[mickael-kerjean]

Genuine question: how reproducible / usable / verifiable are these architectures from the published documentation? Are they similar to PDF/DWG/PSD specifications, where the format look like an open spec at first sight until you attempt to implement it and realize the crucial implementation details are undocumented?

[nl]

It's entirely reproducible from the available documentation (which is why you see vLLM, SGLang, MLX etc all racing to produce optimized implementations).

(As an aside, this is why the "open weights are not open source" thing is a complete misunderstanding. The weights themselves along with the documentation give you enough to fine tune the LLM. You can't rebuild it from scratch, but you can't do this even with the data anyway (because of randomness!))

[FinchNova12]

I agree that if you have the weights you can use/train a model with the same architecture, and that you won't get the exact weights on your own due to randomness. But isn't data an extremely important part of your ability to effectively train/finetune? It might be much harder to get close to the level of the open weight model if you don't have the data that made it, which is why I think the open weights vs open source distinction is useful.

[esperent]

> open weights are not open source" thing is a complete misunderstanding

> You can't rebuild it from scratch

There's is extremely clear and misunderstanding-free.

Open weights is not open source.

[calebkaiser]

Implementing models directly from papers is typically pretty doable (and is of course more straightforward when the full implementation is open sourced). Often there is some amount of specific knowledge, like particular hyperparameters, that is missing and has to be trial and errored by the community, but generally speaking, getting the core model architecture implemented is a reasonable task for most well documented models.

Reproducing the exact training run, however, is basically impossible without the original dataset and training pipeline (here meaning all of the code + infra involved in actually executing the pre and post training loops). Also, it would be exorbitantly expensive to do if you weren't also a lab trying to train a similar model.

But you can still scale the architecture down and experiment as a solo researcher using the published research. There are probably some open source implementations already on GitHub for any given big open model release.

[dannyw]

The architectures are high level concepts and the mechanics usually have enough detail for you to try and implement.

Transformers are very "mendable" in that you can permute the architecture in crazy or random ways, and still basically always end up with get a coherent LLM. The difference comes down to training efficiency, inference efficiency, and usually minor differences in performance.

Hyperparams and stuff, I mean it's standard to do a sweep anyway.

[GaggiX]

The code is open source, and there probably several different implementations.