OmniSenter-Base 16B — the Omni base¶
Status: ✅ published (transitional v1) · HF:
sovthpaw/OmniSenter-Base-16B
Identity¶
| Full name | OmniSenter-Base 16B |
| Type | Multimodal base (Darwin-merged, no agentic core) |
| Total params | 16B |
| Active per token | 16B (dense) |
| Context window | standard |
| Modalities | text + vision + audio + video (text-generation pipeline) |
| Self-evolution | no |
What it is¶
A 16B multimodal base model. Darwin-merged from: - Qwen3-8B (text backbone, agentic capability) - nvidia/Cosmos3-Nano (multimodal capability, vision + video)
It's the Omni base that future OmniSenter 12B and Senter Ohm 32A8B models will be built on top of. Pre-Senter — no notebook, no function calling, no agentic core.
Composition¶
| Component | Source | Role |
|---|---|---|
| Text backbone | Qwen3-8B | Reasoning, instruction following |
| Multimodal | Cosmos3-Nano | Vision, video, audio understanding |
| Total | Darwin-merged | 16B, 7 safetensors shards |
Status: transitional¶
This is the v1 Omni base. When the new architecture ships (OmniSenter 12B as the small Senter, Senter Ohm 32A8B as the flagship), this base will be superseded by the new releases.
Use it now as the multimodal base for downstream Senter fine-tunes. When the new architecture's base ships, switch over.
See also¶
- HF:
sovthpaw/OmniSenter-Base-16B - OmniSenter concept
- Related: OmniSenter 12B · Senter Ohm 32A8B
- Blog:
../../blog/the-omni-family.md(the naming refactor) - Blog:
../../blog/the-omnimodal-fusion.md(the three-component fusion that produces this)