LLaMA-Factory/tests/model/model_utils/test_attention.py

# Copyright 2024 the LlamaFactory team.
#
# 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 os

from transformers.utils import is_flash_attn_2_available, is_torch_sdpa_available

from llamafactory.hparams import get_infer_args
from llamafactory.model import load_model, load_tokenizer


TINY_LLAMA = os.environ.get("TINY_LLAMA", "llamafactory/tiny-random-Llama-3")

INFER_ARGS = {
    "model_name_or_path": TINY_LLAMA,
    "template": "llama3",
}


def test_attention():
    attention_available = ["disabled"]
    if is_torch_sdpa_available():
        attention_available.append("sdpa")

    if is_flash_attn_2_available():
        attention_available.append("fa2")

    llama_attention_classes = {
        "disabled": "LlamaAttention",
        "sdpa": "LlamaSdpaAttention",
        "fa2": "LlamaFlashAttention2",
    }
    for requested_attention in attention_available:
        model_args, _, finetuning_args, _ = get_infer_args({"flash_attn": requested_attention, **INFER_ARGS})
        tokenizer_module = load_tokenizer(model_args)
        model = load_model(tokenizer_module["tokenizer"], model_args, finetuning_args)
        for module in model.modules():
            if "Attention" in module.__class__.__name__:
                assert module.__class__.__name__ == llama_attention_classes[requested_attention]