Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learning · HackerPapers discussions

Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learning — HN submission

Fri, 10 Oct 2025 05:02:48 GMT

Ultrameta Modelling

Hi HN. There are highly specialized languages models abounding. Two that have been personally interesting are: https://arxiv.org/abs/2311.13721 And Today, https://www.reddit.com/r/LocalLLaMA/comments/1o27xsj/introducing_playable1gguf_by_far_the_worlds_best/ The obvious extension of generating models with large corpi, eventually, is generation of models of models. The LMM, the large model model, is trained on models, such as the content of hugging face. Right? All the fine tuners and abliterators, and every variation of a model can be used as training data for eventual large model models.

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learning — mentioned by pjmlp

Wed, 07 May 2025 12:35:13 GMT

Exactly, I also see code generation to current languages as output only an intermediary step, like we had to have those -S switches, or equivalent, to convince developers during the first decades of compiler existence, until optmizing compilers took over. "Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learning" https://arxiv.org/html/2311.13721v3

Paper and discussion history. A mention is not an endorsement.

Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learning — mentioned by pjmlp

Fri, 02 May 2025 14:18:04 GMT

Well, that won't stop folks trying though. "Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learning" https://arxiv.org/html/2311.13721v3

Paper and discussion history. A mention is not an endorsement.