Questions tagged [large-language-models]
The large-language-models tag has no usage guidance, but it has a tag wiki.
98
questions
0
votes
0
answers
19
views
Approaching construction of model that interprets financial reports
I want to train a model to be able to interpret financial reports (from a company). Basically, I want to be able to extract relevant information without needing to read through hundreds of pages of ...
- 1
0
votes
0
answers
9
views
Largest fully open source LLMs including training
It seems that most of the open-sourced large language models (LLMs) like Llama 2 had the model released but not the exact training procedure and training data-sources (exact data revisions) so that ...
- 271
0
votes
1
answer
15
views
What is the current limit for LLMs regarding the size of the context that can be fed to them?
Is there a limitation in current large language models (LLMs) in terms of practical processing time or memory resources when it comes to digesting the context provided by users? What I mean regarding ...
- 101
0
votes
0
answers
17
views
Is beam search the actual obstacle that prevents GPT-style models from doing sophisticated math reasoning?
This is a rather soft question. Some people believe that GPT-style models can eventually solve very complex math problems if the models are large enough, but I'm skeptical about this. Suppose the GPT ...
- 1
0
votes
0
answers
36
views
Intentionally corrupting LLM weights (lobotomy)
It is largely unknown how LLMs work inside. Has anyone scientifically tried to corrupt (open source) model's weights in an organized manner to maybe detect which parts of the model are doing what or ...
- 271
0
votes
0
answers
16
views
Masking during Instruction Tuning for LLM finetuning
I'm currently trying to learn more about LLMs particularly generative decoder only models such as the GPT family of models. I do have one question about masking though.
For me the way masking is ...
-1
votes
0
answers
18
views
LLM's for IMO-The 10m USD Problem [closed]
Just saw that Alex Gerko has launched a $10M challenge for the first AI to win IMO Gold. Link:https://aimoprize.com/
Curious, are there currently any viable LLM's that are even remotely good at ...
0
votes
0
answers
11
views
Could an analysis of GPT4's WAIS score be published?
Similar to this: https://www.scientificamerican.com/article/i-gave-chatgpt-an-iq-test-heres-what-i-discovered/
But more detailed and in depth (subtest breakdown, including image analysis, etc.), WAIS-...
- 113
0
votes
1
answer
45
views
How LLM keeps the context of a chat/thread
How an LLM keeps the context (what has already been entered by the user) of a chat/thread?
For reference, in chat.openai.com, for each chat we create (or a Thread according to their API), the LLM ...
- 103
0
votes
0
answers
17
views
Specifying multiple sources of RAG in different formats
I am considering whether to do RAG or Fine tuning for LLM to answer questions for clients in a small business
The small business is a car rental company that gets
new clients call up about certain ...
- 101
0
votes
0
answers
28
views
Requirements for AI to be able to read a book and accurately recall the facts?
As a fun random project I've been working on is learning languages and making cheatsheets for them, dictionaries, etc.. I would love to encode the grammar rules which have been captured in books ...
- 153
2
votes
1
answer
81
views
Why are some LLMs trained on both CommonCrawl and Wikipedia/StackExchange?
Some LLMs are trained on both CommonCrawl and Wikipedia/StackExchange. Why? Does CommonCrawl already contain Wikipedia/StackExchange?
E.g., from the LLaMa 1 paper:
and from https://huggingface.co/...
- 2,814
0
votes
0
answers
21
views
From Text-To-Speech to LLMs: Providing "writing style"
I've just recently learned about Text-To-Speech models and how they are trained. Unlike LLMs, to a provided pair (text, speech), a feature vector ${f}$, that was generated by more speech of that ...
- 143
2
votes
1
answer
58
views
LLM Hallucinations on In-Context Data
I know of some benchmarks that LLMs do undergo, but I am no expert whatsoever. I think what I am wondering about is closest to TruthfulQA.
The question came up when I heard of combining company data ...
- 143
1
vote
1
answer
39
views
How can LLMs understand and perform meta tasks? (e.g. summarization)
I don't ask how to make it summarize xy but if it is known how a "LLM" understands and performs this meta task at all.
The same is true for prompts like "Be brief" or "Explain ...
- 23
0
votes
0
answers
18
views
How does VRAM usage change when running prompts async through LLMs?
If my LLM uses X GB of VRAM when processing prompts sequentially, does it use 2X GB if I run 2 prompts at once using async support? e.g. using Langchains ...
- 121
1
vote
0
answers
76
views
LLM for Postgres
I have a postgres database with 200+ Tables. Each table contains information about my supply inventory. It also contains columns which are JSON and there are nested JSON as well. There are ...
0
votes
0
answers
65
views
claude 2 doesn't know basic math? [duplicate]
Sometimes, when I see answers like this from large language models, it makes me feel disgusted:
Me: Does Voyager 1 have enough velocity to escape the solar system without using Jupiter's gravity ...
- 1
2
votes
0
answers
37
views
How do I balance context and history when creating prompts for LLM's?
A conversation through the OpenAI API looks something like this
...
- 121
0
votes
0
answers
112
views
Is ChatGPT a viable strategy for solving the P-NP Problem?
According to ZDNet, it is an open question whether a transformer LLM like ChatGPT can facilitate the determination of a solution to the P-NP Problem. (See Can generative AI solve computer science's ...
- 101
0
votes
1
answer
35
views
How to train a seq2seq model to rephrase input text following given rules
I want to train (fine-tune) a seq2seq model to perform the task of rephrasing input following these rules :
1- always follow the pattern "Entity Verb Entity"
2- only use simple sentences : ...
-1
votes
1
answer
248
views
How to get Llama-2 Rotary Embeddings?
I want to get the Llama-2 rotary embeddings. I do print(model) and get the following output:
In the picture I highlight the rotary embeddings.
How can get the ...
- 99
0
votes
0
answers
21
views
What options do I have when performing RAG on similarly phrased chunks?
I'm building a RAG pipeline to extract real estate phrases from excel documents. These phrases are short (2-5 words) and are often phrased differently.
I've manually added in different phrases to each ...
- 121
0
votes
0
answers
14
views
What is the policy model in RLHF for LLMs?
What is the policy model doing explicitly in an LLM with RLHF setup?
From my understanding, LLMs generate in a way that is no different from any of their predecessors: beam search decoding, ...
1
vote
1
answer
286
views
How can I teach a book to an LLM?
I am trying to find out how I can teach the content of a whole, multiple hundert pages book to an LLM so that it "knows" all details and can be queried, give summaries etc. The book is one ...
- 11
0
votes
1
answer
54
views
Why do current language models no longer generate to long or short texts?
One of the biggest strengths of ChatGPT is that it generates fitting text with respect to the input query. It usually stays on topic, anwers the question completely and especially does not start ...
- 1
1
vote
2
answers
98
views
Tips and tricks when training a very large language model?
Have never trained a (very) large language model, so I am wondering if the process is the same as training a (regular) language model, i.e. you prepare the data, set up the architecture, ...
0
votes
1
answer
38
views
How to prompt Open AI for a variable list length?
I struggle to get a variable list out of Open AIs GPT-3.5. Even with a prompt like
...
- 101
-2
votes
2
answers
92
views
How successful are the state-of-the-art (2023) email filters really? [closed]
How successful are the state-of-the-art (2023) email filters really?
Some references claim that spam detection may reach high accuracy in test settings, but I've thought that email filtering should ...
- 105
2
votes
1
answer
659
views
Why do transformers compute the loss over the prompt?
When fine-tuning large language models, which are commonly decoder transformer architectures, sometimes we want to compute the training loss over the entire prompt, sometimes just the completion ...
- 951
0
votes
0
answers
88
views
Does Embeddings and Vector Databases solve the need of having longer context windows?
I am learning to use the OpenAI API to build LLM-based agents. I recently came across the concept of vector databases, which use embeddings to convert text into vectors and store them in a database ...
- 453
0
votes
0
answers
29
views
Create samples out of documents for Causal Language Modelling
I want to create an input source for Causal Language model using Llama 2 model in hugging face. I have a set of documents which are scraped from a specific website and want to fine-tune on them. Each ...
- 1
1
vote
1
answer
146
views
What causes my loss curve to consistently oscillate when training an LLM?
Why is my loss curve consistently oscillating? Every 50 steps it jumps back up. I'm assumming there's a bug in my data, since I'm using this colab notebook that shows a proper train/loss at the bottom....
- 11
0
votes
0
answers
23
views
Any research in "probe-tuning" of LLMs?
Is there any research in "probe-tuning" of LLMs, i.e., tuning LLM's parameter weights such that a specific probe (classifier) is more reliably detecting certain markers throughout the ...
- 101
0
votes
0
answers
20
views
What if in DPR (dense passage retrieval), the answer belongs to more than one passage?
In the DPR paper
the dataset is expected to be in this format D = {<qi, pi+, pi,1-, ... >}
With only one positive passage, but it is possible that the question requires an answer that spans ...
- 1
0
votes
0
answers
29
views
can I finetune a model on Azure for information extraction based on "question", "context", and "answer" training data?
I am working on extracting certain fields from a large corpus.
I was looking at finetuning an LLM on Azure for the task. I think finetuning is the right idea (as opposed to vector databases, or RAG), ...
3
votes
3
answers
1k
views
What will happen if to train an LLM on mathematical exersises?
What will happen if to train an LLM on taking integrals and solving equations? The process of mathematical education can be absolutely automated by a computer algebra system because the verification ...
- 321
0
votes
0
answers
253
views
Optimal Quantity of Training Data for Fine-Tuning an LLM: Is Bigger Always Better?
I am currently working on fine-tuning an LLM for a specific task, and I am trying to determine the optimal size for my training dataset. Intuitively, one might think that the more data, the better. ...
- 554
1
vote
2
answers
127
views
What is the difference betwen fine runing and rlhf for llm?
I am confused about the difference betwen fine runing and rlhf for llm. When to use what? I know RLHF need to creating a reward model which at furst rates responses to align the responses to the ...
- 293
1
vote
1
answer
74
views
Can LLMs be used to discover new laws of logic?
Can LLMs be used to discover new laws of logic?
Stephen Wolfram seems to claim this in What Is ChatGPT Doing … and Why Does It Work?, § "What Really Lets ChatGPT Work?":
is there a general ...
- 163
0
votes
0
answers
47
views
What are the *non-cost-related* reasons RNN+Attention underperform Transformers?
There are obvious trainability and performance challenges with RNNs, such as having to process in serial and BPTT. But let's say we magically had an "optimal" set of weights for the RNN + ...
- 146
2
votes
1
answer
92
views
Could hallucinations be the demise of the AI hype? [closed]
For quite some time now, I have been evaluating ChatGPT's capability to deliver accurate and helpful responses. While its performance is undeniably impressive, the issue of hallucinations poses a ...
- 133
0
votes
0
answers
13
views
Does anyone know about a reference where someone has aggregated the cost optimization strategies for deploying LLMs?
I am looking for a source where someone has mentioned the most commonly used strategies and optimisations to deploy LLMs on consumer hardware. I have read about layer offloading, quantisation methods ...
0
votes
1
answer
600
views
Why does LLM inference cost scale in both input tokens and output tokens?
EDIT
This question was flawed. See my answer with help from commenters.
Original question
This question has been asked in other forums [1] [2] but I'm not sure I understand the claims, which are (...
- 146
0
votes
1
answer
120
views
Fine-Tune Llama on main and auxiliary task
I am trying to fine-tune Llama model on two task at the same time, using hugging face library:
Main task: Causal language model like the model was initially trained for
A classification task based on ...
- 1
-1
votes
1
answer
51
views
Open access Adept-like dataset? (LLM-to-computer-input) [closed]
Here's a demo for Adept ACT-1 for Transformers. I don't doubt that one could create a demo video using zero-shot; actually I tested just now and the basic chat.openai.com interface was able to do some ...
- 146
6
votes
3
answers
789
views
Why can't Lucene search be used to power LLM applications?
w.r.t. LLM applications using the RAG (retriever-augmented-generation) architecture, people have started taken it for granted that it will be powered by a vector database. e.g., see this:
The most ...
- 254
3
votes
2
answers
3k
views
How to Formulate a realiable ChatGPT Prompt for Sentiment Analysis of a Text, and show that it is reliable?
I have a dataset which consists of like.. 400000 sentences and I want give each sentence to ChatGPT so it classifies each sentence as positive or ...
- 131
1
vote
1
answer
708
views
What is considered the pre-fill, and what is considered the decoding phase in this process?
I've seen conflicting information about this online so I'm looking for clarification. I'm dealing with the causal LLaMAF model specifically.
I used to think that a sequence of tokens is generated in, ...
- 11
0
votes
1
answer
427
views
What role does data quality plays in the LLM scaling laws?
DeepMind released the Training Compute-Optimal Large Language Models paper in 2022 which describe some scaling laws for LLMs. As far as I understand this is the most accredited reference to estimate ...
- 123