Text generation · served with vLLM · ↓ 831 downloads a month · View on Hugging Face
mattshumer/Reflection-Llama-3.1-70B needs about 162 GB of GPU memory for conversations of up to 8,192 tokens. It runs on 8x RTX 3090 from $2.30/h right now; best value: 2x RTX PRO 6000D at $3.29/h.
Weights
141 GB
GPU memory needed
about 162 GB
Conversation length
8,192 tokens
Model allows
8,192 tokens
Memory · disk
78.6 GB · 196 GB
Architecture
LlamaForCausalLM
Licence
llama3.1
Hardware and deploy
$3.29/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 Reflection-Llama-3.1-70B need?
About 162 GB for conversations of up to 8,192 tokens: 141.11 GB of weights, the cache for one full-length conversation and the runtime's own memory.
What is the cheapest GPU to run Reflection-Llama-3.1-70B?
Right now 8x RTX 3090 at $2.30 an hour; 2x RTX PRO 6000D at $3.29 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.
Can Reflection-Llama-3.1-70B run on an RTX 4090?
Not on one: it needs about 162 GB and an RTX 4090 has 24 GB. Use a bigger GPU, or a smaller (quantized) version of the model.
How do I deploy Reflection-Llama-3.1-70B 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.
Prices are per hour for the whole machine and its disk, right now. Speed is Vast's DLPerf score of the machine, relative to the cheapest. Best value: the most speed per dollar. Only machines with a Vast reliability above 95% that download at 400 Mbit/s or more are offered: the machine fetches the model itself.