Lightnews — Scholar-powered news

Moritz Laurer @moritzlaurer.bsky.social · Jan 15

Prompt-templates docs: moritzlaurer.github.io/prompt_templ...
Templates on the hub: huggingface.co/datasets/Mor...
Prompt-templates collection: huggingface.co/collections/...
Paper: arxiv.org/pdf/2501.04519

Overview - Prompt Templates

A library for working with prompt templates locally or on the Hugging Face Hub

moritzlaurer.github.io

2

Moritz Laurer @moritzlaurer.bsky.social · Jan 15

—without GPT-4-based data distillation.
💾 While we wait for the release of code and datasets, you can already download the prompts they used from the HF Hub!

Details here 👇

1 1

Moritz Laurer @moritzlaurer.bsky.social · Jan 15

🤖 A Process Preference Model (PPM) enables fine-grained evaluation of intermediate steps, improving training data quality.
🧪 The system underwent four rounds of self-evolution, progressively refining both the policy and reward models to tackle Olympiad-level math problems

1 1

Moritz Laurer @moritzlaurer.bsky.social · Jan 15

📏 The paper introduces rStar-Math, which claims to rival OpenAI o1's math reasoning capabilities by integrating Monte Carlo Tree Search (MCTS) with step-by-step verified reasoning trajectories.

1

Moritz Laurer @moritzlaurer.bsky.social · Jan 15

[email protected]'s rStar-Math paper claims that 🤏 ~7B models can match the math skills of o1 using clever train- and test-time techniques. You can now download their prompt templates from @hf.co !
🧵

1 1

Moritz Laurer @moritzlaurer.bsky.social · Jan 11

- prompt-templates docs: moritzlaurer.github.io/prompt_templ...
- all templates on the HF Hub: huggingface.co/datasets/Mor...
- FACTS paper: storage.googleapis.com/deepmind-med...

Overview - Prompt Templates

A library for working with prompt templates locally or on the Hugging Face Hub

moritzlaurer.github.io

3

Moritz Laurer @moritzlaurer.bsky.social · Jan 11

💾 You can now download and reuse these prompt templates via the prompt-templates library!

🔄 The library simplifies sharing prompt templates on the HF hub or locally via standardized YAML files. Let’s make LLM work more transparent and reproducible by sharing more templates like this!

Links 👇

1 1

Moritz Laurer @moritzlaurer.bsky.social · Jan 11

🧪 The authors tested different prompt templates on held-out data to ensure their generalization.

📚 It's highly educational to read these templates to learn how frontier labs design prompts and understand their limitations.

1 2

Moritz Laurer @moritzlaurer.bsky.social · Jan 11

📏 The paper introduces the FACTS Grounding benchmark for evaluating the factuality of LLM outputs.

🤖 Fact-checking is automated by an ensemble of LLM judges that verify if a response is fully grounded in a factual reference document.

1 1

Moritz Laurer @moritzlaurer.bsky.social · Jan 11

FACTS is a great paper from @GoogleDeepMind on measuring the factuality of LLM outputs. You can now download their prompt templates from @huggingface to improve LLM-based fact-checking yourself!
🧵

1 3

Moritz Laurer @moritzlaurer.bsky.social · Jan 9

Release: github.com/huggingface/...
Mergekit: github.com/arcee-ai/mer...
Mixture of judges paper: huggingface.co/papers/2409....

Moritz Laurer @moritzlaurer.bsky.social · Jan 9

⚖️ Mixture of judges: The new AllTrueJudge combines decisions from multiple binary judges for more nuanced evaluation.

Read the release notes and other resources here 👇

1

Moritz Laurer @moritzlaurer.bsky.social · Jan 9

🛠️ Tool call support: TRL preprocessing now supports tool integration, laying the groundwork for agent fine-tuning with examples like dynamic temperature fetching in prompts.

1

Moritz Laurer @moritzlaurer.bsky.social · Jan 9

Perfect for tasks like stepwise reasoning.
🔀 Model merging: A new callback leverages mergekit to merge models during training, improving performance by blending reference and policy models - optionally pushing merged models to the Hugging Face Hub.

1

Moritz Laurer @moritzlaurer.bsky.social · Jan 9

The TRL v0.13 release is 🔥! My highlight are the new process reward trainer to train models similar to o1 and tool call support:

🧠 Process reward trainer: Enables training of Process-supervised Reward Models (PRMs), which reward the quality of intermediate steps, promoting structured reasoning.

1 2

Moritz Laurer @moritzlaurer.bsky.social · Jan 7

techcrunch.com/2025/01/05/o...

OpenAI is losing money on its pricey ChatGPT Pro plan, CEO Sam Altman says | TechCrunch

OpenAI CEO Sam Altman says that the company is currently losing money on its $200-per-month plan because people use it more than expected.

techcrunch.com

Moritz Laurer @moritzlaurer.bsky.social · Jan 7

on revenue of $3.7 billion last year, with ChatGPT alone once costing an estimated $700,000 per day to operate. 💸🔥
- They build strong models and do great research. Whether this business model will work in the long run is one of the biggest questions in the AI economy.

Source with the numbers 👇

OpenAI is losing money on its pricey ChatGPT Pro plan, CEO Sam Altman says | TechCrunch

OpenAI CEO Sam Altman says that the company is currently losing money on its $200-per-month plan because people use it more than expected.

techcrunch.com

1

Moritz Laurer @moritzlaurer.bsky.social · Jan 7

OpenAI is losing money on the $200/month subscription 🤯. It's crazy how expensive it is to run these largest LLMs:

- ChatGPT Pro costs $200/month ($2,400/year) and is still unprofitable for OpenAI due to higher-than-expected usage.
- OpenAI reportedly expected losses of about $5 billion

1 4 4

Moritz Laurer @moritzlaurer.bsky.social · Jan 6

Base model: huggingface.co/MoritzLaurer...
Large model: huggingface.co/MoritzLaurer...
Updated zeroshot collection: huggingface.co/collections/...
ModernBERT collection with paper: huggingface.co/collections/...

MoritzLaurer/ModernBERT-base-zeroshot-v2.0 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

Moritz Laurer @moritzlaurer.bsky.social · Jan 6

Great work by @answerdotai !

If you’re looking for a high-speed zeroshot classifier, give it a try!

📄 Resources below: 👇

1

Moritz Laurer @moritzlaurer.bsky.social · Jan 6

- 💡 What’s next? I’m preparing a newer version trained on better + longer synthetic data to fully leverage the 8k context window and improve upon the training mix of my older zeroshot-v2.0 models. I also hope that there will be a multilingual variant in the future.

1 1

Moritz Laurer @moritzlaurer.bsky.social · Jan 6

- 📉 Performance tradeoff: It performs slightly worse than DeBERTav3 on average across my zeroshot classification task collection
- 🧠 Use cases: I recommend using it for scenarios requiring speed and a larger context window (8k).

1

Moritz Laurer @moritzlaurer.bsky.social · Jan 6

🚀 Releasing a new zeroshot-classifier based on ModernBERT! Some key takeaways:

- ⚡ Speed & efficiency: It's multiple times faster and uses significantly less memory than DeBERTav3. You can use larger batch sizes and enabling bf16 (instead of fp16) gave me a ~2x speed boost
- 📉 Performance tradeoff:

1 1 4

Moritz Laurer @moritzlaurer.bsky.social · Dec 20

Congrats @answerdotai, @LightOnIO and collaborators like @tomaarsen.com !

Paper and models here 👇https://huggingface.co/collections/answerdotai/modernbert-67627ad707a4acbf33c41deb

1

Moritz Laurer @moritzlaurer.bsky.social · Dec 20

Quite excited by the ModernBERT release! 0.15/0.4B small, 2T modern pre-training data and tokenizer with code, 8k context window, great efficient model for embeddings & classification!

This will probably be the basis for many future SOTA encoders! I can finally stop using DeBERTav3 2021 :D

1 1 8