chore(vendor): import pi 1.3.0

Imported as a linear vendor commit so forge rebases do not replay the upstream root tree.

git-subtree-dir: third_party/pi

git-subtree-split: 71023aec1546d6d677a65669c5665c0927ce0dd0

third_party/pi/packages/coding-agent/docs/llama-cpp.md

@@ -1,4 +1,4 @@
-# llama.cpp
+# Local Models with llama.cpp
 
 Pi supports the [llama.cpp](https://github.com/ggml-org/llama.cpp) router server. The router discovers multiple GGUF models and loads or unloads them on demand.
 
@@ -41,7 +41,7 @@ A single-file model can sit directly in the model directory. Put multimodal and
     └── large-model-Q4_K_M-00003-of-00003.gguf
 ```
 
-Restart the router after manually adding files. For per-model context sizes and other options, use [llama.cpp model presets](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md#model-presets).
+Restart the router after manually adding files. For per-model context sizes and other options, use [llama.cpp model presets](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md#model-presets). For settings tuned to a specific NVIDIA GPU and model, see [Run Pi with NVIDIA RTX and llama.cpp](nvidia-rtx-llama-cpp.md).
 
 ## Configure Pi
 
@@ -82,10 +82,28 @@ Hugging Face search uses `HF_TOKEN` when set, then checks `$HF_TOKEN_PATH`, `$HF
 
 If other models are loaded, Pi asks whether to unload them first or keep them loaded. Pi does not silently unload models and never deletes model files. The router may be shared with other clients, so `/llama` always displays the router's current state.
 
-Only loaded models appear in `/model`. After loading a model, run `/model` to select it for the current Pi session.
+Loaded and sleeping models appear in `/model`. Sleeping models wake automatically when selected. With router autoload enabled, unloaded preset models also appear and load when selected. With `--no-models-autoload`, load a model through `/llama` before selecting it.
 
 If the router disconnects, `/llama` shows **Retry** and **Close**. Retry reconnects and refreshes model state without replaying the interrupted operation.
 
+## Classification
+
+Classifier models answer typed `choice`, `bool`, and `score` questions about JSON state, like TypeSafe's Jev models. The model reaches them from [`codemode`](cli.md#enable-codemode) scripts, and extensions through `ctx.modelRegistry.classify()`; see [Classifier models](models.md#use-classifier-models). Pi lists llama.cpp models as classifiers in two ways:
+
+- **Decision models** such as [Julia-1, Laya, Kev, lev, and OpenJev](https://huggingface.co/collections/ggml-org/decision-models-6abf80cca3c83f127060a769) answer natively through llama.cpp's `/v1/systemone` endpoint. They appear only as classifiers, with the `typesafe-system-one` API, and not in `/model`.
+- **Chat models** are also listed as classifiers with the same ID and the `llama-cpp-classify` API, which reads answers from next-token probabilities as described below.
+
+llama.cpp 0.6.0 and later report decision models in the router's model list: their `architecture.output_modalities` contains `decisions`. The router reads this from the GGUF metadata without loading the model, so Pi recognizes unloaded and sleeping decision models too. Older llama.cpp builds do not report it, and Pi lists their decision models as chat models.
+
+### Chat models as classifiers
+
+The model does not generate an answer. Each question becomes one chat prompt: the state, every question of the request, the state again, and then the question with its answers under single-token labels. Labels are letters for a choice (up to 62 options), `Yes`/`No` for a bool, and digits for a score (up to 10 levels). The second copy of the state is read with the questions in view, which improved accuracy on JevBench with small models. Pi reads the probabilities of the labels as the next token and normalizes them. A choice returns every option's probability and a confidence of `(n * peak - 1) / (n - 1)`; a score returns the expected level.
+
+- Raw label probabilities are usually overconfident. The per-request `temperature` option divides the label logits before normalizing; values above 1 soften the distribution. It changes no answer.
+- Questions run one after another. Everything before the final question is the same for all questions of a request, so the server's prompt cache evaluates it once. The state appears twice, so it needs twice its size in context.
+- Small models may follow instructions written inside the state. The prompt tells the model to judge the state as data, but that is not a guarantee.
+- Hybrid models such as Qwen3.5 cannot rewind a partially cached prompt without context checkpoints. If each question reprocesses the whole state, start the router with `--ctx-checkpoints 32 --checkpoint-min-step 0`.
+
 ## Troubleshooting
 
 Check that the router is reachable:
@@ -96,6 +114,8 @@ curl http://127.0.0.1:8080/models
 ```
 
 - **No models in `/llama`:** Check `--models-dir`, the directory layout, and restart the router.
-- **Model missing from `/model`:** Load it with `/llama` first.
+- **Model missing from `/model` with `--no-models-autoload`:** Load it with `/llama` first.
 - **Load fails or uses too much memory:** Lower `-c` or unload another model.
 - **Server is not in router mode:** Start it without `--model`, `-m`, or `-hf`.
+
+To remove the `llama.cpp` provider and `/llama`, disable `llama.cpp` under Built-in in `pi config`, or set `"extensions": ["-builtin:llama.cpp"]` in [settings](settings.md#resources).