deca-exp/3-ultra-2511
Updated
AI & ML interests
All Deca Experiments here. See @deca_ai for the good stuff.
Recent Activity
ccocks-deca
updated
a
model
21 days ago
ccocks-deca
posted
an
update
3 months ago
Post
290
📢 New Dataset: OpenSynth Battles
We've released OpenSynth Battles, a benchmark dataset featuring generations from five large language models on shared prompts. Each prompt includes:
Responses from:
gpt-oss-120b, deepseek-v3.1-thinking, deepseek-v3.1-instruct, moonshotai/kimi-k2-instruct, and deepseek-r1-0528
Automated scoring by gpt-oss-120b
Useful for model comparison, automated evaluation research, and prompt-level performance analysis.
No data splits included.
🔗 https://huggingface.co/datasets/ccocks-deca/open-synth-battles
We've released OpenSynth Battles, a benchmark dataset featuring generations from five large language models on shared prompts. Each prompt includes:
Responses from:
gpt-oss-120b, deepseek-v3.1-thinking, deepseek-v3.1-instruct, moonshotai/kimi-k2-instruct, and deepseek-r1-0528
Automated scoring by gpt-oss-120b
Useful for model comparison, automated evaluation research, and prompt-level performance analysis.
No data splits included.
🔗 https://huggingface.co/datasets/ccocks-deca/open-synth-battles
ccocks-deca
posted
an
update
4 months ago
Post
2870
12 hours ago:
Annnnnd... here it is! https://huggingface.co/deca-ai/3-alpha-ultra —the largest AI model in the world by
deca-ai
, clocking in at a whopping 4.6T parameters. Apologies for the delay, but we’re stoked to finally drop this, even in its alpha stage. Before you dive in, here are a few things to keep in mind:
1. **No commercial use yet**: We're still working on Deca 2.5 (Proprietary), and releasing Deca 3 for commercial use right now would impact that. Once Deca 3.5 hits in early '26, we’ll be opening it up with a more permissive license.
2. **Built on existing models**: Deca 3 isn’t a ground-up creation—it’s a huge step forward, building on what’s already out there.
3. **It’s experimental**: As much as we’re hyped about its scale, it’s still in testing.
4. **DynaMoE architecture**: Run a (very) small part of the model with 64GB of RAM/VRAM (when quantized - quants coming soon), or the whole thing with 1TB. It’s that scalable.
5. **Not widely supported yet**: Frameworks like vLLM and Transformers aren’t compatible with Deca 3 at the moment, so until we drop the DynaMoE software (beta coming soon), it’s mostly just a concept.
We’re super excited to see what you do with it once the full setup’s ready. Hang tight, and stay tuned!
Something big* coming
*big = biggest in the worldAnnnnnd... here it is! https://huggingface.co/deca-ai/3-alpha-ultra —the largest AI model in the world by
deca-ai
, clocking in at a whopping 4.6T parameters. Apologies for the delay, but we’re stoked to finally drop this, even in its alpha stage. Before you dive in, here are a few things to keep in mind:1. **No commercial use yet**: We're still working on Deca 2.5 (Proprietary), and releasing Deca 3 for commercial use right now would impact that. Once Deca 3.5 hits in early '26, we’ll be opening it up with a more permissive license.
2. **Built on existing models**: Deca 3 isn’t a ground-up creation—it’s a huge step forward, building on what’s already out there.
3. **It’s experimental**: As much as we’re hyped about its scale, it’s still in testing.
4. **DynaMoE architecture**: Run a (very) small part of the model with 64GB of RAM/VRAM (when quantized - quants coming soon), or the whole thing with 1TB. It’s that scalable.
5. **Not widely supported yet**: Frameworks like vLLM and Transformers aren’t compatible with Deca 3 at the moment, so until we drop the DynaMoE software (beta coming soon), it’s mostly just a concept.
We’re super excited to see what you do with it once the full setup’s ready. Hang tight, and stay tuned!
-
- 3 replies
ccocks-deca
posted
an
update
4 months ago
Post
342
Something big* coming...
big = biggest in the world
big = biggest in the world
-
- 3 replies
ccocks-deca
posted
an
update
5 months ago
Post
2597
🚀 Deca 3 Ultra Alpha is coming in the next 72 hours! 🚀
We're on the verge of something monumental. Right now, we're in the final stages of testing, and we're about to drop a game-changing milestone in the open-source AI community. 🎉
In just two weeks, we've managed to almost 4x the size of the largest open-source LLM at that time (and we are still 2.6x bigger than the largest LLM). This is unprecedented and a testament to the power of collaboration, innovation, and the relentless pursuit of pushing AI to its limits.
The future of open-source AI is now. Stay tuned for the release – we’re just getting started.
- Model testing finishes: 24hrs from now
- Model gets uploaded: 30hrs from now
- Related code/inference stack gets published: 70-90hrs from now
We're on the verge of something monumental. Right now, we're in the final stages of testing, and we're about to drop a game-changing milestone in the open-source AI community. 🎉
In just two weeks, we've managed to almost 4x the size of the largest open-source LLM at that time (and we are still 2.6x bigger than the largest LLM). This is unprecedented and a testament to the power of collaboration, innovation, and the relentless pursuit of pushing AI to its limits.
The future of open-source AI is now. Stay tuned for the release – we’re just getting started.
- Model testing finishes: 24hrs from now
- Model gets uploaded: 30hrs from now
- Related code/inference stack gets published: 70-90hrs from now
-
- 8 replies