prism-inference 0.1.23

Bounded Rust CPU inference and tokenization for Prism-compatible GGUF models
# Documentation


[Project overview](../README.md) ยท [Download releases](https://github.com/symbiogenesis/prism-inference/releases)

Prism Inference supplies CPU inference tools and a Rust library. Its initial
generation path targets Swift/Bonsai; broader Prism models are planned.

## Start here


| Task | Guide |
| --- | --- |
| Run a downloaded executable | [CLI and workers]cli-and-workers.md |
| Choose text, token IDs or tokenization-only mode | [Driver input]cli-and-workers.md#generate-from-text-or-token-ids |
| Keep a tokenizer process alive | [Persistent token worker]cli-and-workers.md#persistent-token-worker |
| Integrate a worker or the private HTTP server | [Swift worker and private server]cli-and-workers.md#swift-worker-and-private-http-server |
| Add the Rust library to an application | [Library dependency and tokenizer example]build-and-validation.md#use-the-rust-library |
| Understand budgets and failure recovery | [Runtime contracts]runtime-contracts.md |
| Configure opt-in packed projection workers | [Explicit row workers]parallel-rows.md |
| Review prepared CPU work/storage reductions | [Inference efficiency candidate]inference-efficiency.md |
| Review scoped packed activation preparation | [PQ2 consumption layout]packed-pq2-consumption.md |
| Select separately qualified CPU prompt tiles or GDN chunks | [CPU tiled prefill]cpu-tiled-prefill.md |
| Qualify exact latest-state service reuse | [Latest-prefix service]latest-prefix-service.md |
| Compile, check or package the workspace | [Building and validation]build-and-validation.md |
| Run upstream source/backend and codec checks | [Runtime conformance]prism-upstream-tests.md |
| Capture explicit rank-one adapter correctness evidence | [Adapter correctness path]rank-one-adapter.md |
| Check support and remaining qualification | [Roadmap]roadmap.md |
| Compare the prepared CPU release with its predecessor | [0.1.22 release scope]release-0.1.22.md |

## Technical reference


The [runtime contracts](runtime-contracts.md) retain the implemented API behavior,
model verification, transactional checkpoints, memory accounting and SIMD
diagnostics. The [CLI guide](cli-and-workers.md) covers exact file/stream input,
generation controls, JSONL framing, client failure policy and tokenizer APIs.

The [build guide](build-and-validation.md) describes Cargo features, the pinned
compiler, local focused checks, hosted correctness gates and release receipts.
The [roadmap](roadmap.md) links the inherited research and qualification evidence.

The [GGUF tool](../tools/gguf/README.md), [ternary codecs](../tools/ternary/README.md)
and [Prism storage overlay](../tools/prism/README.md) retain their own reference
documentation. See [NOTICE](../NOTICE) for attribution and source revisions.