Add GenAI ASRPipeline evaluator

[as-suvorov]

• Add GenAI ASRPipeline evaluator
• Add Qwen3ASROptimumPipeline evaluator
• Fix Whisper config for optimum and hf - { num_beams: 1 }. Which aligned with GenAI configuration.

Qwen3-ASR deps requirements:
pip install --no-deps git+https://github.com/huggingface/optimum-intel.git@f48d93fddff8c91e198389c47a6d5974789b67f4
pip install qwen-asr (requires transformers==4.57.6)

Example config.json:

evaluations:
  - name: qwen3-asr-0.6b-genai
    module: custom_evaluators.asr_pipeline_evaluator.ASRPipelineEvaluator
    module_config:
      pipeline_class: GenAIASRPipeline
      launchers:
        - framework: openvino
          device: CPU
      datasets:
        - name: LibriSpeech_test_clean_flac_max_duration_15
          data_source: /home/asuvorov/projects/open_model_zoo/.venv/data/librispeech/test/LibriSpeech/test-clean
          annotation_conversion:
            converter: librispeech
            data_dir: /home/asuvorov/projects/open_model_zoo/.venv/data/librispeech/test/LibriSpeech/test-clean
            annotation_file: /home/asuvorov/projects/open_model_zoo/librispeech_test_clean_max_duration_15_10.jsonl
            use_flac: True
            max_duration: 15
            top_n: 10
          reader:
            type: flac_reader
            mono: True
            dtype: float64
          metrics:
            - type: wer

  - name: qwen3-asr-0.6b-optimum
    module: custom_evaluators.asr_pipeline_evaluator.ASRPipelineEvaluator
    module_config:
      pipeline_class: Qwen3ASROptimumPipeline
      max_new_tokens: 1000
      launchers:
        - framework: openvino
          device: CPU
      datasets:
        - name: LibriSpeech_test_clean_flac_max_duration_15
          data_source: /home/asuvorov/projects/open_model_zoo/.venv/data/librispeech/test/LibriSpeech/test-clean
          annotation_conversion:
            converter: librispeech
            data_dir: /home/asuvorov/projects/open_model_zoo/.venv/data/librispeech/test/LibriSpeech/test-clean
            annotation_file: /home/asuvorov/projects/open_model_zoo/librispeech_test_clean_max_duration_15_10.jsonl
            use_flac: True
            max_duration: 15
            top_n: 10
          reader:
            type: flac_reader
            mono: True
            dtype: float64
          metrics:
            - type: wer

accuracy_check -c checker_config_qwen3_asr_local.yaml -m /home/asuvorov/projects/openvino.genai/.vscode/models/Qwen/Qwen3-ASR-0.6B

Processing info:
model: qwen3-asr-0.6b-genai
launcher: openvino
device: CPU
dataset: LibriSpeech_test_clean_flac_max_duration_15
OpenCV version: 4.13.0
Annotation conversion for LibriSpeech_test_clean_flac_max_duration_15 dataset has been started
Parameters to be used for conversion:
converter: librispeech
data_dir: /home/asuvorov/projects/open_model_zoo/.venv/data/librispeech/test/LibriSpeech/test-clean
annotation_file: /home/asuvorov/projects/open_model_zoo/librispeech_test_clean_max_duration_15_10.jsonl
use_flac: True
max_duration: 15
top_n: 10
Annotation conversion for LibriSpeech_test_clean_flac_max_duration_15 dataset has been finished
10 objects processed in 13.674 seconds                                                                                                                                                                                             
wer: 3.57%
Processing info:
model: qwen3-asr-0.6b-optimum
launcher: openvino
device: CPU
dataset: LibriSpeech_test_clean_flac_max_duration_15
OpenCV version: 4.13.0
The tokenizer you are loading from '/home/asuvorov/projects/openvino.genai/.vscode/models/Qwen/Qwen3-ASR-0.6B' with an incorrect regex pattern: https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503/discussions/84#69121093e8b480e709447d5e. This will lead to incorrect tokenization. You should set the `fix_mistral_regex=True` flag when loading this tokenizer to fix this issue.
Annotation conversion for LibriSpeech_test_clean_flac_max_duration_15 dataset has been started
Parameters to be used for conversion:
converter: librispeech
data_dir: /home/asuvorov/projects/open_model_zoo/.venv/data/librispeech/test/LibriSpeech/test-clean
annotation_file: /home/asuvorov/projects/open_model_zoo/librispeech_test_clean_max_duration_15_10.jsonl
use_flac: True
max_duration: 15
top_n: 10
Annotation conversion for LibriSpeech_test_clean_flac_max_duration_15 dataset has been finished
  0%|                                                                                                                                                                                                              | 0/10 [00:00<?]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
 10%|████████████████████▏                                                                                                                                                                                     | 1/10 [00:03<00:32]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 20%|████████████████████████████████████████▍                                                                                                                                                                 | 2/10 [00:04<00:16]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 30%|████████████████████████████████████████████████████████████▌                                                                                                                                             | 3/10 [00:06<00:12]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 40%|████████████████████████████████████████████████████████████████████████████████▊                                                                                                                         | 4/10 [00:06<00:08]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 50%|█████████████████████████████████████████████████████████████████████████████████████████████████████                                                                                                     | 5/10 [00:08<00:06]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 60%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏                                                                                | 6/10 [00:10<00:06]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 70%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▍                                                            | 7/10 [00:12<00:05]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 80%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▌                                        | 8/10 [00:12<00:02]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
 90%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▊                    | 9/10 [00:14<00:01]The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:151643 for open-end generation.
10 objects processed in 16.451 seconds                                                                                                                                                                                             
wer: 3.57%