Questions tagged [training]

Training is the process of using large amounts of data to enable a large language model to perform a task or answer questions. Use this tag for questions centered around fundamentals of training or how to implement training.

Filter by
Sorted by
Tagged with
5 votes
0 answers
56 views

How to fine-tune an LLM on a very small document

I am currently trying to fine-tune an LLM on a single document (about 1700 characters). I know that generally it is better to use prompt injection or something like a RAG system to provide specific ...
Roy Varon's user avatar
  • 151
0 votes
0 answers
9 views

Can I modify the weight of a particular section of text in an LLM training corpus?

I have a rather odd data format that I want to train a LLM on. I want it to focus on learning specific segments within one piece of training data. The problem is that I can't just delete the rest of ...
user avatar
1 vote
0 answers
40 views

Freezing layers for finetuning

So i'm planning to fine tune a 7b parameter model for a research project. I understand the different steps of model fine tuning, namely Supervised fine tuning - where we train model on curated ...
kaiser's user avatar
  • 111
0 votes
0 answers
14 views

how to load distcp checkpoint files?

I have fintuned full aparmeters of mistral 7-b model, and i have used FDSP in HF accelerate I have a checkpoint which is place in a folder pytorch_model_0, which contains multiple distcp files. how ...
Mustafa Alahmid's user avatar
0 votes
0 answers
31 views

Parallelised GPU training

When training/fine-tuning an LLM model it is necessary to have huge GPU resources (VRAM), as such I was wondering if there was a way in Linux to pool GPU resources if I have, say, 5 separate computers ...
mya205's user avatar
  • 9
3 votes
2 answers
109 views

How does corpus size affect an LLM? Would one trained on just a book still be able to grasp the whole language?

I'm trying to understand how various factors affect LLMs. Specifically the size of the dataset they're trained on. What would be the main difference between: A regular LLM (like ChatGPT) that's ...
laggingreflex's user avatar
0 votes
0 answers
13 views

Interpretation of Evaluation Values of Augmented SBERT Training with EmbeddingSimilarityEvaluator()

I train a BI-Encoder to get an Augmented SBERT and I get a final training result. How can I interpret the following output of the final training result? EmbeddingSimilarityEvaluator: Evaluating the ...
Christian01's user avatar
0 votes
0 answers
24 views

The using of golden dataset in Augmented SBERT Training

I use the training strategy of Augmented SBERT (Domain-Transfer) to train an augmented SBERT. In the code example they use the golden-dataset (STSb) for the training evaluator. Here two code snippets ...
Christian01's user avatar
6 votes
2 answers
160 views

How can I know if an image generation tool might copy someone's art?

I've read that the training sets and outputs of some generative AI services include copyrighted art. I'm interested in trying out image generators while not accidentally copying someone's art who hasn'...
2 votes
2 answers
189 views

What works better: An LLM trained on better texts or an LLM with better prompts?

Assume you have a very large corpus of high-quality documents related to a given topic, and assume you have a pretrained large language model (the foundation model) with training data not containing ...
Hans-Peter Stricker's user avatar
0 votes
1 answer
197 views

Is it possible to create a reliable AI quotation tool using excel spreadsheets with ChatGPT API

I'm working on a project to create an AI Quotation tool using the ChatGPT API trained on my data. The data I have spans the last 10 years and is mostly in Excel and PDF format, but it's not ...
user avatar
11 votes
3 answers
2k views

How do I "teach" a large language model new knowledge?

Suppose I have a copy of a pre-trained transformer-based large language model like Google's T5 or Meta's Llama. Due to the pre-training, it contains a lot of knowledge. However, I want to teach the ...
Ian Campbell's user avatar