FrogNano-4B-2609

Deploy Microsoft FrogNano-4B-2609 on Runpod Serverless with vLLM, fp8 precision, and an 8,192 token max model length.

Overview

FrogNano-4B-2609 is a 4B parameter text model from Microsoft, published as microsoft/frognano-4b-2609. This recipe serves it on Runpod Serverless with vLLM on a single GPU, so you can run long multi-turn, tool-calling agent workloads yourself instead of paying per token.

Recipe

Setting	Value
Engine	vllm
Image	runpod/worker-v1-vllm:v2.27.0
GPU	1x RTX 4090
Precision	fp8
Max model length	8,192

Deploy

Deploy the template from the console: Deploy FrogNano-4B-2609

runpodctl serverless create --hub-id <vllm listing> --model-reference hf://microsoft/FrogNano-4B-2609

Browse related listings in the Hub, or create an account first.

Measured performance

GPU	Price	Throughput	Time to first token	Cold start	Cost per 1M output tokens
1x RTX 4090, fp8	$0.69/hr	104 tokens per second	582 ms	870 s	$1.84

Limits

• License: mit. Review the license terms before you deploy.
• Context length for the model is not stated. This recipe sets a max model length of 8,192, and requests above that limit are rejected.
• The recipe covers text only. It does not configure a network volume, quantization other than fp8, a custom handler, or multi-GPU serving.

See the model page and the deployment walkthrough.