sarvamai/prism
just landed · new, unvettedmodel
New, unvetted: created or first seen in the last few days, not from a trusted org and not past the signal threshold yet.
Sarvam-Prism is a decision model built on Sarvam-105B. Mixture-of-experts, 105B total (128 experts, 8 per token). Context: 128K tokens (config). License: apache-2.0.
- Org
- sarvamai
- Released
- 2026-11-02
- Status
- just landed · new, unvetted
- License
- apache-2.0 (permissive)
- Parameters
- 105.0B
- Architecture
- MoE · 128 experts, 8/token · SarvamMLAForCausalLM
- Context
- 131,072 tokens
- Task
- text-classification · text
- Signal
- 18 HF likes · 0 downloads
Runs on
estimate from params; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~64.3 GB (estimated size): 96GB GPU (RTX Pro 6000 Blackwell), Strix Halo 128GB unified (~96GB usable), Mac 96GB, Mac 128GB
- Q8 ~112.9 GB (estimated size): 1× DGX Spark (128GB), Mac 192GB, Mac 256GB, CPU only, 128GB RAM (slow)
- FP16 ~211.3 GB (estimated size): 2× DGX Spark (256GB), Mac 512GB, CPU only, 256GB RAM (slow)
Links
First published here 2026-11-02 · updated 2026-11-04 · markdown · JSON