import argparse import os import torch from torch.nn.attention import sdpa_kernel, SDPBackend from transformers import AutoConfig, AutoModel, AutoModelForCTC, AutoProcessor, MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING import evaluate from normalizer import data_utils import time from tqdm import tqdm wer_metric = evaluate.load("wer") torch.set_float32_matmul_precision('high') def main(args): model = AutoModel.from_pretrained(args.model_id, torch_dtype=torch.float16, trust_remote_code=True, force_download=True).to(args.device) print(f"Model size: {sum(p.numel() for p in model.parameters()) / 1e9:.2f}B parameters") processor = AutoProcessor.from_pretrained("openai/whisper-large-v3-turbo", force_download=True) model_input_name = processor.model_input_names[0] if model.can_generate(): gen_kwargs = {"max_new_tokens": 224} elif args.max_new_tokens: raise ValueError("`max_new_tokens` should only be set for auto-regressive models, but got a CTC model.") if args.torch_compile: model.forward = torch.compile(model.forward, mode=args.compile_mode, fullgraph=True) if model.can_generate(): # enable static k/v cache for autoregressive models model.generation_config.cache_implementation = "static" def benchmark(batch, min_new_tokens=None): # Load audio inputs audios = [audio["array"] for audio in batch["audio"]] minibatch_size = len(audios) batch["audio_length_s"] = [len(audio) / 16000 for audio in audios] batch["audio_filepath"] = data_utils.extract_audio_filepaths_from_batch(batch, minibatch_size) # START TIMING start_time = time.time() # 1. Pre-Processing # 1.1 Pad audios to max batch size if using torch compile to prevent re-compilations padding_size = None if minibatch_size != args.batch_size and args.torch_compile: padding_size = args.batch_size - minibatch_size padding_audios = [audios[-1] for _ in range(padding_size)] audios.extend(padding_audios) if not model.can_generate(): #or len(audios[0]) > processor.feature_extractor.n_samples: # 1.2 Either CTC pre-processing (normalize to mean 0, std 1), or long-form Whisper processing inputs = processor( audios, sampling_rate=16_000, truncation=False, padding="longest", return_tensors="pt", return_attention_mask=True, ) else: # 1.3 Standard Whisper processing: pad audios to 30-seconds and converted to log-mel inputs = processor(audios, sampling_rate=16_000, return_tensors="pt", device=args.device) inputs = inputs.to(args.device) inputs[model_input_name] = inputs[model_input_name].to(torch.float16) # 2. Model Inference with sdpa_kernel(SDPBackend.MATH if args.torch_compile else SDPBackend.FLASH_ATTENTION): forced_decoder_ids = processor.get_decoder_prompt_ids(language="english", task="transcribe") if model.can_generate(): # 2.1 Auto-regressive generation for encoder-decoder models # Set forced_decoder_ids on generation_config (not as kwarg) for newer transformers model.generation_config.forced_decoder_ids = forced_decoder_ids pred_ids = model.generate(**inputs, **gen_kwargs, min_new_tokens=min_new_tokens) else: # 2.2. Single forward pass for CTC with torch.no_grad(): logits = model(**inputs).logits pred_ids = logits.argmax(-1) # 3. Post-processing # 3.1 Strip padded ids from predictions if padding_size is not None: pred_ids = pred_ids[:-padding_size, ...] # 3.2 Convert token ids to text transcription pred_text = processor.batch_decode(pred_ids, skip_special_tokens=True) # END TIMING runtime = time.time() - start_time # normalize by minibatch size since we want the per-sample time batch["transcription_time_s"] = minibatch_size * [runtime / minibatch_size] batch["predictions"] = pred_text batch["references"] = batch["original_text"] return batch if args.warmup_steps is not None: dataset = data_utils.load_data(args) dataset = data_utils.prepare_data(dataset) num_warmup_samples = args.warmup_steps * args.batch_size if args.streaming: warmup_dataset = dataset.take(num_warmup_samples) else: warmup_dataset = dataset.select(range(min(num_warmup_samples, len(dataset)))) warmup_dataset = iter(warmup_dataset.map(benchmark, batch_size=args.batch_size, batched=True, fn_kwargs={"min_new_tokens": args.max_new_tokens})) for _ in tqdm(warmup_dataset, desc="Warming up..."): continue dataset = data_utils.load_data(args) if args.max_eval_samples is not None and args.max_eval_samples > 0: print(f"Subsampling dataset to first {args.max_eval_samples} samples!") if args.streaming: dataset = dataset.take(args.max_eval_samples) else: dataset = dataset.select(range(min(args.max_eval_samples, len(dataset)))) dataset = data_utils.prepare_data(dataset) dataset = dataset.map( benchmark, batch_size=args.batch_size, batched=True, remove_columns=["audio"], ) all_results = { "audio_length_s": [], "transcription_time_s": [], "predictions": [], "references": [], "audio_filepath": [], } result_iter = iter(dataset) for result in tqdm(result_iter, desc="Samples..."): for key in all_results: all_results[key].append(result[key]) # Write manifest results (WER and RTFX) manifest_path = data_utils.write_manifest( all_results["references"], all_results["predictions"], args.model_id, args.dataset_path, args.dataset, args.split, audio_length=all_results["audio_length_s"], transcription_time=all_results["transcription_time_s"], audio_filepaths=all_results["audio_filepath"], ) print("Results saved at path:", os.path.abspath(manifest_path)) norm_refs = [data_utils.normalizer(r) for r in all_results["references"]] norm_preds = [data_utils.normalizer(p) for p in all_results["predictions"]] wer = wer_metric.compute( references=norm_refs, predictions=norm_preds ) wer = round(100 * wer, 2) rtfx = round(sum(all_results["audio_length_s"]) / sum(all_results["transcription_time_s"]), 2) print("WER:", wer, "%", "RTFx:", rtfx) if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument( "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers", ) parser.add_argument( "--dataset_path", type=str, default="hf-audio/open-asr-leaderboard", help="Dataset path. By default, it is `hf-audio/open-asr-leaderboard`", ) parser.add_argument( "--dataset", type=str, required=True, help="Dataset name. *E.g.* `'librispeech_asr` for the LibriSpeech ASR dataset, or `'common_voice'` for Common Voice. The full list of dataset names " "can be found at `https://huggingface.co/datasets/hf-audio/open-asr-leaderboard`", ) parser.add_argument( "--split", type=str, default="test", help="Split of the dataset. *E.g.* `'validation'` for the dev split, or `'test'` for the test split.", ) parser.add_argument( "--device", type=int, default=-1, help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.", ) parser.add_argument( "--batch_size", type=int, default=16, help="Number of samples to go through each streamed batch.", ) parser.add_argument( "--max_eval_samples", type=int, default=None, help="Number of samples to be evaluated. Put a lower number e.g. 64 for testing this script.", ) parser.add_argument( "--streaming", action="store_true", help="Stream the dataset lazily over the network instead of downloading it in full before the evaluation. Off by default for reproducible benchmark timings.", ) parser.add_argument( "--max_new_tokens", type=int, default=None, help="Maximum number of tokens to generate (for auto-regressive models).", ) parser.add_argument( "--torch_compile", action="store_true", help="Whether to JIT compile the forward pass of the model.", ) parser.add_argument( "--compile_mode", type=str, default="max-autotune", help="Mode for torch compiling model forward pass. Can be either 'default', 'reduce-overhead', 'max-autotune' or 'max-autotune-no-cudagraphs'.", ) parser.add_argument( "--warmup_steps", type=int, default=10, help="Number of warm-up steps to run before launching the timed runs.", ) args = parser.parse_args() main(args)