Granite Switch 4.1 vs pplx-decider

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.

pplx-decider

The hosted Decisions API now serves v1.2, with an OpenAPI description, typed questions, Python and TypeScript SDK methods and $0.02 per million input tokens, and the weights are public under Apache-2.0. Requests naming v1.1 or v1 are now answered by v1.2 with no notice period, and the response does not say which version answered.

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

pplx-decider

1. Send Authorization: Bearer. A key in x-api-key returns 401 on this endpoint.
2. Pin pplx-decider-v1.2 and log it with each result. Old names and aliases route to newer versions, and the response echoes the name sent.
3. Revalidate thresholds set on v1.1. The v1.2 card says it gives different answers to v1.1-style requests.
4. Group questions that are facets of one decision. Send independent candidates, such as documents or tools, in separate requests.
5. Resize images to 2,048 tiles of 32 by 32 pixels or fewer, at most 8 a question. Larger images end in a 504 after about a minute.
6. For the weights, use the repository's pplx_decider package. The card says default causal inference does not reproduce the checkpoint.