busishengui

程序未报错，但是wer结果100%

17

我采用miniLibrispeech作为训练和测试语音数据集进行测试，使用example中的tdnn作为训练模型，整个run的流程并未报错，但是最终wer结果为100%。 $WER 100.00%[20138/20138,0 ins, 20138 del, 0sub]

256G mem is Not Enough （AWQ 4bit LLama 70b）

2

v 0.6.1 ```bash python quantize.py --model_dir ./hg_weight_3999/ --dtype float16 --qformat int4_awq --export_path ./quantized_int4-awq --calib_size 32 ``` ```log Using pad_token, but it is not set yet. [12/15/2023-01:48:38] The model weights are...

在进行推理的时候出现“Error: param is nil”

2

**1. 环境（environment）** - Build OS and Version: Windows - RunTime OS Version: Windows - RunTime DEVICE: COU **2. Github版本** 当前版本 **3. 编译方式(compile method)** CMake完整编译参数(full cmake arguments 默认参数 **4. 编译日志(build log)**...

A bug occurs when the program terminates

7

### Describe the issue It works well when it run in GPU，but it has a bug when it terminates terminate called after throwing an instance of 'onnxruntime::OnnxRuntimeException' what(): /onnxruntime_src/onnxruntime/core/providers/cuda/cuda_call.cc:122 bool...

[Performance] batch inference slower than frame inference

3

### Describe the issue |batch_size|batch_cost_time|frame_cost_time| |----|----|----| |1|207|203| |2|600|406| |3|914|627| |4|1234|855| |5|1570|1106| |6|1868|1267| |7|2297|1779| |8|2657|1779| |9|2953|2096| |10|3320|2260| ### To reproduce ```cpp std::arrayembedding_shape { batch_size , seq_len, embedding_dim }; std::arraymask_shape { batch_size...

是否有计划推出ONNX版本呢？

2

感觉CPU环境下的部署还是用ONNX或者MNN比较好， 在Roadmap中没有看到相应的计划， 未来是否可以呢？

这里有个bug

1

File "/data/renkl1/renkl/Stream-Omni/./stream_omni/eval/run_stream_omni.py", line 270, in eval_model(args) File "/data/renkl1/renkl/Stream-Omni/./stream_omni/eval/run_stream_omni.py", line 191, in eval_model output_ids = model.generate( ^^^^^^^^^^^^^^^ File "/data/renkl1/env/anaconda3/envs/python3.12/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/data/renkl1/renkl/Stream-Omni/stream_omni/model/language_model/stream_omni_llama.py", line 1306,...