# coding=utf-8
# Copyright 2025 The Qwen team, Alibaba Group and 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.
import inspect
import unittest

import numpy as np
from huggingface_hub import hf_hub_download
from parameterized import parameterized

from transformers import (
    Qwen3OmniMoeProcessor,
)
from transformers.testing_utils import (
    require_av,
    require_librosa,
    require_torch,
    require_torchaudio,
    require_torchvision,
    require_vision,
)
from transformers.utils import is_torch_available

from ...test_processing_common import ProcessorTesterMixin, url_to_local_path


if is_torch_available():
    import torch


@require_vision
@require_torch
@require_torchaudio
@require_torchvision
class Qwen3OmniMoeProcessorTest(ProcessorTesterMixin, unittest.TestCase):
    processor_class = Qwen3OmniMoeProcessor
    model_id = "Qwen/Qwen2.5-Omni-7B"

    @classmethod
    def _setup_image_processor(cls):
        image_processor_class = cls._get_component_class_from_processor("image_processor")
        return image_processor_class.from_pretrained(
            cls.model_id, size={"shortest_edge": 28 * 28, "longest_edge": 56 * 56}
        )

    @classmethod
    def _setup_video_processor(cls):
        video_processor_class = cls._get_component_class_from_processor("video_processor")
        return video_processor_class.from_pretrained(
            cls.model_id, size={"shortest_edge": 28 * 28, "longest_edge": 56 * 56}
        )

    def prepare_audio_inputs(self, batch_size: int = 3):
        """This function prepares a list of numpy audios."""
        audio_inputs = [np.random.rand(160000) * 2 - 1] * batch_size
        return audio_inputs

    @require_torch
    def _test_apply_chat_template(
        self,
        modality: str,
        batch_size: int,
        return_tensors: str,
        input_name: str,
        processor_name: str,
        input_data: list[str],
    ):
        processor = self.get_processor()
        if processor.chat_template is None:
            self.skipTest("Processor has no chat template")

        if processor_name not in self.processor_class.get_attributes():
            self.skipTest(f"{processor_name} attribute not present in {self.processor_class}")

        batch_messages = [
            [
                {
                    "role": "user",
                    "content": [{"type": "text", "text": "Describe this."}],
                },
            ]
        ] * batch_size

        # Test that jinja can be applied
        formatted_prompt = processor.apply_chat_template(batch_messages, add_generation_prompt=True, tokenize=False)
        self.assertEqual(len(formatted_prompt), batch_size)

        # Test that tokenizing with template and directly with `self.tokenizer` gives same output
        formatted_prompt_tokenized = processor.apply_chat_template(
            batch_messages, add_generation_prompt=True, tokenize=True, return_tensors=return_tensors
        )
        add_special_tokens = True
        if processor.tokenizer.bos_token is not None and formatted_prompt[0].startswith(processor.tokenizer.bos_token):
            add_special_tokens = False
        tok_output = processor.tokenizer(
            formatted_prompt, return_tensors=return_tensors, add_special_tokens=add_special_tokens
        )
        expected_output = tok_output.input_ids
        self.assertListEqual(expected_output.tolist(), formatted_prompt_tokenized.tolist())

        # Test that kwargs passed to processor's `__call__` are actually used
        tokenized_prompt_100 = processor.apply_chat_template(
            batch_messages,
            add_generation_prompt=True,
            tokenize=True,
            padding="max_length",
            truncation=True,
            return_tensors=return_tensors,
            max_length=100,
        )
        self.assertEqual(len(tokenized_prompt_100[0]), 100)

        # Test that `return_dict=True` returns text related inputs in the dict
        out_dict_text = processor.apply_chat_template(
            batch_messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors=return_tensors,
        )
        self.assertTrue(all(key in out_dict_text for key in ["input_ids", "attention_mask"]))
        self.assertEqual(len(out_dict_text["input_ids"]), batch_size)
        self.assertEqual(len(out_dict_text["attention_mask"]), batch_size)

        # Test that with modality URLs and `return_dict=True`, we get modality inputs in the dict
        for idx, url in enumerate(input_data[:batch_size]):
            batch_messages[idx][0]["content"] = [batch_messages[idx][0]["content"][0], {"type": modality, "url": url}]

        out_dict = processor.apply_chat_template(
            batch_messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors=return_tensors,
            num_frames=2,  # by default no more than 2 frames, otherwise too slow
        )
        input_name = getattr(self, input_name)
        self.assertTrue(input_name in out_dict)
        self.assertEqual(len(out_dict["input_ids"]), batch_size)
        self.assertEqual(len(out_dict["attention_mask"]), batch_size)

        if modality == "video":
            # qwen pixels don't scale with bs same way as other models, calculate expected video token count based on video_grid_thw
            expected_video_token_count = 0
            for thw in out_dict["video_grid_thw"]:
                expected_video_token_count += thw[0] * thw[1] * thw[2]
            mm_len = expected_video_token_count
        elif modality == "audio":
            mm_len = batch_size
        else:
            mm_len = batch_size * 1200
        self.assertEqual(len(out_dict[input_name]), mm_len)

        return_tensor_to_type = {"pt": torch.Tensor, "np": np.ndarray, None: list}
        for k in out_dict:
            self.assertIsInstance(out_dict[k], return_tensor_to_type[return_tensors])

    @unittest.skip("Skipping but this one is important, should be fixed ASAP")
    @parameterized.expand([(1, "pt"), (2, "pt")])
    def test_apply_chat_template_image(self, batch_size: int, return_tensors: str):
        pass

    @require_av
    def test_apply_chat_template_video_frame_sampling(self):
        processor = self.get_processor()
        if processor.chat_template is None:
            self.skipTest("Processor has no chat template")

        signature = inspect.signature(processor.__call__)
        if "videos" not in {*signature.parameters.keys()} or (
            signature.parameters.get("videos") is not None
            and signature.parameters["videos"].annotation == inspect._empty
        ):
            self.skipTest("Processor doesn't accept videos at input")

        messages = [
            [
                {
                    "role": "user",
                    "content": [
                        {"type": "text", "text": "What is shown in this video?"},
                    ],
                },
            ]
        ]

        formatted_prompt = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
        self.assertEqual(len(formatted_prompt), 1)

        formatted_prompt_tokenized = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True)
        expected_output = processor.tokenizer(formatted_prompt, return_tensors=None).input_ids
        self.assertListEqual(expected_output, formatted_prompt_tokenized)

        out_dict = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True)
        self.assertListEqual(list(out_dict.keys()), ["input_ids", "attention_mask"])

        # Add video URL for return dict and load with `num_frames` arg
        messages[0][0]["content"].append(
            {
                "type": "video",
                "url": url_to_local_path(
                    "https://huggingface.co/datasets/raushan-testing-hf/videos-test/resolve/main/tiny_video.mp4"
                ),
            }
        )
        num_frames = 3
        out_dict_with_video = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            num_frames=num_frames,
        )
        self.assertTrue(self.videos_input_name in out_dict_with_video)
        self.assertEqual(len(out_dict_with_video[self.videos_input_name]), 7728)

        # Load with `fps` arg
        fps = 1
        out_dict_with_video = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            fps=fps,
        )
        self.assertTrue(self.videos_input_name in out_dict_with_video)
        self.assertEqual(len(out_dict_with_video[self.videos_input_name]), 7728)

        # Load with `fps` and `num_frames` args, should raise an error
        with self.assertRaises(ValueError):
            out_dict_with_video = processor.apply_chat_template(
                messages,
                add_generation_prompt=True,
                tokenize=True,
                return_dict=True,
                fps=fps,
                num_frames=num_frames,
            )

        # Load without any arg should load the whole video
        out_dict_with_video = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
        )
        self.assertTrue(self.videos_input_name in out_dict_with_video)
        self.assertEqual(len(out_dict_with_video[self.videos_input_name]), 23184)

        # Load video as a list of frames (i.e. images). NOTE: each frame should have same size
        # because we assume they come from one video
        messages[0][0]["content"][-1] = {
            "type": "video",
            "url": [
                "https://www.ilankelman.org/stopsigns/australia.jpg",
                "https://www.ilankelman.org/stopsigns/australia.jpg",
            ],
        }
        out_dict_with_video = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
        )
        self.assertTrue(self.videos_input_name in out_dict_with_video)
        self.assertEqual(len(out_dict_with_video[self.videos_input_name]), 7600)

        # When the inputs are frame URLs/paths we expect that those are already
        # sampled and will raise an error is asked to sample again.
        with self.assertRaises(ValueError):
            out_dict_with_video = processor.apply_chat_template(
                messages,
                add_generation_prompt=True,
                tokenize=True,
                return_dict=True,
                do_sample_frames=True,
                num_frames=num_frames,
            )

    @require_librosa
    @require_av
    def test_chat_template_audio_from_video(self):
        processor = self.get_processor()
        if processor.chat_template is None:
            self.skipTest("Processor has no chat template")

        signature = inspect.signature(processor.__call__)
        if "videos" not in {*signature.parameters.keys()} or (
            signature.parameters.get("videos") is not None
            and signature.parameters["videos"].annotation == inspect._empty
        ):
            self.skipTest(f"{self.processor_class} does not support video inputs")

        if "feature_extractor" not in self.processor_class.get_attributes():
            self.skipTest(f"feature_extractor attribute not present in {self.processor_class}")

        video_file_path = hf_hub_download(
            repo_id="raushan-testing-hf/videos-test", filename="sample_demo_1.mp4", repo_type="dataset"
        )
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "video", "path": video_file_path},
                    {"type": "text", "text": "Which of these animals is making the sound?"},
                ],
            },
            {
                "role": "assistant",
                "content": [{"type": "text", "text": "It is a cow."}],
            },
            {
                "role": "user",
                "content": [
                    {"type": "text", "text": "Tell me all about this animal."},
                ],
            },
        ]

        formatted_prompt = processor.apply_chat_template([messages], add_generation_prompt=True, tokenize=False)
        self.assertEqual(len(formatted_prompt), 1)  # batch size=1

        out_dict = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors="pt",
            load_audio_from_video=True,
        )
        self.assertTrue(self.audio_input_name in out_dict)
        self.assertTrue(self.videos_input_name in out_dict)

        # should always have input_ids and attention_mask
        self.assertEqual(len(out_dict["input_ids"]), 1)  # batch-size=1
        self.assertEqual(len(out_dict["attention_mask"]), 1)  # batch-size=1
        self.assertEqual(len(out_dict[self.audio_input_name]), 1)  # 1 audio in the conversation
        self.assertEqual(len(out_dict[self.videos_input_name]), 145912)  # 1 video in the conversation