NeMo

What do you like best about NeMo?

It gives single unified application for training, deploying and building a conversation AI bots using either already available or finetuning a model. It also gives flexibility to fine tune a LLM using same application. It has almost all the widely used LLM models for use with just one single API key. Since it uses NVIDIA GPU at its peaks, the performance is far better than other RAG building frameworks and it outperforms my previous experience from mistral ai use with bare langchain integration.

What do you dislike about NeMo?

Docs are clear but it can still be betterm especially the index part and there should be appendix at the end which can be referred later on to quickly look for deployment processes for quickly jumping to that page, this is especailly useful here as the application is in early phase and the only support for now is the given documentations and not wide community forum.

What problems is NeMo solving and how is that benefiting you?

NeMo is an application from NVIDIA which allows training, configuring and building a RAG application from within one framework both on cloud and on personal hardwares. The application are very efficient on high end NVIDIA GPUs that it outperforms other RAG builduing frameworks and this allows me to use my GPU at its max limit to run some cool RAG chat bots locally in my machines.

What do you like best about NeMo?

The most helpful thing about Nemo is to let developers can utilize built-in features and deep learning models with a lot of features ranging from NLP to Speech tasks. The code flow is easy to understand for newbies. Moreover, they provided us with a lot of clear documentation for each task, which developers can easy to use and follow. Nvidia has published another useful resource for deploying models as Triton Inference Server.

What do you dislike about NeMo?

However, It seems that if you are working in low-resource languages such as Vietnamese, you have to understand the workflow and modify them; this will make to utilize the pre-trained models. Moreover, at Nemo, contributors prioritize the models from Nvidia team; you should wait for them to update new or SOTA models. I hope that Nvidia team will release more features such as real-time Speech recognition, Text-to-speech, ... to help developers like me in production environment.

What problems is NeMo solving and how is that benefiting you?

As I mentioned before, I can get benefits from utilizing built-in features in a range of tasks from NLP to Speech tasks. In my case, I utilized Nemo to develop Vietnamese Speech recognition and Text-to-speech applications. This helped me to reduce my working time for developing these kind of models.

What do you like best about NeMo?

It is open source and built on PyTorch backend, supporting Speech Recognition, Natural Language Processing, and Voice Recognition. Training the models is a breeze as it requires fewer lines of code to deploy the trained model.

What do you dislike about NeMo?

There is still scope for improvement in the pre-trained model as it still does not correctly identify the words/pronunciation. It's sometimes hit-and-miss. Also, it requires substantial memory to run the models.

What problems is NeMo solving and how is that benefiting you?

We are deploying Voice bots and chatbots that use NeMo out of the box for our use cases, making deployment smoother. There is a faster turn-around as we have one model less to train.

What do you like best about NeMo?

It is very important to convert texts to speeches and Nemo is a nice and easy tool to make it

What do you dislike about NeMo?

It is very helpfull but asks for payment

What problems is NeMo solving and how is that benefiting you?

It will make easy speeches from text. It will easy for online seminars