gpu●aiDocs
CH·02CLI reference

gpu embeddings

Create text embeddings (vectors) for one or more inputs

Create text embeddings (vectors) for one or more inputs

Synopsis

Creates embeddings for one or more text inputs using a serverless embedding model, over the OpenAI-compatible POST /v1/embeddings endpoint. The call is always synchronous — embeddings have no streaming form. Requires a key with the serverless:write scope.

Each trailing argument is one input. With no trailing arguments, inputs are read from piped stdin, one input per line (blank lines are skipped) — so cat corpus.txt | gpu embeddings returns one vector per line rather than a single vector for the whole file.

The default table output prints one line per input with the vector's width and a three-float preview; a full vector is thousands of floats and is not useful on a terminal. Use -o json to get the complete response, including every vector, for piping into a file or another program.

Billing is on input tokens only, at the model's listed input rate, rounded up to the next whole cent per request — so every call is billed at least 1¢; batch inputs into one call to pay the listed per-token rate. An embeddings call emits no completion tokens, so the response's usage block carries prompt_tokens and total_tokens only.

gpu embeddings <model> [text...] [flags]

Examples

# Embed a single string
gpu embeddings gpuai/qwen3-embedding-8b "hello world"

# Embed several inputs in one call (one vector per argument)
gpu embeddings gpuai/qwen3-embedding-8b "first sentence" "second sentence"

# Embed a file, one input per line
cat corpus.txt | gpu embeddings gpuai/qwen3-embedding-8b

# Full vectors as JSON, saved to a file
gpu embeddings gpuai/qwen3-embedding-8b "hello" -o json > vectors.json

# Request a narrower vector (only models supporting truncated embeddings)
gpu embeddings gpuai/qwen3-embedding-8b "hello" --dimensions 1024

# List the embedding models you can call
gpu models --modality embedding

Table output looks like:

[0] dims=4096 first3=[-0.019, 0.0071, -0.033]
[1] dims=4096 first3=[0.0042, -0.026, 0.011]
2 inputs, 9 prompt tokens

Options

      --dimensions int   Requested vector width; only models supporting truncated embeddings honour it
  -h, --help             help for embeddings

Options inherited from parent commands

      --api-base string   API base URL (override with GPUAI_API_BASE env) (default "https://api.gpu.ai/v1")
      --debug             Enable debug logging to stderr
  -o, --output string     Output format: table|json (default table on TTY, json otherwise)

Limits

  • Up to 2048 inputs per request, bounded so the response stays under
64 MiB (about 680 inputs at 4096 dimensions); over-limit requests are rejected with 400 before any processing.
  • encoding_format supports float only; base64 is rejected with a
400 unsupported_parameter.
  • A chat model id on this endpoint returns 404 model_not_found — the id is
valid, but not on this surface.

SEE ALSO

  • gpu - GPU.ai command-line interface
  • gpu models - List and inspect serverless inference models

← The gpu CLI