Problem downloading the en_core_web_trf model on AWS EC2 Ubuntu Instance

[ansh-lehri]

I am trying to download the en_core_web_trf model in the EC2 Ubuntu instance but it is not getting downloading. The system is not throwing any error rather it is completely downloading the 460MB "en_core_web_trf" model but is not finally collecting required dependencies and creating wheels.

Commands used to install spaCy and en_core_web_trf were:

pip install -U spacy --- to download spaCy
python3 -m spacy download en_core_web_trf ---- to download en_core_web_trf model

On executing the above command following process happens and gets killed or stopped after this.

Collecting en-core-web-trf==3.1.0
Downloading https://github.com/explosion/spacy-models/releases/download/en_cor e_web_trf-3.1.0/en_core_web_trf-3.1.0-py3-none-any.whl (460.2 MB)
|████████████████████████████████| 460.2 MB 43.7 MB/s eta 0:00:01

On doing python3 -m spacy info, I get:

spaCy version 3.1.0
Location /home/ubuntu/environment/try/lib/python3.8/site-packages/spacy
Platform Linux-5.4.0-1045-aws-x86_64-with-glibc2.29
Python version 3.8.10
Pipelines

Even when no error is thrown while downloading the "trf" model, it is still not getting added into the spaCy pipelines.

I am not able to understand what is going wrong here. Is there a problem with the EC2 instance I am using or some problem in commands?

EC2 configurations:

Family Type vCPUs Memory(GiB) Instance Storage(GB) Network Performance IPv6 Support
t2 t2.micro 1 1 EBS(only) Low to Moderate Yes

Please help to rectify this issue.

[polm]

Sorry you're having trouble with this, we haven't heard of a problem like this before.

Do other models work?

Does the model show up in pip freeze?

Is there a directory with the model name in /home/ubuntu/environment/try/lib/python3.8/site-packages/?

[johachi]

I had the same/similar problem with en_core_web_trf not installing. The solution was the same (increase storage). In my case, I had to increase the swap size in Docker Desktop.

[imhans33]

@polm I am also trying to deploy my custom NER entity model in aws EC2. So how much RAM and storage would be required for this. Right now i am using a blank model to create custom model. But later i will try with transformer models too. Will medium or large instance will be sufficient ?

[polm]

@imhans33

1. Don't ping individual maintainers.
2. Your question is only tangentially related to this one, you should have opened a new question.

Required memory depends on your model training settings, docs processed, and other factors, so it's hard to say. Our main advice is that for training with Transformers we recommend 10GB of GPU RAM, though for inference you can get away with less. For more detailed numbers I would recommend you run a pipeline on a local machine and measure memory usage.

[imhans33]

Is there any other platform apart from aws, azure etc where we can deploy spacy models and test it with few developers from different locations.

[polm]

spaCy is just Python code, you can put it on anything. I guess you could use HuggingFace hub? Like here:

https://huggingface.co/spacy/en_core_web_lg