Update README.md

teamtee · web-flow · commit 856f430eb719 · 2025-06-16T13:47:06.000+08:00
diff --git a/examples/asr_librispeech/README.md b/examples/asr_librispeech/README.md
@@ -79,7 +79,7 @@ If you're interested in training with DeepSpeed, refer to the script `finetune_w
 }
 ```
 
-Note that when using `zero-0`/`1`/`2`/`3`, the DeepSpeed model is saved as `pytorch_model.bin`
+Note that when using `zero-0`/`1`/`2`/`3`, the DeepSpeed model is saved as `pytorch_model.bin`, and you should change "++ckpt_path=$ckpt_path/model.pt" to " ++ckpt_path=$ckpt_path/pytorch_model.bin" in the script to use the model during inference.
 If you use bf16/fp16 training in DeepSpeed and encounter NaN in train/eval loss, check the autocast in `src/slam_llm/utils/deepspeed_utils.py`:
 
 ```python
@@ -96,4 +96,4 @@ You can refer to the paper for more results.
   journal={arXiv preprint arXiv:2402.08846},
   year={2024}
 }
-```
+```

Original file line number	Diff line number	Diff line change
@@ -79,7 +79,7 @@ If you're interested in training with DeepSpeed, refer to the script `finetune_w
`79`	`79`	`}`
`80`	`80`	```
`81`	`81`
`82`		-Note that when using `zero-0`/`1`/`2`/`3`, the DeepSpeed model is saved as `pytorch_model.bin`
	`82`	+Note that when using `zero-0`/`1`/`2`/`3`, the DeepSpeed model is saved as `pytorch_model.bin`, and you should change "++ckpt_path=$ckpt_path/model.pt" to " ++ckpt_path=$ckpt_path/pytorch_model.bin" in the script to use the model during inference.
`83`	`83`	If you use bf16/fp16 training in DeepSpeed and encounter NaN in train/eval loss, check the autocast in `src/slam_llm/utils/deepspeed_utils.py`:
`84`	`84`
`85`	`85`	```python
`@@ -96,4 +96,4 @@ You can refer to the paper for more results.`
`96`	`96`	`journal={arXiv preprint arXiv:2402.08846},`
`97`	`97`	`year={2024}`
`98`	`98`	`}`
`99`		-```
	`99`	+```