lodestones/Kroma
Text-to-Image • Updated • 400
The Brainrotted Nerd
DontPutFishInIt
AI & ML interests
None yet
Recent Activity
liked a model 4 days ago
lodestones/Kroma liked a model 6 days ago
ukisai/Swift-1.5-Qwen3.8-Flash-Next-GGUF liked a model about 2 months ago
Qwen/Qwen3.8-Flash-NextOrganizations
liked a
model 4 days ago
liked a
model 6 days ago
liked 2
models about 2 months ago
Will we ever get a 12b or 26b version for those of us with 12-16gb vram?
👍 1
5
#2 opened 2 months ago
by DontPutFishInIt
liked 2
datasets 3 months ago
liked a
model 3 months ago
liked 2
models 4 months ago
liked 3
models 5 months ago
reacted to omarkamali's post with 👍 6 months ago
Post
4642
We got Qwen 3.5 to count Rs in Strawberry correctly! 🚨
Building on Sawtone, we’ve been testing a different way to feed language into an LLM to build the next generation of multilingual AI.
The usual setup gives the model tokenized text and asks it to perform various linguistic tasks. That works surprisingly well, until it doesn’t. Accents disappear. Words get mangled. Internal structure gets blurred away. And the cost of that gets higher once you move into multilingual and lower-resource settings.
So we tried adding a second path.
In addition to the normal text input, the model also receives Sawtone: a byte-level word representation that preserves how a word is written, how it sounds, and how it is structured.
Same LLM. Better interface.
In this proof of concept with Qwen 3.5 0.8B, that pushed our eval from 64% to 88%. The gains showed up exactly where tokenized models usually get shaky: diacritics, character order, exact spelling, and other form-sensitive behavior.
Sawtone itself is tokenizer-free, byte-level, and pre-trained across 507 languages.
Still early, but promising!
Building on Sawtone, we’ve been testing a different way to feed language into an LLM to build the next generation of multilingual AI.
The usual setup gives the model tokenized text and asks it to perform various linguistic tasks. That works surprisingly well, until it doesn’t. Accents disappear. Words get mangled. Internal structure gets blurred away. And the cost of that gets higher once you move into multilingual and lower-resource settings.
So we tried adding a second path.
In addition to the normal text input, the model also receives Sawtone: a byte-level word representation that preserves how a word is written, how it sounds, and how it is structured.
Same LLM. Better interface.
In this proof of concept with Qwen 3.5 0.8B, that pushed our eval from 64% to 88%. The gains showed up exactly where tokenized models usually get shaky: diacritics, character order, exact spelling, and other form-sensitive behavior.
Sawtone itself is tokenizer-free, byte-level, and pre-trained across 507 languages.
Still early, but promising!
- 5 replies
liked a
model 7 months ago
New activity in Dans-DiscountModels/32b-glm4-dans-personality-engine-v1.3.0-TestArticle-1 7 months ago
GGUF?
👀 1
#1 opened 7 months ago
by DontPutFishInIt
any chance of a larger successor model? (24B, 32B, ETC.)?
#2 opened 7 months ago
by DontPutFishInIt
liked a
Space 10 months ago
liked a
model about 1 year ago
liked a
Space about 1 year ago