| | |
|---|
| | | deep_update(model_conf, kwargs.get("model_conf", {})) |
|---|
| | | deep_update(model_conf, kwargs) |
|---|
| | | model = model_class(**model_conf, vocab_size=vocab_size) |
|---|
| | | model.to(device) |
|---|
| | | |
|---|
| | | # init_param |
|---|
| | | init_param = kwargs.get("init_param", None) |
|---|
| | |
|---|
| | | # fp16 |
|---|
| | | if kwargs.get("fp16", False): |
|---|
| | | model.to(torch.float16) |
|---|
| | | elif kwargs.get("bf16", False): |
|---|
| | | model.to(torch.bfloat16) |
|---|
| | | model.to(device) |
|---|
| | | return model, kwargs |
|---|
| | | |
|---|
| | | def __call__(self, *args, **cfg): |
|---|
| | |
|---|
| | | # f"time_speech_total_per_sample: {time_speech_total_per_sample: 0.3f}, " |
|---|
| | | # f"time_escape_total_per_sample: {time_escape_total_per_sample:0.3f}") |
|---|
| | | |
|---|
| | | if len(results_sorted) != n: |
|---|
| | | results_ret_list.append({"key": key, "text": "", "timestamp": []}) |
|---|
| | | logging.info("decoding, utt: {}, empty result".format(key)) |
|---|
| | | continue |
|---|
| | | restored_data = [0] * n |
|---|
| | | for j in range(n): |
|---|
| | | index = sorted_data[j][1] |
|---|
