A WireMock simulation of Hugging Face's REST API — covering dense, sparse and unpooled embeddings, reranking, sequence classification, tokenization, and the OpenAI-compatible embeddings route.
A shared instance of this template runs on a public endpoint — call it right now from curl, an SDK, or an AI agent. Responses are fixed examples and behavior can't be changed; rate limits apply.
Jump to the endpoints ↓Run this template as a mock API in WireMock Cloud for a private instance you control — edit stubs, add stateful scenarios, use dynamic response templating, and inspect requests in the log.
Open in WireMock Cloud (free) →Live sandbox · no signup, no API key
A running WireMock Cloud instance of this template, callable right now from a terminal, a script, or an AI agent.
Endpoints
POST Decode input ids [decode]
curl -s -X POST 'https://oz63kz.wiremockapi.cloud/decode' \
-H 'Content-Type: application/json' \
-d '{}'GET Health check method [health]
curl -s -X GET 'https://oz63kz.wiremockapi.cloud/health'POST Get Embeddings. Returns a 424 status code if the model is not an embedding model. [embed]
curl -s -X POST 'https://oz63kz.wiremockapi.cloud/embed' \
-H 'Content-Type: application/json' \
-d '{}'Spec catalog — Text Embeddings Inference (1.9.4) · 12 endpoints
| Method | Path | Summary |
|---|---|---|
| POST | /decode | Decode input ids |
| POST | /embed | Get Embeddings. Returns a 424 status code if the model is not an embedding model. |
| POST | /embed_all | Get all Embeddings without Pooling. |
| POST | /embed_sparse | Get Sparse Embeddings. Returns a 424 status code if the model is not an embedding model with SPLADE pooling. |
| GET | /health | Health check method |
| GET | /info | Text Embeddings Inference endpoint info |
| GET | /metrics | Prometheus metrics scrape endpoint |
| POST | /predict | Get Predictions. Returns a 424 status code if the model is not a Sequence Classification model |
| POST | /rerank | Get Ranks. Returns a 424 status code if the model is not a Sequence Classification model with |
| POST | /similarity | Get Sentence Similarity. Returns a 424 status code if the model is not an embedding model. |
| POST | /tokenize | Tokenize inputs |
| POST | /v1/embeddings | OpenAI compatible route. Returns a 424 status code if the model is not an embedding model. |
Unauthenticated WireMock Cloud demo sandbox for Hugging Face — not an official Hugging Face sandbox, and it returns simulated example data only.
The template stands in for a self-hosted Text Embeddings Inference server, following the TEI 1.9.4 spec. There's nothing to create or store, so every response is computed from the request instead: embed returns one vector per input, rerank returns index and score pairs for the texts you send, and the OpenAI-style route wraps vectors in a list object with the model name. A retrieval pipeline wired to TEI can run its full indexing path without a GPU behind it.
TEI serves an OpenAI-compatible embeddings route, so the OpenAI client is the shortest way to call it; it only needs the base URL and a placeholder key:
Vector-store integrations such as LangChain's OpenAI embeddings class accept the same base URL, so a RAG pipeline can switch between a local TEI container, this mock and a production server through configuration.
That makes ingestion jobs and retrieval APIs testable in CI on machines that could never load the model, with no container start-up time.
from openai import OpenAI
# TEI serves an OpenAI-compatible /v1/embeddings route;
# it's unauthenticated, so the api_key is only a placeholder
client = OpenAI(
base_url="https://oz63kz.wiremockapi.cloud/v1",
api_key="unused",
)
result = client.embeddings.create(model="tei", input="What is deep learning?")
print(len(result.data[0].embedding))TEI is self-hosted and unauthenticated by default, and so is the mock. If your deployment puts TEI behind a gateway that checks tokens, model that gateway separately.
An empty batch answers 400 with error_type Empty, and more than 32 inputs answers 413 with error_type Validation. On the OpenAI-style route both use the OpenAI error shape, so client-side batching logic can be checked against TEI's limits.
The tokenizer 422, backend 424, overload 429 and unhealthy 503 never appear, because no input reliably triggers them. Add those as stubs in your own WireMock Cloud copy to test back-off behavior.
Vendor names identify APIs represented by WireMock template sources. This page does not imply vendor endorsement, certification, partnership, or official integration status.