2 11 4

spuun is trying

spuun

https://discord.thisisartunion.com

AI & ML interests

NLG and AI applications for artistic purposes. Specifically for the Art Union Discord server.

Recent Activity

published a Space 7 days ago

spuun/mwsamanaga

reacted to grimjim's post with 🚀 14 days ago

This recent paper points to an explanation for the unreasonable effectiveness of Frankenmerges: https://huggingface.co/papers/2502.05171 Specifically, the duplication of layers in Frankenmerges serves a purpose similar to what occurs in their recurrent-depth architecture. Successful frankenmerges that operate without additional fine-tuning are able to recover or "heal" from any damage due to abrupt transitions between layer blocks. Operational replicated layer blocks can provide functional benefits grounded in latent reasoning. Frankenmerges can also result in hybrid reasoning, by splicing together the latent reasoning of different models. Back in April 2024, I was able to duplicate a few layers in the Llama 3 8B model, turning it into a 9B model, without harming benchmarks significantly, despite any transition damage. https://huggingface.co/grimjim/llama-3-experiment-v1-9B My informal experimentation suggested that latent reasoning circuits could occupy continguous stacks of 2-4 layers, though the result was highly sensitive to the choice of transition location between layers.

upvoted a paper 14 days ago

Can 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling

View all activity

Organizations

None yet

spuun's activity

published a Space 7 days ago

Mwsamanaga

🏆

Some other CTF thingy, dw abt it

reacted to grimjim's post with 🚀 14 days ago

Post

2038

This recent paper points to an explanation for the unreasonable effectiveness of Frankenmerges: Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach (2502.05171)

Specifically, the duplication of layers in Frankenmerges serves a purpose similar to what occurs in their recurrent-depth architecture. Successful frankenmerges that operate without additional fine-tuning are able to recover or "heal" from any damage due to abrupt transitions between layer blocks. Operational replicated layer blocks can provide functional benefits grounded in latent reasoning. Frankenmerges can also result in hybrid reasoning, by splicing together the latent reasoning of different models.

Back in April 2024, I was able to duplicate a few layers in the Llama 3 8B model, turning it into a 9B model, without harming benchmarks significantly, despite any transition damage.
grimjim/llama-3-experiment-v1-9B
My informal experimentation suggested that latent reasoning circuits could occupy continguous stacks of 2-4 layers, though the result was highly sensitive to the choice of transition location between layers.