Issues related to the tokenizer and chat template are the main reasons why merged models underperformed on the Open LLM Leaderboard v2. The evals don't change anything. A model merged to maximize MMLU will also perform well on these new evals.
LLMS
-
NeuralDaredevil LLM Outperforms Llama 3.1 Models
By
–
Haha NeuralDaredevil is a beast, outperforming most Llama 3.1 models. All of that despite being based on Llama 3.0 and the performance drop due to the abliteration process. Thanks to Dampfinchen for fixing an issue with the tokenizer that messed up the previous IFEval score.
-
Fine-tuning and Model Merging Techniques Explained
By
–
Excellent breakdown of my fine-tuning and model merging talk by @TheTuringPost 🙂
-
15 ChatGPT Prompts to Land Your Dream Job Faster
By
–
I'm shocked people still don't use ChatGPT for job search! ChatGPT can help you land your dream job twice as fast. Copy and paste these 15 ChatGPT prompts to land your Dream Job:
-
Claude may inject invisible copyright-avoidance instructions
By
–
Claude sometimes invisibly adds instructions to avoid reciting copyrighted works to the user’s prompt, presenting them to the model as though the user wrote them. Thus, these rules need not apply to Batman:
-
LLMs Training on Book Summaries and Study Materials
By
–
LLMs would *already* have trained on stuff like this for major books, since they would have seen 2 page summaries, 'Cliff's Notes', chapter summaries, etc.
-
LLMs Recursive Summarization Capabilities and Limitations
By
–
That's an interesting hypothesis! What do you think of this counter-point: AI systems today can get very close to your idea in training, by using an LLM to recursively summarise a long document.
-
Commentary on LLM prompting and counting reliability
By
–
As many have pointed out, there are much better ways to prompt this. Also, it’s silly to use an LLM here, as there’s little hope of making them 100% reliable at counting. What’s notable IMO isn’t that it can’t count, but that it doesn’t see it can’t count (and e.g. try its REPL)
-
Fine-tuning and Merging LLMs: Techniques and Best Practices
By
–
If you're interested in fine-tuning and merging LLMs, here's a well-edited video of a talk I gave in July. We talk about: – How to create a dataset
– SFT techniques
– How to merge models Thanks to @Tunehq_ai for the invitation! https://
youtu.be/8aFqLVQjHTw?si
=K8KpM55G_93wwZmM
… -
ChatGPT-4o does the unimaginable and is the best
By
–
ChatGPT-4o can do what people can't even imagine. Its just the best