StableLM VS sparsegpt

https://github.com/Stability-AI/StableLM?tab=readme-ov-file#...

https://mistral.ai/news/announcing-mistral-7b/

looking at the 3b results (here https://github.com/Stability-AI/StableLM#stablelm-alpha-v2 ?), it looks like Mistral (which outperforms Llama-2 13b) is far more powerful

Does this mean Stability gave up on StableLM?

I notice that the repo hasn’t been updated since April, and a question asking for an update has been ignored for at least a month: https://github.com/Stability-AI/StableLM/issues/83

I'm not "ignoring" StableLM, if anything it's the impetus for my post. The alpha models were so bad and unusable that it seems they may have simply abandoned the project. It's clear they basically didn't know what they were doing, which is silly for a company of their size and specialization.

1) StableLM released a checkpoint at 800B for their 3B and 7B at 800B tokens with 4096 context size, but perform very poorly on different benchmarks and finetuning is discouraged with such a weak base model

It is the best open-source model currently available. Falcon-40B outperforms LLaMA, StableLM, RedPajama, MPT, etc. See the OpenLLM Leaderboard.

Someone shared this link on another recent post

Github: https://github.com/Stability-AI/StableLM

SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot (https://arxiv.org/abs/2301.00774)

There is this : https://github.com/IST-DASLab/sparsegpt

Check the paper here, it's intersting: https://arxiv.org/abs/2301.00774

There's no chance that we've peeked from a bang for buck sense - we still haven't adequately investigated sparse networks.

Relevantish: https://arxiv.org/abs/2301.00774

The fact that we can reach those levels of sparseness with pruning also indicates that we're not doing a very good job of generating the initial network conditions.

Being able to come up with trainable initial settings for sparse networks across different topologies is hard, but given that we've had a degree of success with pre-trained networks, pre-training and pre-pruning might also allow for sparse networks with minimally compromised learning capabilities.

If it's possible to pre-train composable network modules, it might also be feasible to define trainable sparse networks with significantly relaxed topological constraints.

Training uses gradient descent, so you want to have good precision during that process. But once you have the overall structure of the network, https://arxiv.org/abs/2210.17323 (GPTQ) showed that you can cut down the precision quite a bit without losing a lot of accuracy. It seems you can cut down further for larger models. For the 13B Llama-based ones, going below 5 bit per parameter is noticeably worse, but for 30B models you can do 4 bits.

The same group did another paper https://arxiv.org/abs/2301.00774 which shows that in addition to reducing the precision of each parameter, you can also prune out a bunch of parameters entirely. It's harder to apply this optimization because models are usually loaded into RAM densely, but I hope someone figures out how to do it for popular models.