feat(minor): customize `postinstall` behavior (#582)

* feat: customize `postinstall` behavior
* feat: experimental support for context KV cache type configurations
* feat: support `NVFP4` quants

feat: automatic checkpoints for models that need it (#573)

* feat: automatic checkpoints for models that need it (such as Qwen 3.5 due to its hybrid architecture)
* feat(`QwenChatWrapper`): Qwen 3.5 support
* feat(`inspect gpu` command): detect and report missing prebuilt binary modules and custom npm registry
* feat: initial disk cache dir option for future optimizations (disabled for now)
* fix: Qwen 3.5 memory estimation
* fix: grammar use with HarmonyChatWrapper
* fix: add mistral think segment detection
* fix: compress excessively long segments from the current response on context shift instead of throwing an error
* fix: default thinking budget to 75% of the context size to prevent low-quality responses
* fix: bugs

fix: Electron template (#566)

feat(`getLlama`): `build: "autoAttempt"` (#564)

* feat(`getLlama`): `build: "autoAttempt"`
* feat: get rid of octokit
* fix(CLI): disable Direct I/O by default
* fix: Bun segmentation fault on process exit with undisposed `Llama`
* fix: detect glibc inside Nix
* fix: stricter CI build flag
* chore: update `simple-git`
* chore: switch off of `tsconfig.json` deprecated configs
* docs: clarify `getLlama`'s `build` option logic

fix: macOS 14 prebuilt binaries (#559)

fix: export missing types (#557)

feat: Exclude Top Choices (XTC) (#553)

* feat: Exclude Top Choices (XTC) support
* feat: DRY (Don't Repeat Yourself) repeat penalty support
* feat: Tiny Aya support
* fix: adjust the default VRAM padding config to reserve enough memory for compute buffers
* fix: adapt to breaking `llama.cpp` changes
* fix: support function call syntax with optional whitespace prefix
* fix: find the provided cmake path
* fix: change the default value of `useDirectIo` to `false`
* fix: Vulkan device dedupe

fix: adapt to `llama.cpp` changes (#547)

* fix: adapt to `llama.cpp` changes
* fix: change the level of common logs

feat(`LlamaCompletion`): `stopOnAbortSignal` (#538)

* feat(`LlamaCompletion`): `stopOnAbortSignal`
* feat(`LlamaModel`): `useDirectIo`
* fix: support new CUDA 13.1 archs
* fix: build the prebuilt binaries with CUDA 13.1 instead of 13.0
* docs: stopping a text completion generation

docs: fix cmake dependencies link (#534)