# Qwen2.5-0.5B on AMD

A 0.5B dense chat model on one AMD Instinct MI325X on a bring-your-own cluster.

Source: https://docs.modelplane.ai/recipes/qwen2.5-0.5b-amd/

<!-- vale write-good.Passive = NO -->
A 0.5B dense chat model on one AMD Instinct MI325X on a bring-your-own
cluster: one `Standalone` engine, no cache, weights pulled straight from
Hugging Face. The model is deliberately small - the point of this run is the
AMD serving path, not the model. The class claims the GPU through DRA
(`gpu.amd.com`); the serving stack installs the AMD GPU operator in DRA mode
when a class claims that driver, against the amdgpu kernel driver the node
image already carries. The engine is AMD's ROCm vLLM.

`source: Existing` is the one cloud source that installs the AMD stack today.
Bring the cluster with the amdgpu driver on its GPU nodes, label the pool's
nodes `modelplane.ai/pool=gpu-mi325x`, and apply the platform side first, then
the ML side.

## Validated deployments

{{< validated-deployments >}}

## Platform

{{< manifests "recipes/qwen2.5-0.5b-amd/inference-class.yaml" >}}

{{< manifests "recipes/qwen2.5-0.5b-amd/inference-cluster.yaml" >}}

## Deployment

{{< manifests "recipes/qwen2.5-0.5b-amd/model-deployment.yaml" >}}

{{< manifests "recipes/qwen2.5-0.5b-amd/model-service.yaml" >}}
<!-- vale write-good.Passive = YES -->