Library trl==0.11.4 is missing in the latest Hugging Face Neuron Deep Learning AMI (Ubuntu 22.04)

[jianyinglangaws]

System Info

The missing library leads to compilation error of the Qwen3 Lora finetuning tutorial example https://huggingface.co/docs/optimum-neuron/en/training_tutorials/finetune_qwen3.


[rank26]: Traceback (most recent call last):                                                                                                                                                                       
[rank26]:   File "/home/ubuntu/optimum-neuron/examples/training/qwen3/finetune_qwen3.py", line 128, in <module>                                                                                                    
[rank26]:     train(                                                                                                                                                                                               
[rank26]:   File "/home/ubuntu/optimum-neuron/examples/training/qwen3/finetune_qwen3.py", line 86, in train                                                                                                        
[rank26]:     sft_config = NeuronSFTConfig(                                                                                                                                                                        
[rank26]: TypeError: NeuronSFTConfig.__init__() got an unexpected keyword argument 'max_seq_length'                                                                                                                    
[rank8]: Traceback (most recent call last):                                                                                                                                                                        
[rank8]:   File "/home/ubuntu/optimum-neuron/examples/training/qwen3/finetune_qwen3.py", line 128, in <module>                                                                                                     
[rank8]:     train(                                                                                                                                                                                                
[rank8]:   File "/home/ubuntu/optimum-neuron/examples/training/qwen3/finetune_qwen3.py", line 86, in train                                                                                                         
[rank8]:     sft_config = NeuronSFTConfig(                                                                                                                                                                         
[rank8]: TypeError: NeuronSFTConfig.__init__() got an unexpected keyword argument 'max_seq_length'               

Can someone please help fix this?

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction (minimal, reproducible, runnable)

1. Launch the instance trn1.32xlarge with Huggingface DLAMI huggingface-neuron-2025-07-18T12-25-10Z-692efe1a-8d5c-4033-bcbc-5d99f2d4ae6a This DLAMI is based on Neuron SDK 2.24

   Neuron Library List:

(aws_neuronx_venv_pytorch_2_7) ubuntu@ip-172-31-45-102:~/optimum-neuron$ pip list | grep neuron 
libneuronxla 2.2.4410.0+835a67fb 
neuronx-cc 2.19.8089.0+8ab9f450 
neuronx-distributed 0.13.14393+b8569585 
optimum-neuron 0.3.0 
tensorboard-plugin-neuronx 2.6.117.0 
torch-neuronx 2.7.0.2.8.6734+ac864f72 
(aws_neuronx_venv_pytorch_2_7) ubuntu@ip-172-31-45-102:~/optimum-neuron$ pip list | grep optimum 
optimum 1.24.0 
optimum-neuron 0.3.0

2. Download the optimum-neuron source code

    git clone https://github.com/huggingface/optimum-neuron.git
    cd ~/[optimum-neuron](https://github.com/huggingface/optimum-neuron/tree/main)/[examples](https://github.com/huggingface/optimum-neuron/tree/main/examples)/[training](https://github.com/huggingface/optimum-neuron/tree/main/examples/training)/qwen3

3. Compile neuron_parallel_compile ./finetune_qwen3.sh and you would encounter the error.

Expected behavior

The compilation should complete without error messages.

[jimburtoft]

@jianyinglangaws

If you want to do training, you need to do

pip install optimum-neuron[training]

That should get all the dependencies you need.

We do have it in some of the walkthroughs.

https://github.com/huggingface/optimum-neuron/blob/e84e0c61fb80b6f6dafc5906578b286a122a7c3a/docs/source/training_tutorials/sft_lora_finetune_llm.mdx#1-setup-aws-environment

Maybe we need to document it somewhere else too?

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue was closed because it has been stalled for 5 days with no activity.