# Prepare text for a retrieval workflow | EasyNet

Source: https://easynet.run/cases/warm-gpu-model

# Prepare text for a retrieval workflow

A deterministic embedding stub checks input bounds, provider selection, and a fixed-size result without loading a model or using a GPU.
[All source cases](https://easynet.run/cases)[Inspect the pinned project](https://github.com/EasyRemote/EasyRemote/tree/44c768ca26d0f92a5ab9d1d03342895c29c27dc6/gallery/projects/01_team_gpu_pool_load_balancing)

## 30-second boundary walkthrough

A conceptual animation of the checked-in fixture: bounded text enters, the provider-local deterministic stub runs, and an eight-number result returns.

01A caller submits bounded text to embed_text.

02The fixture selects the named local provider.

03A deterministic BLAKE2b stub executes, no model or GPU is used.

04Only the eight-number result returns to the caller.

## The problem

A team needs to validate the callable boundary and provider selection before connecting a real resident model.

## What gets published

Expose embed_text with a non-empty input capped at 4,096 characters.

## What stays local

The deterministic implementation, process state, and machine access.

## Business data crossing the call boundary

One bounded text input and a fixed-size JSON vector.

## What this shows

After the interface passes, a real model can replace the stub while callers keep the same input and result shape.

## What it doesn't prove

The implementation is a deterministic BLAKE2b stub. It does not load a model, use a GPU, or prove scheduling, quotas, latency, or deployment.

## Public source evidence for this case

Evidence level

Source layout and declarations are anonymously inspectable

Pinned public EasyRemote revision

44c768ca26d0f92a5ab9d1d03342895c29c27dc6

Public inspection

[Public inspection](https://easynet.run/docs/source-check#run-check)

Expected check

Required files, Python syntax, and @node.register declaration are checked without installing dependencies

Runtime invocation

Not established by this source fixture; record separate discovery, Invocation, and terminal Receipt evidence

## Run in a compatible source workspace

This Provider/Client walkthrough requires compatible EasyRemote, EasyNet-Cli, and EasyNet-Axon sibling source checkouts, a compatible Hub, and a paired, running Runtime. The source is public; clean installation from public packages is not yet verified. Technical Alpha offers guided setup. Use the pinned EasyRemote revision below; this page does not pin the other components. Runtime admission still determines whether the caller may execute the function.

```
# From EasyRemote revision 44c768ca26d0f92a5ab9d1d03342895c29c27dc6
cd gallery/projects/01_team_gpu_pool_load_balancing
# Terminal 1
uv sync --frozen
uv run python node.py

# Terminal 2 (same project)
EASYREMOTE_GPU_NODE=gpu-1 uv run python client.py
```

[Connect your own function](https://easynet.run/docs/connect/function)

Read the setup first. You need compatible sibling source checkouts, a compatible Hub, and a paired, running Runtime. Technical Alpha offers guided help; public-package clean installation is not yet verified.
[All source cases](https://easynet.run/cases)[Request technical alpha](https://easynet.run/pilot)