sequelbox/Superpotion-DeepSeek-V3.2-Speciale
Viewer
• Updated
• 57.8k • 39 • 7
t.d.a.g. PRO
sequelbox
AI & ML interests
open source, infinite games. (they/them)
Recent Activity
replied to their post about 9 hours ago
IMPORTANT corrective note/apology: the initial upload of ValiantLabs/Qwen3.5-27B-Esper3.1 contained improperly merged weights, meaning it was effectively just Qwen 3.5. We've merged properly now and re-uploaded the correct weights to the existing repository.
The model link is here: https://huggingface.co/ValiantLabs/Qwen3.5-27B-Esper3.1
This is not at all the fault of Qwen 3.5, transformers, or anything other than our own flawed upload pipeline and insufficient post-upload validation. We have to do better. We'll immediately improve our validation procedures to be more rigorous. This will never happen again.
We are proud to build quickly: by providing a quick finetune of a new high-performance model like Qwen 3.5, we seek to provide value not only in the model's direct use but especially to our fellow open-source creators, who can learn from our initial training attempt. We are proud that building fast helps you build fast. In this case, our poor work has produced the opposite result - we've wasted your time. We are really sorry to everyone, but especially our fellow builders.
We feel about one inch tall right now, but we're going to get back to work and do better. Our crew deserves better and so do our users.
Humbly, your captain,
t.d.a.g. posted an
update
about 9 hours ago
IMPORTANT corrective note/apology: the initial upload of ValiantLabs/Qwen3.5-27B-Esper3.1 contained improperly merged weights, meaning it was effectively just Qwen 3.5. We've merged properly now and re-uploaded the correct weights to the existing repository.
The model link is here: https://huggingface.co/ValiantLabs/Qwen3.5-27B-Esper3.1
This is not at all the fault of Qwen 3.5, transformers, or anything other than our own flawed upload pipeline and insufficient post-upload validation. We have to do better. We'll immediately improve our validation procedures to be more rigorous. This will never happen again.
We are proud to build quickly: by providing a quick finetune of a new high-performance model like Qwen 3.5, we seek to provide value not only in the model's direct use but especially to our fellow open-source creators, who can learn from our initial training attempt. We are proud that building fast helps you build fast. In this case, our poor work has produced the opposite result - we've wasted your time. We are really sorry to everyone, but especially our fellow builders.
We feel about one inch tall right now, but we're going to get back to work and do better. Our crew deserves better and so do our users.
Humbly, your captain,
t.d.a.g. liked
a model about 9 hours ago
ValiantLabs/Qwen3.5-27B-Esper3.1 Organizations
