bottlecapai/ThinkingCap-NVIDIA-Nemotron-3.5-Lightning-30B-A3B
archived · low-signalmodel
ThinkingCap uses 23% fewer thinking tokens and completes 25 tasks in that time, compared to 11 for the base model. Mixture-of-experts, 32B total, 3B active (128 experts, 6 per token). Context: 256K tokens (config). License: openmdw-1.1.
- Org
- bottlecapai
- Released
- 2026-10-07
- Status
- archived · low-signal
- License
- openmdw-1.1 (permissive)
- Parameters
- 31.6B total, 3.0B active
- Architecture
- MoE · 128 experts, 6/token · NemotronHForCausalLM
- Context
- 262,144 tokens
- Task
- text-generation · text
- Signal
- 9 HF likes · 166 downloads
- Archived
- 2026-10-24 (no longer listed by the collector; stats as last seen)
Runs on
real quant files; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~26.8 GB: 32GB GPU (5090), Mac 36GB, Mac 48GB
- Q8 ~36.4 GB: 48GB GPU (RTX 6000 Ada / A6000), Mac 64GB (M1 Max), CPU only, 64GB RAM (slow)
- FP16 ~67.3 GB: 96GB GPU (RTX Pro 6000 Blackwell), 1× DGX Spark (128GB), 2× DGX Spark (256GB), Strix Halo 128GB unified (~96GB usable), Mac 96GB, Mac 128GB, Mac 192GB, Mac 256GB, Mac 512GB, CPU only, 128GB RAM (slow), CPU only, 256GB RAM (slow)
Links
First published here 2026-10-07 · updated 2026-10-24 · markdown · JSON