LLMS

→ View original post on X — @hwchase17

Fine-tuning Models for Custom Tone: Elon Musk Example

By

@hwchase17

–

26 August 2023 18h02

One of my favorite use cases for fine-tuning is getting a model to speak in a custom tone This is kinda hard to achieve with just prompting/few-shot Here's an example of fine-tuning a model to tweet like Elon Musk (compare it to what you get by just prompting)

→ View original post on X — @replicate

Code Llama API Available on Replicate with Multiple Models

By

@replicate

–

26 August 2023 17h58

Run Code Llama with an API on Replicate. Code Llama can generate and discuss code. It's the best open-source model for things like debugging and code completion. • 7b https://
replicate.com/replicate/code
llama-7b
…
• 7b Python https://
replicate.com/replicate/code
llama-7b-python
…
• 13b

→ View original post on X — @thom_wolf

HumanEval Benchmark: Assessing AI Code Generation Capabilities

By

@thom_wolf

–

26 August 2023 16h50

Time to read again Loubna’s nice post from last week diving in the HumanEval benchmark

RAG Technique: Solving Modern LLM Limitations Without Fine-Tuning

By

–

26 August 2023 16h37

Let’s dive in to one of the most-talked about AI techniques—retrieval augmented generation (RAG)— that developers are using as a precursor to fine-tuning. It helps solve the recent issue of modern LLMs while being not-really-that-hard to implement.

RAG Closes Gap Between Open Source and API Providers

By

–

26 August 2023 16h26

And using RAG could close the gap even further between the performance of open source alongside the API providers, particularly for companies that don’t want to hand over control to a provider and are looking for a cheaper, faster, and perhaps more importantly, predictable tool.

Open Source AI Models Accessible Without Fine-Tuning Resources

By

–

26 August 2023 16h25

It's a tantalizing prospect for companies that are exploring the use of open source models, but don't have the resources (personnel or financial) to fine-tune or pre-train a model. It works right out of the box without any significantly advanced technical requirements.

Adding Memory to Frozen LLM Models Reduces Costs

By

–

26 August 2023 16h23

It offers the ability to endow a kind of "memory" to models that are frozen in time, as well as lower the overall cost of LLM usage. Or as @bobvanluijt told me, a way to make a stateless tool more stateful.