Can you run Sarvam 105B locally?

Sarvam family · 2026 · 106B parameters (10.3B activated per token) · Hugging Face model card

How much VRAM does Sarvam 105B need at each quantization?

Sarvam 105B needs 70.0 GB of VRAM at Q4_K_M. The table below lists weights-only size and total VRAM including overhead for each common quantization level with 4k and 32k token contexts.

Multi-head latent attention (sarvam_mla) with kv_lora_rank compression; config exposes no num_key_value_heads. Head dim 576 is the MLA combined q/kv projection dim.

Method: weights = parameters × bits-per-weight (Q4_K_M ≈ 4.8, Q5_K_M ≈ 5.7, Q6_K ≈ 6.6, Q8_0 ≈ 8.5, FP16 = 16), plus 10% loading overhead, plus KV cache from the verified architecture config (32 layers, KV heads, 576 head dim). Source: HF-verified 2026-09-30 via plain api (gated=false safetensors.total=106031767424 params=106.03B ctx=131072 license=apache-2.0 repo=sarvamai/sarvam-105b@abe264bdb8fb2244daadaa59ebc2020b3c139e89 layers=32 hidden=4096 kv_heads=None heads=64 vocab=262144). MoE: num_experts=128 num_experts_per_tok=8 active_params_b=10.3 (vendor model card (10.3B active); designation 105b; config num_experts=128 routed 8 per token). VRAM figures are documented calculations (Q4-class ~0.6 GB per B params, on active params for MoE rows), not measurements. release_date is the HF repo lastModified date observed on the verification pass. No tok/s or benchmark scores invented.

Which GPUs can run Sarvam 105B locally?

At Q4_K_M with a 4k context, Sarvam 105B needs 70.0 GB of VRAM. The lists below are computed live from our GPU database and grouped by how much headroom the card has. Disclosure: CompareAIHardware.com participates in the Amazon Associates program and earns from qualifying purchases through links on this page. Affiliate relationships do not influence our recommendations.

Runs Sarvam 105B comfortably (25%+ VRAM headroom)

Minimum GPUs that fit Sarvam 105B at Q4_K_M

These GPUs hold the model but leave little headroom — keep contexts short.

Needs 2+ GPUs to run Sarvam 105B

One of these cards is too small on its own, but a pair (tensor or pipeline parallel, ~90% efficiency) covers the 70.0 GB requirement. See our multi-GPU guide for setup.

Can a Mac run Sarvam 105B?

Yes — these Apple Silicon machines fit Sarvam 105B at Q4_K_M, since macOS lets the GPU use about 75% of unified memory. Generation speed is bound by memory bandwidth, so the GB/s column matters as much as capacity.

Can Sarvam 105B run on mini PCs, Jetson, or NPU devices?

These edge and NPU devices from our database have enough memory for Sarvam 105B at Q4_K_M. Their memory bandwidth is far below discrete GPUs, so expect a fraction of desktop generation speed.

Which pre-built systems can run Sarvam 105B?

These mini PCs and workstations from our database fit Sarvam 105B at Q4_K_M. Usable-memory figures are conservative: Windows shares about half of system RAM with the GPU by default (Linux can expose more), macOS lets Apple Silicon GPUs use about 75% of unified memory, and Linux unified-memory systems such as GB10 expose roughly 90%. Generation speed is bound by memory bandwidth, so compare the GB/s column before buying.

Frequently asked questions