How much GPU memory does sarvam-m need?

About 56.1 GB for conversations of up to 8,192 tokens: 47.14 GB of weights, the cache for one full-length conversation and the runtime's own memory.

What is the cheapest GPU to run sarvam-m?

Right now one Nvidia A800 at $0.90 an hour; RTX PRO 6000 at $1.15 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.

Can sarvam-m run on an RTX 4090?

Not on one: it needs about 56.1 GB and an RTX 4090 has 24 GB. Use a bigger GPU, or a smaller (quantized) version of the model.

How do I deploy sarvam-m as an API?

Pick a machine on its GridPup page and press Deploy: GridPup rents the GPU, downloads the model from Hugging Face and serves it at an https address with keys and OpenAI's chat API, billed by the hour.

Models / sarvam-m / sarvam-m ~23.6B

sarvamai/sarvam-m

Text generation · served with vLLM · ↓ 3.2K downloads a month · View on Hugging Face

sarvamai/sarvam-m needs about 56.1 GB of GPU memory for conversations of up to 8,192 tokens. It runs on one Nvidia A800 from $0.90/h right now; best value: RTX PRO 6000 at $1.15/h.

Weights: 47 GB
GPU memory needed: about 56.1 GB
Conversation length: 8,192 tokens
Model allows: 32,768 tokens
Memory · disk: 31.6 GB · 92 GB
Architecture: MistralForCausalLM
Licence: apache-2.0

Versions of sarvam-m ~23.6B

The same model packaged differently: a quantized version needs less GPU memory and runs on a cheaper machine.

Version	Runtime	Weights	GPU memory	Cheapest machine now
Originalsarvamai/sarvam-m	vLLM	47 GB	56.1 GB	$0.89/h Nvidia A800
onnx (Prince-1)Prince-1/sarvam-m-onnx	ONNX Runtime GenAI	47 GB	57.1 GB	$0.89/h Nvidia A800

Hardware and deploy

Endpoint name

$1.15/h from your balance while it runs, billed by the hour; stop it from its page. GridPup rents the machine, which downloads the model from Hugging Face (usually ready in 10 to 15 minutes), and gives you an https address, keys and a Playground.

Questions

How much GPU memory does sarvam-m need?: About 56.1 GB for conversations of up to 8,192 tokens: 47.14 GB of weights, the cache for one full-length conversation and the runtime's own memory.
What is the cheapest GPU to run sarvam-m?: Right now one Nvidia A800 at $0.90 an hour; RTX PRO 6000 at $1.15 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.
Can sarvam-m run on an RTX 4090?: Not on one: it needs about 56.1 GB and an RTX 4090 has 24 GB. Use a bigger GPU, or a smaller (quantized) version of the model.
How do I deploy sarvam-m as an API?: Pick a machine on its GridPup page and press Deploy: GridPup rents the GPU, downloads the model from Hugging Face and serves it at an https address with keys and OpenAI's chat API, billed by the hour.

Loading…

sarvamai/sarvam-m

Text generation · served with vLLM · ↓ 3.2K downloads a month · View on Hugging Face

sarvamai/sarvam-m needs about 56.1 GB of GPU memory for conversations of up to 8,192 tokens. It runs on one Nvidia A800 from $0.90/h right now; best value: RTX PRO 6000 at $1.15/h.

Weights

47 GB

GPU memory needed

about 56.1 GB

Conversation length

8,192 tokens

Model allows

32,768 tokens

Memory · disk

31.6 GB · 92 GB

Architecture

MistralForCausalLM

Licence

apache-2.0

Versions of sarvam-m ~23.6B

The same model packaged differently: a quantized version needs less GPU memory and runs on a cheaper machine.

Version	Runtime	Weights	GPU memory	Cheapest machine now
Originalsarvamai/sarvam-m	vLLM	47 GB	56.1 GB	$0.89/h Nvidia A800
onnx (Prince-1)Prince-1/sarvam-m-onnx	ONNX Runtime GenAI	47 GB	57.1 GB	$0.89/h Nvidia A800

Hardware and deploy

Endpoint name

Questions

How much GPU memory does sarvam-m need?: About 56.1 GB for conversations of up to 8,192 tokens: 47.14 GB of weights, the cache for one full-length conversation and the runtime's own memory.
What is the cheapest GPU to run sarvam-m?: Right now one Nvidia A800 at $0.90 an hour; RTX PRO 6000 at $1.15 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.
Can sarvam-m run on an RTX 4090?: Not on one: it needs about 56.1 GB and an RTX 4090 has 24 GB. Use a bigger GPU, or a smaller (quantized) version of the model.
How do I deploy sarvam-m as an API?: Pick a machine on its GridPup page and press Deploy: GridPup rents the GPU, downloads the model from Hugging Face and serves it at an https address with keys and OpenAI's chat API, billed by the hour.