Granite Switch 4.1 vs SPEV

Granite Switch 4.1

One self-hosted Apache 2.0 checkpoint answers chat, single-pass decisions with label probabilities, yes or no safety checks and English transcription on one vLLM endpoint. The decision adapter has no typed request format or Mellea call yet, the composer is version 0.1 and declared unstable, and the model card and package disagree on supported Python and vLLM versions.

SPEV

SPEV answers typed questions over text, audio or an image in one call, with a public OpenAPI schema and Apache-2.0 weights on Hugging Face. It is a day old, has an 8,192-token context, no status page or changelog, and Sprag's terms let it train on API content unless the customer opts out by email.

Granite Switch 4.1

1. Pass chat_template_kwargs: {"adapter_name": "decision"} or "guardian-core". Without it the request is ordinary chat on the base model
2. For a decision, send the state and the ### DECISION question as two user turns, request logprob_token_ids for the option labels, and set max_completion_tokens to 1
3. Use single-token option labels. Split more than 128 labels into chunks with the identical prompt and apply one softmax over all returned logprobs
4. Parse guardian-core answers as JSON, {"score": "yes"} or {"score": "no"}, where yes means the criterion is met
5. Add authentication before exposing the vLLM server. Neither vLLM nor the checkpoint requires a key by default

SPEV

1. Send model: "spev" to POST https://api.sprag.ai/v1/decisions with a bearer key. /v1/systemone returns 410
2. Ask every question about one piece of evidence in one request. Answers come back in question order
3. Send audio as base64 in input_audio with format, not as a data URL, and at most one image across all messages
4. Keep evidence and questions under 8,192 tokens. Over the limit returns 400 model_limit_exceeded
5. Back off on 429 and 503 for Retry-After seconds. Limits are 60 requests a minute per organisation by default
6. Handle an answer of type refusal, which carries no probabilities