sometimesanotion

Now imagine this as a hashtag generator and so a RAG search can find great context. :)

Neat! I've transitioned from wanting more from a model's one-shot answers to breaking things down and walking through the problem with cached context. This effectively means simulating most of the thinking block, but by tool usage and RAG.

I'm happily using our models from months ago to do it. If anything - even Lamarck 0.7's use of thinking blocks are a bit much. I'm using Lamarck 0.7 Fusion (my best GPQA model, though it didn't break your record and is best used where modest IFEVAL isn't a blocker) and /nothink with ValiantLab's Qwen3 models in concert.

I suspect I'll try some merges soon to give this toolchain better models, leaderboard or no leaderboard!

I've been using Esper3 8B and 14B for first-pass code review. I am quite pleased.

Have you considered fine-tuning a 1.7B or smaller model for autocomplete?

I've been thinking a lot about using small caches of embeddings for local RAG lately. Have you considered an HTTP caching proxy like Squid as a low-impact source? It would retrieve what a user is reading anyway, and what's in their field of interest. A browser extension to signal some limited ingestion when a page is bookmarked might fit a lot of use cases.

For many reasons, smart management of context windows is my top priority with AI now!

@Inschrift-Spruch-Raum ,I am looking through recent PRs to mergekit, and I am optimistic that Lamarck's recipes will be working again soon!

When that happens, there will be two efforts: one to make a compelling non-CoT model, and another to blend CoT in right amounts.

Lamarck's multilingual capabilities improved noticeably from light influence of Krystalan/DRT-14B in v0.6, and merging from other CoT models like DeepSeek R1 is a matter of careful moderation. I will always put the overall apparent quality of translation, prose, and reasoning first.

No worries! See, I agree, the recipe behind Lamarck is pretty good, and there's a lot more to get out of it. It'll likely depend on getting multiple mergekit versions working on the pipeline. The new mergekit's fusion and sce merges offer some interesting potential, but I use fine-grained sliced merges to control the mix of branches, which last I checked, work only with older mergekit and bitsnbytes.

By now there are ample upgrades to try. I did feel Lamarck v0.7 was a proof-of-concept and had plenty of headroom to grow!

Yes, there's a reason for that! I ran into a mergekit bug with the use of slices and della_linear merge methods which are key to intentionally crafting Lamarck's releases. @Crystalcareai , @Arcee , are there fixes in the queue?

https://huggingface.co/posts/sometimesanotion/507492798113402

The numbers are in! The results are fascinating.

Though IFEVAL skewed low compared to the ancestor model's average, and Lamarckvergence's improved MATH didn't come through, this model is strong in several ways. The GPQA score suggests as much. These are scores I'm pretty sure I can improve without giving up much of the interesting gains.

What's more, my subjective impression is that its prose and consistency get a boost from Chocolatine. @jpacifico , I think arcee_fusion is a merge method that has a lot to offer for your future base models! This also bodes very well for the next several merges to come.

I've been doing all my LoRA work on AMD hardware with Linux; I'm looking forward to your notes! I sometimes still do it on CPU because it's easy to renice the task priority so the foreground tasks stay snappy.

The main challenge I have is keeping a solid ROCm bitsandbytes install when other packages want updates.

Thank you! If you liked this, and want 14B models with even smoother command of French, I'm a huge fan of https://huggingface.co/jpacifico/Chocolatine-2-14B-Instruct-v2.0.3!

Right-sizing language models is something I'm really here for. I find that a 1.5B parameter model fronting simple questions from a backing RAG source that a larger model gradually works on is more scalable. Classic information sources and stores can be QA'd, and they don't have such huge energy footprints.

AI will work out better if we give humans, classic code, SLMs, and frontier LLMs the roles they're right-sized for, and ensure data privacy and individual dignity at every stage of the contract.