Add paraformer npu scripts and tiny fix (wenet-e2e#2563)

* add paraformer npu scripts and tiny fix * add npu setup and rnnt scripts
Mddct · Aug 6, 2024 · b2f59ef · b2f59ef
1 parent bc58f83
commit b2f59ef
Show file tree

Hide file tree

Showing 8 changed files with 436 additions and 9 deletions.
diff --git a/README.md b/README.md
@@ -68,6 +68,28 @@ conda install conda-forge::sox
 pip install torch==2.2.2+cu121 torchaudio==2.2.2+cu121 -f https://download.pytorch.org/whl/torch_stable.html
 ```
 
+<details><summary><b>For Ascend NPU users:</b></summary>
+
+- Install CANN: please follow this [link](https://ascend.github.io/docs/sources/ascend/quick_install.html) to install CANN toolkit and kernels.
+
+- Install WeNet with torch-npu dependencies:
+
+``` sh
+pip install -e .[torch-npu]
+```
+
+- Related version control table:
+
+| Requirement  |      Minimum     | Recommend   |
+| ------------ | ---------------- | ----------- |
+| CANN         | 8.0.RC2.alpha003 | latest      |
+| torch        | 2.1.0            | 2.2.0       |
+| torch-npu    | 2.1.0            | 2.2.0       |
+| torchaudio   | 2.1.0            | 2.2.0       |
+| deepspeed    | 0.13.2           | latest      |
+
+</details>
+
 - Install other python packages
 
 ``` sh

diff --git a/docs/python_package.md b/docs/python_package.md
@@ -31,6 +31,7 @@ You can specify the following parameters.
 * `--align`: force align the input audio and transcript
 * `--label`: the input label to align
 * `--paraformer`: use the best Chinese model
+* `--device`: specify the backend accelerator (cuda/npu/cpu)
 
 ## Python Programming Usage
 

diff --git a/examples/aishell/paraformer/run_npu.sh b/examples/aishell/paraformer/run_npu.sh
@@ -0,0 +1,179 @@
+#!/bin/bash
+
+# Copyright 2019 Mobvoi Inc. All Rights Reserved.
+. ./path.sh || exit 1;
+
+# Automatically detect number of npus
+if command -v npu-smi info &> /dev/null; then
+  num_npus=$(npu-smi info -l | grep "Total Count" | awk '{print $4}')
+  npu_list=$(seq -s, 0 $((num_npus-1)))
+else
+  num_npus=-1
+  npu_list="-1"
+fi
+
+# You can also manually specify ASCEND_RT_VISIBLE_DEVICES
+# if you don't want to utilize all available NPU resources.
+export ASCEND_RT_VISIBLE_DEVICES="${npu_list}"
+echo "ASCEND_RT_VISIBLE_DEVICES is ${ASCEND_RT_VISIBLE_DEVICES}"
+
+stage=0
+stop_stage=2
+
+# You should change the following two parameters for multiple machine training,
+# see https://pytorch.org/docs/stable/elastic/run.html
+HOST_NODE_ADDR="localhost:0"
+num_nodes=1
+job_id=2024
+
+# data_type can be `raw` or `shard`. Typically, raw is used for small dataset,
+# `shard` is used for large dataset which is over 1k hours, and `shard` is
+# faster on reading data and training.
+data_type=raw
+
+train_set=train
+
+train_config=conf/train_paraformer_dynamic.yaml
+checkpoint=exp/paraformer/large/wenet_paraformer.init-ctc.init-embed.pt
+dir=exp/finetune_paraformer_dynamic
+tensorboard_dir=tensorboard
+num_workers=8
+prefetch=500
+
+# use average_checkpoint will get better result
+average_checkpoint=true
+decode_checkpoint=$dir/final.pt
+average_num=5
+decode_modes="ctc_greedy_search ctc_prefix_beam_search paraformer_greedy_search"
+decode_device=0
+decoding_chunk_size=-1
+decode_batch=16
+ctc_weight=0.3
+reverse_weight=0.5
+max_epoch=100
+
+train_engine=torch_fsdp
+
+# model+optimizer or model_only, model+optimizer is more time-efficient but
+# consumes more space, while model_only is the opposite
+deepspeed_config=../whisper/conf/ds_stage1.json
+deepspeed_save_states="model+optimizer"
+
+. tools/parse_options.sh || exit 1;
+
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+  mkdir -p $dir
+  num_npus=$(echo $ASCEND_RT_VISIBLE_DEVICES | awk -F "," '{print NF}')
+  # Use "hccl" for npu if it works, otherwise use "gloo"
+  # NOTE(xcsong): deepspeed fails with gloo, see
+  #   https://github.com/microsoft/DeepSpeed/issues/2818
+  dist_backend="hccl"
+
+  # train.py rewrite $train_config to $dir/train.yaml with model input
+  # and output dimension, and $dir/train.yaml will be used for inference
+  # and export.
+  echo "$0: using ${train_engine}"
+
+  # NOTE(xcsong): Both ddp & deepspeed can be launched by torchrun
+  # NOTE(xcsong): To unify single-node & multi-node training, we add
+  #               all related args. You should change `nnodes` &
+  #               `rdzv_endpoint` for multi-node, see
+  #               https://pytorch.org/docs/stable/elastic/run.html#usage
+  #               https://github.com/wenet-e2e/wenet/pull/2055#issuecomment-1766055406
+  #               `rdzv_id` - A user-defined id that uniquely identifies the worker group for a job.
+  #                           This id is used by each node to join as a member of a particular worker group.
+  #               `rdzv_endpoint` - The rendezvous backend endpoint; usually in form <host>:<port>.
+  # NOTE(xcsong): In multi-node training, some clusters require special NCCL variables to set prior to training.
+  #               For example: `NCCL_IB_DISABLE=1` + `NCCL_SOCKET_IFNAME=enp` + `NCCL_DEBUG=INFO`
+  #               without NCCL_IB_DISABLE=1
+  #                   RuntimeError: NCCL error in: ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1269, internal error, NCCL Version xxx
+  #               without NCCL_SOCKET_IFNAME=enp  (IFNAME could be get by `ifconfig`)
+  #                   RuntimeError: The server socket has failed to listen on any local network address. The server socket has failed to bind to [::]:xxx
+  #               ref: https://github.com/google/jax/issues/13559#issuecomment-1343573764
+  echo "$0: num_nodes is $num_nodes, proc_per_node is $num_npus"
+  torchrun --nnodes=$num_nodes --nproc_per_node=$num_npus \
+           --rdzv_id=$job_id --rdzv_backend="c10d" --rdzv_endpoint=$HOST_NODE_ADDR \
+    wenet/bin/train.py \
+      --device "npu" \
+      --train_engine ${train_engine} \
+      --config $train_config \
+      --data_type  $data_type \
+      --train_data data/$train_set/data.list \
+      --cv_data data/dev/data.list \
+      ${checkpoint:+--checkpoint $checkpoint} \
+      --model_dir $dir \
+      --tensorboard_dir ${tensorboard_dir} \
+      --ddp.dist_backend $dist_backend \
+      --num_workers ${num_workers} \
+      --prefetch ${prefetch} \
+      --pin_memory \
+      --deepspeed_config ${deepspeed_config} \
+      --deepspeed.save_states ${deepspeed_save_states}
+fi
+
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+  if [ "$deepspeed_save_states" = "model+optimizer" ]; then
+    for subdir in $(find "$dir" -maxdepth 1 -type d | grep -v "^$dir$")
+    do
+      # NOTE(xcsong): zero_to_fp32.py is automatically generated by deepspeed
+      tag=$(basename "$subdir")
+      echo "$tag"
+      python3 ${dir}/zero_to_fp32.py \
+        ${dir} ${dir}/${tag}.pt -t ${tag}
+      rm -rf ${dir}/${tag}
+    done
+  fi
+fi
+
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+  # Test model, please specify the model you want to test by --checkpoint
+  if [ ${average_checkpoint} == true ]; then
+    decode_checkpoint=$dir/avg_${average_num}_maxepoch_${max_epoch}.pt
+    echo "do model average and final checkpoint is $decode_checkpoint"
+    python wenet/bin/average_model.py \
+      --dst_model $decode_checkpoint \
+      --src_path $dir  \
+      --num ${average_num} \
+      --max_epoch ${max_epoch} \
+      --val_best
+  fi
+  # Please specify decoding_chunk_size for unified streaming and
+  # non-streaming model. The default value is -1, which is full chunk
+  # for non-streaming inference.
+  base=$(basename $decode_checkpoint)
+  result_dir=$dir/${base}_chunk${decoding_chunk_size}_ctc${ctc_weight}_reverse${reverse_weight}
+  mkdir -p ${result_dir}
+  python wenet/bin/recognize.py --device "npu" \
+    --modes $decode_modes \
+    --config $dir/train.yaml \
+    --data_type $data_type \
+    --test_data data/test/data.list \
+    --checkpoint $decode_checkpoint \
+    --beam_size 10 \
+    --batch_size ${decode_batch} \
+    --blank_penalty 0.0 \
+    --ctc_weight $ctc_weight \
+    --reverse_weight $reverse_weight \
+    --result_dir $result_dir \
+    ${decoding_chunk_size:+--decoding_chunk_size $decoding_chunk_size}
+  for mode in ${decode_modes}; do
+    python tools/compute-wer.py --char=1 --v=1 \
+      data/test/data.list $result_dir/$mode/text > $result_dir/$mode/wer
+  done
+fi
+
+
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+  # Export the best model you want
+  # NOTE (MengqingCao): if RuntimeError "Expected a value of type 'Tuple[Tensor, Tensor]'
+  #    for argument 'hx' but instead found type 'Tensor (inferred)'." occured,
+  #    modify the function "def lstm_forward(self, input1,hx = None):" to
+  #    "def lstm_forward(self, input1, hx: Optional[tuple[torch.Tensor, torch.Tensor]] = None):"
+  #    in torch-npu/utils/module.py
+  # revert this note when torch-npu fix it. sa: https://gitee.com/ascend/pytorch/pulls/12818
+  python wenet/bin/export_jit.py \
+    --config $dir/train.yaml \
+    --checkpoint $dir/avg_${average_num}_maxepoch_${max_epoch}.pt \
+    --output_file $dir/final.zip \
+    --output_quant_file $dir/final_quant.zip
+fi