertaXLModel) * [YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos#transformers.YolosModel) +* [helium](https://huggingface.co/docs/transformers/main/en/model_doc/heliumtransformers.HeliumModel) diff --git a/src/transformers/__init__.py b/src/transformers/__init__.py index 20b8180f9d..1cf0f88ad6 100755 --- a/src/transformers/__init__.py +++ b/src/transformers/__init__.py @@ -498,6 +498,7 @@ _import_structure = { "GroupViTTextConfig", "GroupViTVisionConfig", ], + "models.helium": ["HeliumConfig"], "models.herbert": ["HerbertTokenizer"], "models.hiera": ["HieraConfig"], "models.hubert": ["HubertConfig"], @@ -2506,6 +2507,15 @@ else: "GroupViTVisionModel", ] ) + _import_structure["models.helium"].extend( + [ + "HeliumForCausalLM", + "HeliumForSequenceClassification", + "HeliumForTokenClassification", + "HeliumModel", + "HeliumPreTrainedModel", + ] + ) _import_structure["models.hiera"].extend( [ "HieraBackbone", @@ -5529,6 +5539,7 @@ if TYPE_CHECKING: GroupViTTextConfig, GroupViTVisionConfig, ) + from .models.helium import HeliumConfig from .models.herbert import HerbertTokenizer from .models.hiera import HieraConfig from .models.hubert import HubertConfig @@ -7371,6 +7382,13 @@ if TYPE_CHECKING: GroupViTTextModel, GroupViTVisionModel, ) + from .models.helium import ( + HeliumForCausalLM, + HeliumForSequenceClassification, + HeliumForTokenClassification, + HeliumModel, + HeliumPreTrainedModel, + ) from .models.hiera import ( HieraBackbone, HieraForImageClassification, diff --git a/src/transformers/convert_slow_tokenizer.py b/src/transformers/convert_slow_tokenizer.py index 030e3a6664..c821d2db63 100644 --- a/src/transformers/convert_slow_tokenizer.py +++ b/src/transformers/convert_slow_tokenizer.py @@ -1446,6 +1446,95 @@ class MoshiConverter(SpmConverter): return pre_tokenizers.Metaspace(replacement=replacement, prepend_scheme=prepend_scheme, split=False) +class HeliumConverter(SpmConverter): + handle_byte_fallback = True + + def __init__(self, vocab_file=None, *args): + requires_backends(self, "protobuf") + + Converter.__init__(self, vocab_file) + + model_pb2 = import_protobuf() + + m = model_pb2.ModelProto() + with open(vocab_file, "rb") as f: + m.ParseFromString(f.read()) + self.proto = m + + def tokenizer(self, proto): + vocab_scores = self.vocab(proto) + tokenizer = Tokenizer( + Unigram( + vocab_scores, + unk_id=self.unk_id(proto), + byte_fallback=self.handle_byte_fallback, + ) + ) + # control tokens are special + # user defined symbols are not + # both user and control tokens are AddedTokens + # Add user defined symbols (type == 4) from sentencepiece (https://github.com/google/sentencepiece/blob/6225e08edb2577757163b3f5dbba4c0b670ef445/src/sentencepiece_model.proto#L299C29-L299C33) + spm_added_tokens = [ + (id, p.piece, p.type == 3 or p.piece in self.special_tokens) + for id, p in enumerate(proto.pieces) + if p.type in [3, 4] + ] + tokenizer.add_tokens( + [ + AddedToken(token, normalized=False, special=special, single_word=True) + for id, token, special in sorted(spm_added_tokens, key=lambda x: x[0]) + ] + ) + tokenizer.add_tokens([AddedToken("\n", normalized=False, special=False)]) + tokenizer.enable_padding(pad_token="", pad_id=3) + return tokenizer + + def vocab(self, proto): + vocab = [] + for piece in proto.pieces: + if piece.piece == "<0x0A>": + vocab += [("\n", piece.score)] + else: + vocab += [(piece.piece, piece.score)] + return vocab + + def unk_id(self, proto): + unk_id = 0 + return unk_id + + def decoder(self, replacement, add_prefix_space): + sequence = [ + decoders.Replace("▁", " "), + decoders.ByteFallback(), + decoders.Fuse(), + ] + sequence += [decoders.Strip(content=" ", left=1)] + return decoders.Sequence(sequence) + + def normalizer(self, proto): + return normalizers.Sequence([normalizers.Prepend(" "), normalizers.Replace(r" ", "▁")]) + + def pre_tokenizer(self, replacement, add_prefix_space): + return pre_tokenizers.Sequence([pre_tokenizers.Split("\n", "contiguous")]) + + def post_processor(self): + return processors.TemplateProcessing( + single=[ + "", + "$A", + ], + pair=[ + "", + "$A", + "", + "$B", + ], + special_tokens=[ + ("", 1), + ], + ) + + # Copied from transformers.models.gpt2.tokenization_gpt2.bytes_to_unicode def bytes_to_unicode(): """ diff --git a/src/transformers/models/__init__.py b/src/transformers/models/__init__.py index 7db328f87a..b84a6c6e48 100644 --- a/src/transformers/models/__init__.py +++ b/src/transformers/models/__init__.py @@ -117,6 +117,7 @@ from . import ( granitemoe, grounding_dino, groupvit, + helium, herbert, hiera, hubert, diff --git a/src/transformers/models/auto/configuration_auto.py b/src/transformers/models/auto/configuration_auto.py index 985fe59582..659fc44151 100644 --- a/src/transformers/models/auto/configuration_auto.py +++ b/src/transformers/models/auto/configuration_auto.py @@ -137,6 +137,7 @@ CONFIG_MAPPING_NAMES = OrderedDict( ("graphormer", "GraphormerConfig"), ("grounding-dino", "GroundingDinoConfig"), ("groupvit", "GroupViTConfig"), + ("helium", "HeliumConfig"), ("hiera", "HieraConfig"), ("hubert", "HubertConfig"), ("ibert", "IBertConfig"), @@ -458,6 +459,7 @@ MODEL_NAMES_MAPPING = OrderedDict( ("graphormer", "Graphormer"), ("grounding-dino", "Grounding DINO"), ("groupvit", "GroupViT"), + ("helium", "Helium"), ("herbert", "HerBERT"), ("hiera", "Hiera"), ("hubert", "Hubert"), diff --git a/src/transformers/models/auto/modeling_auto.py b/src/transformers/models/auto/modeling_auto.py index bf54a6ce97..53733c6a47 100644 --- a/src/transformers/models/auto/modeling_auto.py +++ b/src/transformers/models/auto/modeling_auto.py @@ -132,6 +132,7 @@ MODEL_MAPPING_NAMES = OrderedDict( ("graphormer", "GraphormerModel"), ("grounding-dino", "GroundingDinoModel"), ("groupvit", "GroupViTModel"), + ("helium", "HeliumModel"), ("hiera", "HieraModel"), ("hubert", "HubertModel"), ("ibert", "IBertModel"), @@ -517,6 +518,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict( ("gptj", "GPTJForCausalLM"), ("granite", "GraniteForCausalLM"), ("granitemoe", "GraniteMoeForCausalLM"), + ("helium", "HeliumForCausalLM"), ("jamba", "JambaForCausalLM"), ("jetmoe", "JetMoeForCausalLM"), ("llama", "LlamaForCausalLM"), @@ -989,6 +991,7 @@ MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict( ("gpt_neo", "GPTNeoForSequenceClassification"), ("gpt_neox", "GPTNeoXForSequenceClassification"), ("gptj", "GPTJForSequenceClassification"), + ("helium", "HeliumForSequenceClassification"), ("ibert", "IBertForSequenceClassification"), ("jamba", "JambaForSequenceClassification"), ("jetmoe", "JetMoeForSequenceClassification"), @@ -1182,6 +1185,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict( ("gpt_bigcode", "GPTBigCodeForTokenClassification"), ("gpt_neo", "GPTNeoForTokenClassification"), ("gpt_neox", "GPTNeoXForTokenClassification"), + ("helium", "HeliumForTokenClassification"), ("ibert", "IBertForTokenClassification"), ("layoutlm", "LayoutLMForTokenClassification"), ("layoutlmv2", "LayoutLMv2ForTokenClassification"), diff --git a/src/transformers/models/auto/tokenization_auto.py b/src/transformers/models/auto/tokenization_auto.py index 2e26cea971..9ce9edd06c 100644 --- a/src/transformers/models/auto/tokenization_auto.py +++ b/src/transformers/models/auto/tokenization_auto.py @@ -226,6 +226,7 @@ else: ("gptsan-japanese", ("GPTSanJapaneseTokenizer", None)), ("grounding-dino", ("BertTokenizer", "BertTokenizerFast" if is_tokenizers_available() else None)), ("groupvit", ("CLIPTokenizer", "CLIPTokenizerFast" if is_tokenizers_available() else None)), + ("helium", (None, "PreTrainedTokenizerFast" if is_tokenizers_available() else None)), ("herbert", ("HerbertTokenizer", "HerbertTokenizerFast" if is_tokenizers_available() else None)), ("hubert", ("Wav2Vec2CTCTokenizer", None)), ("ibert", ("RobertaTokenizer", "RobertaTokenizerFast" if is_tokenizers_available() else None)), diff --git a/src/transformers/models/helium/__init__.py b/src/transformers/models/helium/__init__.py new file mode 100644 index 0000000000..73d0966e5c --- /dev/null +++ b/src/transformers/models/helium/__init__.py @@ -0,0 +1,27 @@ +# Copyright 2024 The HuggingFace 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. +from typing import TYPE_CHECKING + +from ...utils import _LazyModule +from ...utils.import_utils import define_import_structure + + +if TYPE_CHECKING: + from .configuration_helium import * + from .modeling_helium import * +else: + import sys + + _file = globals()["__file__"] + sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__) diff --git a/src/transformers/models/helium/configuration_helium.py b/src/transformers/models/helium/configuration_helium.py new file mode 100644 index 0000000000..58ed1c106c --- /dev/null +++ b/src/transformers/models/helium/configuration_helium.py @@ -0,0 +1,140 @@ +# coding=utf-8 +# Copyright 2024 The Kyutai and HuggingFace Inc. teams. 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. + +from ...configuration_utils import PretrainedConfig + + +class HeliumConfig(PretrainedConfig): + r""" + This is the configuration class to store the configuration of a [`HeliumModel`]. It is used to instantiate an Helium + model according to the specified arguments, defining the model architecture. Instantiating a configuration with the + defaults will yield a similar configuration to that of the Helium 2b model. + e.g. [kyutai/helium-2b](https://huggingface.co/kyutai/helium-2b) + Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the + documentation from [`PretrainedConfig`] for more information. + Args: + vocab_size (`int`, *optional*, defaults to 48000): + Vocabulary size of the Helium model. Defines the number of different tokens that can be represented by the + `inputs_ids` passed when calling [`HeliumModel`] + hidden_size (`int`, *optional*, defaults to 2560): + Dimension of the hidden representations. + intermediate_size (`int`, *optional*, defaults to 7040): + Dimension of the MLP representations. + num_hidden_layers (`int`, *optional*, defaults to 24): + Number of hidden layers in the Transformer decoder. + num_attention_heads (`int`, *optional*, defaults to 20): + Number of attention heads for each attention layer in the Transformer decoder. + num_key_value_heads (`int`, *optional*, defaults to 20): + This is the number of key_value heads that should be used to implement Grouped Query Attention. If + `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if + `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When + converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed + by meanpooling all the original heads within that group. For more details checkout [this + paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to + `num_attention_heads`. + head_dim (`int`, *optional*, defaults to 128): + The attention head dimension. + hidden_act (`str` or `function`, *optional*, defaults to `"silu"`): + The legacy activation function. It is overwritten by the `hidden_activation`. + attention_dropout (`float`, *optional*, defaults to 0.0): + The dropout ratio for the attention probabilities. + max_position_embeddings (`int`, *optional*, defaults to 4096): + The maximum sequence length that this model might ever be used with. + initializer_range (`float`, *optional*, defaults to 0.02): + The standard deviation of the truncated_normal_initializer for initializing all weight matrices. + rms_norm_eps (`float`, *optional*, defaults to 1e-08): + The epsilon used by the rms normalization layers. + use_cache (`bool`, *optional*, defaults to `True`): + Whether or not the model should return the last key/values attentions (not used by all models). Only + relevant if `config.is_decoder=True`. + tie_word_embeddings (`bool`, *optional*, defaults to `False`): + Whether to tie weight embeddings + rope_theta (`float`, *optional*, defaults to 100000.0): + The base period of the RoPE embeddings. + pad_token_id (`int`, *optional*, defaults to 3): + Padding token id. + eos_token_id (`int` | `list`, *optional*, defaults to 2): + End of stream token id. + bos_token_id (`int`, *optional*, defaults to 1): + Beginning of stream token id. + attention_bias (`bool`, *optional*, defaults to `False`): + Whether to use a bias in the query, key, value and output projection layers during self-attention. + mlp_bias (`bool`, *optional*, defaults to `False`): + Whether to use a bias in up_proj, down_proj and gate_proj layers in the MLP layers. + ```python + >>> from transformers import HeliumModel, HeliumConfig + >>> # Initializing a Helium 2b style configuration + >>> configuration = HeliumConfig() + >>> # Initializing a model from the Helium 2b style configuration + >>> model = HeliumModel(configuration) + >>> # Accessing the model configuration + >>> configuration = model.config + ```""" + + model_type = "helium" + keys_to_ignore_at_inference = ["past_key_values"] + + def __init__( + self, + vocab_size=48000, + hidden_size=2560, + intermediate_size=7040, + num_hidden_layers=24, + num_attention_heads=20, + num_key_value_heads=20, + head_dim=128, + hidden_act="silu", + attention_dropout=0.0, + max_position_embeddings=4096, + initializer_range=0.02, + rms_norm_eps=1e-8, + use_cache=True, + tie_word_embeddings=False, + rope_theta=100000.0, + pad_token_id=3, + eos_token_id=2, + bos_token_id=1, + attention_bias=False, + mlp_bias=False, + **kwargs, + ): + self.vocab_size = vocab_size + self.max_position_embeddings = max_position_embeddings + self.hidden_size = hidden_size + self.intermediate_size = intermediate_size + self.num_hidden_layers = num_hidden_layers + self.num_attention_heads = num_attention_heads + self.num_key_value_heads = num_key_value_heads + self.head_dim = head_dim + self.hidden_act = hidden_act + self.initializer_range = initializer_range + self.rms_norm_eps = rms_norm_eps + self.use_cache = use_cache + self.rope_theta = rope_theta + self.attention_bias = attention_bias + self.attention_dropout = attention_dropout + self.mlp_bias = mlp_bias + + super().__init__( + pad_token_id=pad_token_id, + bos_token_id=bos_token_id, + eos_token_id=eos_token_id, + tie_word_embeddings=tie_word_embeddings, + **kwargs, + ) + + +__all__ = ["HeliumConfig"] diff --git a/src/transformers/models/helium/modeling_helium.py b/src/transformers/models/helium/modeling_helium.py new file mode 100644 index 0000000000..7eed89b4af --- /dev/null +++ b/src/transformers/models/helium/modeling_helium.py @@ -0,0 +1,1065 @@ +# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 +# This file was automatically generated from src/transformers/models/helium/modular_helium.py. +# Do NOT edit this file manually as any edits will be overwritten by the generation of +# the file from the modular. If any change should be done, please apply the change to the +# modular_helium.py file directly. One of our CI enforces this. +# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 +# coding=utf-8 +# Copyright 2024 The Kyutai and HuggingFace Inc. teams. 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. +import math +from typing import Callable, List, Optional, Tuple, Union + +import torch +import torch.nn as nn + +from ...activations import ACT2FN +from ...cache_utils import Cache, DynamicCache, StaticCache +from ...generation import GenerationMixin +from ...modeling_attn_mask_utils import AttentionMaskConverter +from ...modeling_flash_attention_utils import FlashAttentionKwargs +from ...modeling_outputs import ( + BaseModelOutputWithPast, + CausalLMOutputWithPast, + SequenceClassifierOutputWithPast, + TokenClassifierOutput, +) +from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS +from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel +from ...processing_utils import Unpack +from ...utils import ( + LossKwargs, + add_code_sample_docstrings, + add_start_docstrings, + add_start_docstrings_to_model_forward, + logging, + replace_return_docstrings, +) +from .configuration_helium import HeliumConfig + + +logger = logging.get_logger(__name__) + +_CHECKPOINT_FOR_DOC = "google/helium-7b" +_CONFIG_FOR_DOC = "HeliumConfig" + + +class HeliumRMSNorm(nn.Module): + def __init__(self, hidden_size, eps=1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(hidden_size)) + self.variance_epsilon = eps + + def forward(self, hidden_states): + input_dtype = hidden_states.dtype + hidden_states = hidden_states.to(torch.float32) + variance = hidden_states.pow(2).mean(-1, keepdim=True) + hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon) + return (self.weight.to(torch.float32) * hidden_states).to(input_dtype) + + def extra_repr(self): + return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}" + + +class HeliumRotaryEmbedding(nn.Module): + def __init__(self, config: HeliumConfig, device=None): + super().__init__() + # BC: "rope_type" was originally "type" + if hasattr(config, "rope_scaling") and config.rope_scaling is not None: + self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type")) + else: + self.rope_type = "default" + self.max_seq_len_cached = config.max_position_embeddings + self.original_max_seq_len = config.max_position_embeddings + + self.config = config + self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type] + + inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self.original_inv_freq = self.inv_freq + + def _dynamic_frequency_update(self, position_ids, device): + """ + dynamic RoPE layers should recompute `inv_freq` in the following situations: + 1 - growing beyond the cached sequence length (allow scaling) + 2 - the current sequence length is in the original scale (avoid losing precision with small sequences) + """ + seq_len = torch.max(position_ids) + 1 + if seq_len > self.max_seq_len_cached: # growth + inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, seq_len=seq_len) + self.register_buffer("inv_freq", inv_freq, persistent=False) # TODO joao: may break with compilation + self.max_seq_len_cached = seq_len + + if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len: # reset + # This .to() is needed if the model has been moved to a device after being initialized (because + # the buffer is automatically moved, but not the original copy) + self.original_inv_freq = self.original_inv_freq.to(device) + self.register_buffer("inv_freq", self.original_inv_freq, persistent=False) + self.max_seq_len_cached = self.original_max_seq_len + + @torch.no_grad() + def forward(self, x, position_ids): + if "dynamic" in self.rope_type: + self._dynamic_frequency_update(position_ids, device=x.device) + + # Core RoPE block + inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1) + position_ids_expanded = position_ids[:, None, :].float() + # Force float32 (see https://github.com/huggingface/transformers/pull/29285) + device_type = x.device.type + device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu" + with torch.autocast(device_type=device_type, enabled=False): + freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2) + emb = torch.cat((freqs, freqs), dim=-1) + cos = emb.cos() + sin = emb.sin() + + # Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention + cos = cos * self.attention_scaling + sin = sin * self.attention_scaling + + return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype) + + +class HeliumMLP(nn.Module): + def __init__(self, config): + super().__init__() + self.config = config + self.hidden_size = config.hidden_size + self.intermediate_size = config.intermediate_size + self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias) + self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias) + self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.mlp_bias) + self.act_fn = ACT2FN[config.hidden_act] + + def forward(self, x): + down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x)) + return down_proj + + +def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor: + """ + This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, + num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim) + """ + batch, num_key_value_heads, slen, head_dim = hidden_states.shape + if n_rep == 1: + return hidden_states + hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim) + return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim) + + +def eager_attention_forward( + module: nn.Module, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + attention_mask: Optional[torch.Tensor], + scaling: float, + dropout: float = 0.0, + **kwargs, +): + key_states = repeat_kv(key, module.num_key_value_groups) + value_states = repeat_kv(value, module.num_key_value_groups) + + attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling + if attention_mask is not None: + causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] + attn_weights = attn_weights + causal_mask + + attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype) + attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training) + attn_output = torch.matmul(attn_weights, value_states) + attn_output = attn_output.transpose(1, 2).contiguous() + + return attn_output, attn_weights + + +def rotate_half(x): + """Rotates half the hidden dims of the input.""" + x1 = x[..., 0::2] + x2 = x[..., 1::2] + return torch.stack((-x2, x1), dim=-1).flatten(-2) + + +def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1): + """Applies Rotary Position Embedding to the query and key tensors. + + Args: + q (`torch.Tensor`): The query tensor. + k (`torch.Tensor`): The key tensor. + cos (`torch.Tensor`): The cosine part of the rotary embedding. + sin (`torch.Tensor`): The sine part of the rotary embedding. + position_ids (`torch.Tensor`, *optional*): + Deprecated and unused. + unsqueeze_dim (`int`, *optional*, defaults to 1): + The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and + sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note + that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and + k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes + cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have + the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2. + Returns: + `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding. + """ + cos = cos.unsqueeze(unsqueeze_dim) + sin = sin.unsqueeze(unsqueeze_dim) + + # Interleave them instead of usual shape + cos = cos[..., : cos.shape[-1] // 2].repeat_interleave(2, dim=-1) + sin = sin[..., : sin.shape[-1] // 2].repeat_interleave(2, dim=-1) + + q_embed = (q * cos) + (rotate_half(q) * sin) + k_embed = (k * cos) + (rotate_half(k) * sin) + + return q_embed, k_embed + + +class HeliumAttention(nn.Module): + """Multi-headed attention from 'Attention Is All You Need' paper""" + + def __init__(self, config: HeliumConfig, layer_idx: Optional[int] = None): + super().__init__() + self.config = config + self.layer_idx = layer_idx + self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads) + self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads + self.scaling = 1 / math.sqrt(self.head_dim) + self.attention_dropout = config.attention_dropout + self.is_causal = True + + self.q_proj = nn.Linear( + config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias + ) + self.k_proj = nn.Linear( + config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias + ) + self.v_proj = nn.Linear( + config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias + ) + self.o_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False) + + def forward( + self, + hidden_states: torch.Tensor, + position_embeddings: Tuple[torch.Tensor, torch.Tensor], + attention_mask: Optional[torch.Tensor], + past_key_value: Optional[Cache] = None, + cache_position: Optional[torch.LongTensor] = None, + **kwargs: Unpack[FlashAttentionKwargs], + ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: + input_shape = hidden_states.shape[:-1] + hidden_shape = (*input_shape, -1, self.head_dim) + + query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2) + key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2) + value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2) + + cos, sin = position_embeddings + query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) + + if past_key_value is not None: + # sin and cos are specific to RoPE models; cache_position needed for the static cache + cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} + key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) + + attention_interface: Callable = eager_attention_forward + if self.config._attn_implementation != "eager": + if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False): + logger.warning_once( + "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to " + 'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' + ) + else: + attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation] + + attn_output, attn_weights = attention_interface( + self, + query_states, + key_states, + value_states, + attention_mask, + dropout=0.0 if not self.training else self.attention_dropout, + scaling=self.scaling, + **kwargs, + ) + + attn_output = attn_output.reshape(*input_shape, -1).contiguous() + attn_output = self.o_proj(attn_output) + return attn_output, attn_weights + + +class HeliumDecoderLayer(nn.Module): + def __init__(self, config: HeliumConfig, layer_idx: Optional[int] = None): + super().__init__() + self.hidden_size = config.hidden_size + + self.self_attn = HeliumAttention(config=config, layer_idx=layer_idx) + + self.mlp = HeliumMLP(config) + self.input_layernorm = HeliumRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.post_attention_layernorm = HeliumRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + + def forward( + self, + hidden_states: torch.Tensor, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_value: Optional[Cache] = None, + output_attentions: Optional[bool] = False, + use_cache: Optional[bool] = False, + cache_position: Optional[torch.LongTensor] = None, + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC + **kwargs: Unpack[FlashAttentionKwargs], + ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: + residual = hidden_states + + hidden_states = self.input_layernorm(hidden_states) + + # Self Attention + hidden_states, self_attn_weights = self.self_attn( + hidden_states=hidden_states, + attention_mask=attention_mask, + position_ids=position_ids, + past_key_value=past_key_value, + output_attentions=output_attentions, + use_cache=use_cache, + cache_position=cache_position, + position_embeddings=position_embeddings, + **kwargs, + ) + hidden_states = residual + hidden_states + + # Fully Connected + residual = hidden_states + hidden_states = self.post_attention_layernorm(hidden_states) + hidden_states = self.mlp(hidden_states) + hidden_states = residual + hidden_states + + outputs = (hidden_states,) + if output_attentions: + outputs += (self_attn_weights,) + + return outputs + + +HELIUM_START_DOCSTRING = r""" + This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the + library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads + etc.) + + This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. + Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage + and behavior. + + Parameters: + config ([`HeliumConfig`]): + Model configuration class with all the parameters of the model. Initializing with a config file does not + load the weights associated with the model, only the configuration. Check out the + [`~PreTrainedModel.from_pretrained`] method to load the model weights. +""" + + +@add_start_docstrings( + "The bare Helium Model outputting raw hidden-states without any specific head on top.", + HELIUM_START_DOCSTRING, +) +class HeliumPreTrainedModel(PreTrainedModel): + config_class = HeliumConfig + base_model_prefix = "model" + supports_gradient_checkpointing = True + _no_split_modules = ["HeliumDecoderLayer"] + _skip_keys_device_placement = ["past_key_values"] + _supports_flash_attn_2 = True + _supports_sdpa = True + _supports_flex_attn = True + _supports_cache_class = True + _supports_quantized_cache = True + _supports_static_cache = True + + def _init_weights(self, module): + std = self.config.initializer_range + if isinstance(module, nn.Linear): + module.weight.data.normal_(mean=0.0, std=std) + if module.bias is not None: + module.bias.data.zero_() + elif isinstance(module, nn.Embedding): + module.weight.data.normal_(mean=0.0, std=std) + if module.padding_idx is not None: + module.weight.data[module.padding_idx].zero_() + + +HELIUM_INPUTS_DOCSTRING = r""" + Args: + input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`): + Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide + it. + + Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and + [`PreTrainedTokenizer.__call__`] for details. + + [What are input IDs?](../glossary#input-ids) + attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*): + Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`: + + - 1 for tokens that are **not masked**, + - 0 for tokens that are **masked**. + + [What are attention masks?](../glossary#attention-mask) + + Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and + [`PreTrainedTokenizer.__call__`] for details. + + If `past_key_values` is used, optionally only the last `input_ids` have to be input (see + `past_key_values`). + + If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`] + and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more + information on the default strategy. + + - 1 indicates the head is **not masked**, + - 0 indicates the head is **masked**. + position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): + Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, + config.n_positions - 1]`. + + [What are position IDs?](../glossary#position-ids) + past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *optional*): + Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention + blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values` + returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`. + + Two formats are allowed: + - a [`~cache_utils.Cache`] instance, see our + [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache); + - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of + shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy + cache format. + + The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the + legacy cache format will be returned. + + If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't + have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `input_ids` + of shape `(batch_size, sequence_length)`. + inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): + Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This + is useful if you want more control over how to convert `input_ids` indices into associated vectors than the + model's internal embedding lookup matrix. + use_cache (`bool`, *optional*): + If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see + `past_key_values`). + output_attentions (`bool`, *optional*): + Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned + tensors for more detail. + output_hidden_states (`bool`, *optional*): + Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for + more detail. + return_dict (`bool`, *optional*): + Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. + cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*): + Indices depicting the position of the input sequence tokens in the sequence. Contrarily to `position_ids`, + this tensor is not affected by padding. It is used to update the cache in the correct position and to infer + the complete sequence length. +""" + + +@add_start_docstrings( + "The bare Helium Model outputting raw hidden-states without any specific head on top.", + HELIUM_START_DOCSTRING, +) +class HeliumModel(HeliumPreTrainedModel): + """ + Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`HeliumDecoderLayer`] + + Args: + config: HeliumConfig + """ + + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.padding_idx = config.pad_token_id + self.vocab_size = config.vocab_size + + self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx) + self.layers = nn.ModuleList( + [HeliumDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] + ) + self.norm = HeliumRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.rotary_emb = HeliumRotaryEmbedding(config) + self.gradient_checkpointing = False + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.embed_tokens + + def set_input_embeddings(self, value): + self.embed_tokens = value + + @add_start_docstrings_to_model_forward(HELIUM_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Cache] = None, + inputs_embeds: Optional[torch.FloatTensor] = None, + use_cache: Optional[bool] = None, + output_attentions: Optional[bool] = None, + output_hidden_states: Optional[bool] = None, + return_dict: Optional[bool] = None, + cache_position: Optional[torch.LongTensor] = None, + **flash_attn_kwargs: Unpack[FlashAttentionKwargs], + ) -> Union[Tuple, BaseModelOutputWithPast]: + output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions + output_hidden_states = ( + output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states + ) + use_cache = use_cache if use_cache is not None else self.config.use_cache + return_dict = return_dict if return_dict is not None else self.config.use_return_dict + + if (input_ids is None) ^ (inputs_embeds is not None): + raise ValueError("You must specify exactly one of input_ids or inputs_embeds") + + if self.gradient_checkpointing and self.training and use_cache: + logger.warning_once( + "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`." + ) + use_cache = False + + if inputs_embeds is None: + inputs_embeds = self.embed_tokens(input_ids) + + if use_cache and past_key_values is None: + past_key_values = DynamicCache() + + if cache_position is None: + past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 + cache_position = torch.arange( + past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device + ) + + if position_ids is None: + position_ids = cache_position.unsqueeze(0) + + causal_mask = self._update_causal_mask( + attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions + ) + + hidden_states = inputs_embeds + + # create position embeddings to be shared across the decoder layers + position_embeddings = self.rotary_emb(hidden_states, position_ids) + + # decoder layers + all_hidden_states = () if output_hidden_states else None + all_self_attns = () if output_attentions else None + + for decoder_layer in self.layers[: self.config.num_hidden_layers]: + if output_hidden_states: + all_hidden_states += (hidden_states,) + + if self.gradient_checkpointing and self.training: + layer_outputs = self._gradient_checkpointing_func( + decoder_layer.__call__, + hidden_states, + causal_mask, + position_ids, + past_key_values, + output_attentions, + use_cache, + cache_position, + position_embeddings, + ) + else: + layer_outputs = decoder_layer( + hidden_states, + attention_mask=causal_mask, + position_ids=position_ids, + past_key_value=past_key_values, + output_attentions=output_attentions, + use_cache=use_cache, + cache_position=cache_position, + position_embeddings=position_embeddings, + **flash_attn_kwargs, + ) + + hidden_states = layer_outputs[0] + + if output_attentions: + all_self_attns += (layer_outputs[1],) + + hidden_states = self.norm(hidden_states) + + # add hidden states from the last decoder layer + if output_hidden_states: + all_hidden_states += (hidden_states,) + + output = BaseModelOutputWithPast( + last_hidden_state=hidden_states, + past_key_values=past_key_values if use_cache else None, + hidden_states=all_hidden_states, + attentions=all_self_attns, + ) + return output if return_dict else output.to_tuple() + + def _update_causal_mask( + self, + attention_mask: torch.Tensor, + input_tensor: torch.Tensor, + cache_position: torch.Tensor, + past_key_values: Cache, + output_attentions: bool, + ): + if self.config._attn_implementation == "flash_attention_2": + if attention_mask is not None and (attention_mask == 0.0).any(): + return attention_mask + return None + + # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in + # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail + # to infer the attention mask. + past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 + using_static_cache = isinstance(past_key_values, StaticCache) + + # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward + if self.config._attn_implementation == "sdpa" and not using_static_cache and not output_attentions: + if AttentionMaskConverter._ignore_causal_mask_sdpa( + attention_mask, + inputs_embeds=input_tensor, + past_key_values_length=past_seen_tokens, + is_training=self.training, + ): + return None + + dtype, device = input_tensor.dtype, input_tensor.device + sequence_length = input_tensor.shape[1] + if using_static_cache: + target_length = past_key_values.get_max_cache_shape() + else: + target_length = ( + attention_mask.shape[-1] + if isinstance(attention_mask, torch.Tensor) + else past_seen_tokens + sequence_length + 1 + ) + + # In case the provided `attention` mask is 2D, we generate a causal mask here (4D). + causal_mask = self._prepare_4d_causal_attention_mask_with_cache_position( + attention_mask, + sequence_length=sequence_length, + target_length=target_length, + dtype=dtype, + device=device, + cache_position=cache_position, + batch_size=input_tensor.shape[0], + ) + + if ( + self.config._attn_implementation == "sdpa" + and attention_mask is not None + and attention_mask.device.type == "cuda" + and not output_attentions + ): + # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when + # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path. + # Details: https://github.com/pytorch/pytorch/issues/110213 + min_dtype = torch.finfo(dtype).min + causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype) + + return causal_mask + + @staticmethod + def _prepare_4d_causal_attention_mask_with_cache_position( + attention_mask: torch.Tensor, + sequence_length: int, + target_length: int, + dtype: torch.dtype, + device: torch.device, + cache_position: torch.Tensor, + batch_size: int, + **kwargs, + ): + """ + Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape + `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing. + + Args: + attention_mask (`torch.Tensor`): + A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape + `(batch_size, 1, query_length, key_value_length)`. + sequence_length (`int`): + The sequence length being processed. + target_length (`int`): + The target length: when generating with static cache, the mask should be as long as the static cache, + to account for the 0 padding, the part of the cache that is not filled yet. + dtype (`torch.dtype`): + The dtype to use for the 4D attention mask. + device (`torch.device`): + The device to plcae the 4D attention mask on. + cache_position (`torch.Tensor`): + Indices depicting the position of the input sequence tokens in the sequence. + batch_size (`torch.Tensor`): + Batch size. + """ + if attention_mask is not None and attention_mask.dim() == 4: + # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing. + causal_mask = attention_mask + else: + min_dtype = torch.finfo(dtype).min + causal_mask = torch.full( + (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device + ) + if sequence_length != 1: + causal_mask = torch.triu(causal_mask, diagonal=1) + causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1) + causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1) + if attention_mask is not None: + causal_mask = causal_mask.clone() # copy to contiguous memory for in-place edit + mask_length = attention_mask.shape[-1] + padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :] + padding_mask = padding_mask == 0 + causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill( + padding_mask, min_dtype + ) + + return causal_mask + + +class KwargsForCausalLM(FlashAttentionKwargs, LossKwargs): ... + + +class HeliumForCausalLM(HeliumPreTrainedModel, GenerationMixin): + _tied_weights_keys = ["lm_head.weight"] + _tp_plan = {"lm_head": "colwise_rep"} + + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.model = HeliumModel(config) + self.vocab_size = config.vocab_size + self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False) + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.model.embed_tokens + + def set_input_embeddings(self, value): + self.model.embed_tokens = value + + def get_output_embeddings(self): + return self.lm_head + + def set_output_embeddings(self, new_embeddings): + self.lm_head = new_embeddings + + def set_decoder(self, decoder): + self.model = decoder + + def get_decoder(self): + return self.model + + @add_start_docstrings_to_model_forward(HELIUM_INPUTS_DOCSTRING) + @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None, + inputs_embeds: Optional[torch.FloatTensor] = None, + labels: Optional[torch.LongTensor] = None, + use_cache: Optional[bool] = None, + output_attentions: Optional[bool] = None, + output_hidden_states: Optional[bool] = None, + return_dict: Optional[bool] = None, + cache_position: Optional[torch.LongTensor] = None, + num_logits_to_keep: int = 0, + **kwargs: Unpack[KwargsForCausalLM], + ) -> Union[Tuple, CausalLMOutputWithPast]: + r""" + Args: + labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): + Labels for computing the masked language modeling loss. Indices should either be in `[0, ..., + config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored + (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. + + num_logits_to_keep (`int`, *optional*): + Calculate logits for the last `num_logits_to_keep` tokens. If `0`, calculate logits for all + `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that + token can save memory, which becomes pretty significant for long sequences or large vocabulary size. + + Returns: + + Example: + + ```python + >>> from transformers import AutoTokenizer, HeliumForCausalLM + + >>> model = HeliumForCausalLM.from_pretrained("google/helium-7b") + >>> tokenizer = AutoTokenizer.from_pretrained("google/helium-7b") + + >>> prompt = "What is your favorite condiment?" + >>> inputs = tokenizer(prompt, return_tensors="pt") + + >>> # Generate + >>> generate_ids = model.generate(inputs.input_ids, max_length=30) + >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0] + "What is your favorite condiment?" + ```""" + output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions + output_hidden_states = ( + output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states + ) + return_dict = return_dict if return_dict is not None else self.config.use_return_dict + + # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn) + outputs = self.model( + input_ids=input_ids, + attention_mask=attention_mask, + position_ids=position_ids, + past_key_values=past_key_values, + inputs_embeds=inputs_embeds, + use_cache=use_cache, + output_attentions=output_attentions, + output_hidden_states=output_hidden_states, + return_dict=return_dict, + cache_position=cache_position, + **kwargs, + ) + + hidden_states = outputs[0] + # Only compute necessary logits, and do not upcast them to float if we are not computing the loss + logits = self.lm_head(hidden_states[:, -num_logits_to_keep:, :]) + + loss = None + if labels is not None: + loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs) + + if not return_dict: + output = (logits,) + outputs[1:] + return (loss,) + output if loss is not None else output + + return CausalLMOutputWithPast( + loss=loss, + logits=logits, + past_key_values=outputs.past_key_values, + hidden_states=outputs.hidden_states, + attentions=outputs.attentions, + ) + + +@add_start_docstrings( + """ + The Helium Model transformer with a sequence classification head on top (linear layer). + + [`HeliumForSequenceClassification`] uses the last token in order to do the classification, as other causal models + (e.g. GPT-2) do. + + Since it does classification on the last token, it requires to know the position of the last token. If a + `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If + no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the + padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in + each row of the batch). + """, + HELIUM_START_DOCSTRING, +) +class HeliumForSequenceClassification(HeliumPreTrainedModel): + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.num_labels = config.num_labels + self.model = HeliumModel(config) + self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False) + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.model.embed_tokens + + def set_input_embeddings(self, value): + self.model.embed_tokens = value + + @add_start_docstrings_to_model_forward(HELIUM_INPUTS_DOCSTRING) + def forward( + self, + input_ids: Optional[torch.LongTensor] = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None, + inputs_embeds: Optional[torch.FloatTensor] = None, + labels: Optional[torch.LongTensor] = None, + use_cache: Optional[bool] = None, + output_attentions: Optional[bool] = None, + output_hidden_states: Optional[bool] = None, + return_dict: Optional[bool] = None, + ) -> Union[Tuple, SequenceClassifierOutputWithPast]: + r""" + labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): + Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., + config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If + `config.num_labels > 1` a classification loss is computed (Cross-Entropy). + """ + return_dict = return_dict if return_dict is not None else self.config.use_return_dict + + transformer_outputs = self.model( + input_ids, + attention_mask=attention_mask, + position_ids=position_ids, + past_key_values=past_key_values, + inputs_embeds=inputs_embeds, + use_cache=use_cache, + output_attentions=output_attentions, + output_hidden_states=output_hidden_states, + return_dict=return_dict, + ) + hidden_states = transformer_outputs[0] + logits = self.score(hidden_states) + + if input_ids is not None: + batch_size = input_ids.shape[0] + else: + batch_size = inputs_embeds.shape[0] + + if self.config.pad_token_id is None and batch_size != 1: + raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.") + if self.config.pad_token_id is None: + sequence_lengths = -1 + else: + if input_ids is not None: + # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility + sequence_lengths = torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1 + sequence_lengths = sequence_lengths % input_ids.shape[-1] + sequence_lengths = sequence_lengths.to(logits.device) + else: + sequence_lengths = -1 + + pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths] + + loss = None + if labels is not None: + loss = self.loss_function(logits=logits, labels=labels, pooled_logits=pooled_logits, config=self.config) + + if not return_dict: + output = (pooled_logits,) + transformer_outputs[1:] + return ((loss,) + output) if loss is not None else output + + return SequenceClassifierOutputWithPast( + loss=loss, + logits=pooled_logits, + past_key_values=transformer_outputs.past_key_values, + hidden_states=transformer_outputs.hidden_states, + attentions=transformer_outputs.attentions, + ) + + +@add_start_docstrings( + """ + The Helium Model transformer with a token classification head on top (a linear layer on top of the hidden-states + output) e.g. for Named-Entity-Recognition (NER) tasks. + """, + HELIUM_START_DOCSTRING, +) +class HeliumForTokenClassification(HeliumPreTrainedModel): + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.num_labels = config.num_labels + self.model = HeliumModel(config) + if getattr(config, "classifier_dropout", None) is not None: + classifier_dropout = config.classifier_dropout + elif getattr(config, "hidden_dropout", None) is not None: + classifier_dropout = config.hidden_dropout + else: + classifier_dropout = 0.1 + self.dropout = nn.Dropout(classifier_dropout) + self.score = nn.Linear(config.hidden_size, config.num_labels) + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.model.embed_tokens + + def set_input_embeddings(self, value): + self.model.embed_tokens = value + + @add_start_docstrings_to_model_forward(HELIUM_INPUTS_DOCSTRING) + @add_code_sample_docstrings( + checkpoint=_CHECKPOINT_FOR_DOC, + output_type=TokenClassifierOutput, + config_class=_CONFIG_FOR_DOC, + ) + def forward( + self, + input_ids: Optional[torch.LongTensor] = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[List[torch.FloatTensor]] = None, + inputs_embeds: Optional[torch.FloatTensor] = None, + labels: Optional[torch.LongTensor] = None, + use_cache: Optional[bool] = None, + output_attentions: Optional[bool] = None, + output_hidden_states: Optional[bool] = None, + return_dict: Optional[bool] = None, + ) -> Union[Tuple, TokenClassifierOutput]: + r""" + labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): + Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., + config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If + `config.num_labels > 1` a classification loss is computed (Cross-Entropy). + """ + return_dict = return_dict if return_dict is not None else self.config.use_return_dict + + outputs = self.model( + input_ids, + attention_mask=attention_mask, + position_ids=position_ids, + past_key_values=past_key_values, + inputs_embeds=inputs_embeds, + use_cache=use_cache, + output_attentions=output_attentions, + output_hidden_states=output_hidden_states, + return_dict=return_dict, + ) + sequence_output = outputs[0] + sequence_output = self.dropout(sequence_output) + logits = self.score(sequence_output) + + loss = None + if labels is not None: + loss = self.loss_function(logits, labels, self.config) + + if not return_dict: + output = (logits,) + outputs[2:] + return ((loss,) + output) if loss is not None else output + + return TokenClassifierOutput( + loss=loss, + logits=logits, + hidden_states=outputs.hidden_states, + attentions=outputs.attentions, + ) + + +__all__ = [ + "HeliumPreTrainedModel", + "HeliumModel", + "HeliumForCausalLM", + "HeliumForSequenceClassification", + "HeliumForTokenClassification", +] diff --git a/src/transformers/models/helium/modular_helium.py b/src/transformers/models/helium/modular_helium.py new file mode 100644 index 0000000000..0c9be5ec80 --- /dev/null +++ b/src/transformers/models/helium/modular_helium.py @@ -0,0 +1,171 @@ +# coding=utf-8 +# Copyright 2024 The Kyutai and HuggingFace Inc. teams. 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. +import math +from typing import Optional + +import torch +import torch.nn as nn +import torch.utils.checkpoint + +from ...utils import logging +from ..gemma.modeling_gemma import ( + GemmaForCausalLM, + GemmaForSequenceClassification, + GemmaForTokenClassification, +) +from ..granite.modeling_granite import ( + GraniteAttention, +) +from ..llama.modeling_llama import ( + LlamaDecoderLayer, + LlamaMLP, + LlamaModel, + LlamaPreTrainedModel, + LlamaRotaryEmbedding, +) +from .configuration_helium import HeliumConfig + + +logger = logging.get_logger(__name__) + + +class HeliumRMSNorm(nn.Module): + def __init__(self, hidden_size, eps=1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(hidden_size)) + self.variance_epsilon = eps + + def forward(self, hidden_states): + input_dtype = hidden_states.dtype + hidden_states = hidden_states.to(torch.float32) + variance = hidden_states.pow(2).mean(-1, keepdim=True) + hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon) + return (self.weight.to(torch.float32) * hidden_states).to(input_dtype) + + def extra_repr(self): + return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}" + + +class HeliumRotaryEmbedding(LlamaRotaryEmbedding): + pass + + +class HeliumMLP(LlamaMLP): + pass + + +def rotate_half(x): + """Rotates half the hidden dims of the input.""" + x1 = x[..., 0::2] + x2 = x[..., 1::2] + return torch.stack((-x2, x1), dim=-1).flatten(-2) + + +def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1): + """Applies Rotary Position Embedding to the query and key tensors. + + Args: + q (`torch.Tensor`): The query tensor. + k (`torch.Tensor`): The key tensor. + cos (`torch.Tensor`): The cosine part of the rotary embedding. + sin (`torch.Tensor`): The sine part of the rotary embedding. + position_ids (`torch.Tensor`, *optional*): + Deprecated and unused. + unsqueeze_dim (`int`, *optional*, defaults to 1): + The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and + sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note + that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and + k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes + cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have + the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2. + Returns: + `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding. + """ + cos = cos.unsqueeze(unsqueeze_dim) + sin = sin.unsqueeze(unsqueeze_dim) + + # Interleave them instead of usual shape + cos = cos[..., : cos.shape[-1] // 2].repeat_interleave(2, dim=-1) + sin = sin[..., : sin.shape[-1] // 2].repeat_interleave(2, dim=-1) + + q_embed = (q * cos) + (rotate_half(q) * sin) + k_embed = (k * cos) + (rotate_half(k) * sin) + + return q_embed, k_embed + + +class HeliumAttention(GraniteAttention): + def __init__(self, config: HeliumConfig, layer_idx: Optional[int] = None): + super().__init__(config, layer_idx) + self.o_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False) + self.scaling = 1 / math.sqrt(self.head_dim) + + +class HeliumDecoderLayer(LlamaDecoderLayer): + def __init__(self, config: HeliumConfig, layer_idx: Optional[int] = None): + super().__init__() + + self.mlp = HeliumMLP(config) + self.input_layernorm = HeliumRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.post_attention_layernorm = HeliumRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + + +class HeliumPreTrainedModel(LlamaPreTrainedModel): + pass + + +class HeliumModel(HeliumPreTrainedModel, LlamaModel): + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.layers = nn.ModuleList( + [HeliumDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] + ) + self.norm = HeliumRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.rotary_emb = HeliumRotaryEmbedding(config) + self.gradient_checkpointing = False + + # Initialize weights and apply final processing + self.post_init() + + +class HeliumForCausalLM(GemmaForCausalLM): + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.model = HeliumModel(config) + self.post_init() + + +class HeliumForSequenceClassification(GemmaForSequenceClassification): + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.model = HeliumModel(config) + self.post_init() + + +class HeliumForTokenClassification(GemmaForTokenClassification): + def __init__(self, config: HeliumConfig): + super().__init__(config) + self.model = HeliumModel(config) + self.post_init() + + +__all__ = [ + "HeliumPreTrainedModel", + "HeliumModel", + "HeliumForCausalLM", + "HeliumForSequenceClassification", + "HeliumForTokenClassification", +] diff --git a/src/transformers/utils/dummy_pt_objects.py b/src/transformers/utils/dummy_pt_objects.py index 6a9cd232eb..bac6220a71 100644 --- a/src/transformers/utils/dummy_pt_objects.py +++ b/src/transformers/utils/dummy_pt_objects.py @@ -4981,6 +4981,41 @@ class GroupViTVisionModel(metaclass=DummyObject): requires_backends(self, ["torch"]) +class HeliumForCausalLM(metaclass=DummyObject): + _backends = ["torch"] + + def __init__(self, *args, **kwargs): + requires_backends(self, ["torch"]) + + +class HeliumForSequenceClassification(metaclass=DummyObject): + _backends = ["torch"] + + def __init__(self, *args, **kwargs): + requires_backends(self, ["torch"]) + + +class HeliumForTokenClassification(metaclass=DummyObject): + _backends = ["torch"] + + def __init__(self, *args, **kwargs): + requires_backends(self, ["torch"]) + + +class HeliumModel(metaclass=DummyObject): + _backends = ["torch"] + + def __init__(self, *args, **kwargs): + requires_backends(self, ["torch"]) + + +class HeliumPreTrainedModel(metaclass=DummyObject): + _backends = ["torch"] + + def __init__(self, *args, **kwargs): + requires_backends(self, ["torch"]) + + class HieraBackbone(metaclass=DummyObject): _backends = ["torch"] diff --git a/tests/models/helium/__init__.py b/tests/models/helium/__init__.py new file mode 100644 index 0000000000..e69de29bb2 diff --git a/tests/models/helium/test_modeling_helium.py b/tests/models/helium/test_modeling_helium.py new file mode 100644 index 0000000000..3ad2cf7366 --- /dev/null +++ b/tests/models/helium/test_modeling_helium.py @@ -0,0 +1,110 @@ +# coding=utf-8 +# Copyright 2024 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 Helium model.""" + +import unittest + +from transformers import AutoModelForCausalLM, AutoTokenizer, HeliumConfig, is_torch_available +from transformers.testing_utils import ( + require_read_token, + require_torch, + slow, + torch_device, +) + +from ...test_configuration_common import ConfigTester +from ..gemma.test_modeling_gemma import GemmaModelTest, GemmaModelTester + + +if is_torch_available(): + import torch + + from transformers import ( + HeliumForCausalLM, + HeliumForSequenceClassification, + HeliumForTokenClassification, + HeliumModel, + ) + + +class HeliumModelTester(GemmaModelTester): + if is_torch_available(): + config_class = HeliumConfig + model_class = HeliumModel + for_causal_lm_class = HeliumForCausalLM + for_sequence_class = HeliumForSequenceClassification + for_token_class = HeliumForTokenClassification + + +@require_torch +class HeliumModelTest(GemmaModelTest, unittest.TestCase): + all_model_classes = ( + (HeliumModel, HeliumForCausalLM, HeliumForSequenceClassification, HeliumForTokenClassification) + if is_torch_available() + else () + ) + all_generative_model_classes = (HeliumForCausalLM,) if is_torch_available() else () + pipeline_model_mapping = ( + { + "feature-extraction": HeliumModel, + "text-classification": HeliumForSequenceClassification, + "token-classification": HeliumForTokenClassification, + "text-generation": HeliumForCausalLM, + "zero-shot": HeliumForSequenceClassification, + } + if is_torch_available() + else {} + ) + test_headmasking = False + test_pruning = False + _is_stateful = True + model_split_percents = [0.5, 0.6] + + def setUp(self): + self.model_tester = HeliumModelTester(self) + self.config_tester = ConfigTester(self, config_class=HeliumConfig, hidden_size=37) + + +@slow +# @require_torch_gpu +class HeliumIntegrationTest(unittest.TestCase): + input_text = ["Hello, today is a great day to"] + # This variable is used to determine which CUDA device are we using for our runners (A10 or T4) + # Depending on the hardware we get different logits / generations + cuda_compute_capability_major_version = None + + @classmethod + def setUpClass(cls): + if is_torch_available() and torch.cuda.is_available(): + # 8 is for A100 / A10 and 7 for T4 + cls.cuda_compute_capability_major_version = torch.cuda.get_device_capability()[0] + + @require_read_token + def test_model_2b(self): + model_id = "kyutai/helium-1-preview" + EXPECTED_TEXTS = [ + "Hello, today is a great day to start a new project. I have been working on a new project for a while now and I have" + ] + + model = AutoModelForCausalLM.from_pretrained( + model_id, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16, revision="refs/pr/1" + ).to(torch_device) + tokenizer = AutoTokenizer.from_pretrained(model_id, revision="refs/pr/1") + inputs = tokenizer(self.input_text, return_tensors="pt", padding=True).to(torch_device) + + output = model.generate(**inputs, max_new_tokens=20, do_sample=False) + output_text = tokenizer.batch_decode(output, skip_special_tokens=True) + + self.assertEqual(output_text, EXPECTED_TEXTS)