From 5ec95e5482867de02eece6615ff51662c7ed9b37 Mon Sep 17 00:00:00 2001
From: Yifan Yang <64255737+yfyeung@users.noreply.github.com>
Date: Wed, 23 Apr 2025 16:18:38 +0800
Subject: [PATCH] Fix SpeechLLM recipe (#1926)

---
 egs/speech_llm/ASR_LLM/RESULTS.md |  8 ++++----
 egs/speech_llm/ASR_LLM/prepare.sh | 13 ++++++++-----
 egs/speech_llm/ASR_LLM/shared     |  1 +
 3 files changed, 13 insertions(+), 9 deletions(-)
 mode change 100644 => 100755 egs/speech_llm/ASR_LLM/prepare.sh
 create mode 120000 egs/speech_llm/ASR_LLM/shared

diff --git a/egs/speech_llm/ASR_LLM/RESULTS.md b/egs/speech_llm/ASR_LLM/RESULTS.md
index 830c70397..01c48a82e 100644
--- a/egs/speech_llm/ASR_LLM/RESULTS.md
+++ b/egs/speech_llm/ASR_LLM/RESULTS.md
@@ -42,8 +42,8 @@ huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_aishel
 # For multi-hans fine-tuned whisper model
 # huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_multi-hans-zh_whisper v1.1/whisper-large-v2-multi-hans-zh-epoch-3-avg-10.pt
 
-# huggingface-clie download  --local-dir models/qwen     Qwen/Qwen2-7B-Instruct
-huggingface-clie download  --local-dir models/qwen     Qwen/Qwen2-1.5B-Instruct
+# huggingface-cli download  --local-dir models/qwen     Qwen/Qwen2-7B-Instruct
+huggingface-cli download  --local-dir models/qwen     Qwen/Qwen2-1.5B-Instruct
 
 # First, we only train the projector and freeze other modules.
 torchrun --nproc_per_node 8 ./whisper_llm_zh/train.py \
@@ -57,7 +57,7 @@ torchrun --nproc_per_node 8 ./whisper_llm_zh/train.py \
   --use-flash-attn True \
   --use-lora False --unfreeze-llm False
 
-# Then we jointly train the projector and LLM LoRA modules.
+# Then, we jointly train the projector and LLM LoRA modules.
 torchrun --nproc_per_node 8 ./whisper_llm_zh/train.py \
   --max-duration 200 \
   --exp-dir ./whisper_llm_zh/exp_test \
@@ -81,7 +81,7 @@ huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_aishel
 # For multi-hans fine-tuned whisper model
 # huggingface-cli download --local-dir models/whisper    yuekai/icefall_asr_multi-hans-zh_whisper v1.1/whisper-large-v2-multi-hans-zh-epoch-3-avg-10.pt
 
-huggingface-clie download  --local-dir models/qwen     Qwen/Qwen2-7B-Instruct
+huggingface-cli download  --local-dir models/qwen     Qwen/Qwen2-7B-Instruct
 
 mkdir -p whisper_llm_zh/exp_aishell_whisper_qwen2_1.5B
 ln -s models/checkpoint/epoch-10-avg-5.pt whisper_llm_zh/exp_aishell_whisper_qwen2_1.5B/epoch-999.pt
diff --git a/egs/speech_llm/ASR_LLM/prepare.sh b/egs/speech_llm/ASR_LLM/prepare.sh
old mode 100644
new mode 100755
index 6f5ed5448..8ca3c1c36
--- a/egs/speech_llm/ASR_LLM/prepare.sh
+++ b/egs/speech_llm/ASR_LLM/prepare.sh
@@ -7,6 +7,9 @@ set -eou pipefail
 
 stage=0
 stop_stage=0
+
+. shared/parse_options.sh || exit 1
+
 # All files generated by this script are saved in "data".
 # You can safely remove "data" and rerun this script to regenerate it.
 mkdir -p data
@@ -23,7 +26,7 @@ if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
 
   # pip install huggingface_hub['cli']
   # for aishell 1
-  huggingface-cli download --local-dir data yuekai/aishell_whisper_fbank_lhotse
+  huggingface-cli download --repo-type dataset --local-dir data yuekai/aishell_whisper_fbank_lhotse
 
 fi
 
@@ -31,9 +34,9 @@ if [ $stage -le 1 ] && [ $stop_stage -ge 1 ]; then
   log "stage 1: Download whisper-large-v2 multi-hans-zh fbank feature from huggingface"
 
   # for multi-hans-zh
-  huggingface-cli download --local-dir data/fbank yuekai/wenetspeech_whisper_fbank_lhotse
-  huggingface-cli download --local-dir data/fbank yuekai/multi_hans_zh_whisper_fbank_lhotse
-  huggingface-cli download --local-dir data/fbank yuekai/alimeeting_aishell4_training_whisper_fbank_lhotse
+  huggingface-cli download --repo-type dataset --local-dir data/fbank yuekai/wenetspeech_whisper_fbank_lhotse
+  huggingface-cli download --repo-type dataset --local-dir data/fbank yuekai/multi_hans_zh_whisper_fbank_lhotse
+  huggingface-cli download --repo-type dataset --local-dir data/fbank yuekai/alimeeting_aishell4_training_whisper_fbank_lhotse
 fi
 
 if [ $stage -le 2 ] && [ $stop_stage -ge 2 ]; then
@@ -41,6 +44,6 @@ if [ $stage -le 2 ] && [ $stop_stage -ge 2 ]; then
 
   # for speechio test sets
   mkdir data_speechio
-  huggingface-cli download --local-dir data_speechio yuekai/icefall_asr_speechio
+  huggingface-cli download --repo-type model --local-dir data_speechio yuekai/icefall_asr_speechio
   mv data_speechio/fbank/* data/fbank
 fi
diff --git a/egs/speech_llm/ASR_LLM/shared b/egs/speech_llm/ASR_LLM/shared
new file mode 120000
index 000000000..4cbd91a7e
--- /dev/null
+++ b/egs/speech_llm/ASR_LLM/shared
@@ -0,0 +1 @@
+../../../icefall/shared
\ No newline at end of file