You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: website/docs/guide/context/pages.mdx
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -192,7 +192,7 @@ raw.truncate_working_page_tokens(n); // drop the last n tokens within the wor
192
192
193
193
Working pages must be full (every slot written) before they can commit. Calling `commit_working_pages(k)` while the k-th page is still partial returns an error. Most code lets `flush()` decide when to commit; the manual form is for hand-written forward-pass loops.
194
194
195
-
For when to use these operations, see [Inputs: tokens, positions, masks](../forward/inputs).
195
+
For when to use these operations, see [Tokens, positions, and masks](../forward/inputs).
Copy file name to clipboardExpand all lines: website/docs/guide/context/scheduling.mdx
+4Lines changed: 4 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -9,6 +9,10 @@ import TabItem from '@theme/TabItem';
9
9
10
10
# Scheduling and budgets
11
11
12
+
:::warning[Experimental]
13
+
The credit auction and bidding API are a prototype. Interfaces, defaults, and capabilities change frequently.
14
+
:::
15
+
12
16
Pie's engine schedules forward passes across all live processes via a per-step credit auction. Each context bids for compute on every tick, and the engine fills its batch by clearing the highest bids first. This page covers the bidding API. **Most inferlets do not need to touch it.** The default `Generator` auto-bids using a budget-exhausting strategy that spreads the wallet over the horizon. Read on if you want to write a custom bid strategy or understand what the default is doing. Read this after [Pages](./pages).
Copy file name to clipboardExpand all lines: website/docs/guide/forward/adapters.mdx
+4Lines changed: 4 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -9,6 +9,10 @@ import TabItem from '@theme/TabItem';
9
9
10
10
# Adapters
11
11
12
+
:::warning[Experimental]
13
+
Adapter support is a prototype. Interfaces, defaults, and capabilities change frequently.
14
+
:::
15
+
12
16
LoRA adapters are runtime-managed overlays scoped to a model. The full lifecycle (create, open, fork, save, load, destroy) is available in all three SDKs. This page covers using an adapter at inference. Read this after [The forward pass](./overview).
Copy file name to clipboardExpand all lines: website/docs/guide/forward/inputs.mdx
+2-2Lines changed: 2 additions & 2 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,10 +1,10 @@
1
1
---
2
-
title: "Inputs: tokens, positions, masks"
2
+
title: "Tokens, positions, and masks"
3
3
description: Drive the input side of a forward pass with tokens, custom positions, and BRLE attention masks
4
4
sidebar_position: 2
5
5
---
6
6
7
-
# Inputs: tokens, positions, masks
7
+
# Tokens, positions, and masks
8
8
9
9
A forward pass takes a list of input positions, the token IDs at those positions, and an optional attention mask. This page covers the three input modalities. Read this after [The forward pass](./overview).
Copy file name to clipboardExpand all lines: website/docs/guide/io/mcp.mdx
+4Lines changed: 4 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -9,6 +9,10 @@ import TabItem from '@theme/TabItem';
9
9
10
10
# MCP
11
11
12
+
:::warning[Experimental]
13
+
MCP support is a prototype. Interfaces, defaults, and capabilities change frequently.
14
+
:::
15
+
12
16
The [Model Context Protocol](https://modelcontextprotocol.io/) is an open standard for exposing tools, resources, and prompt templates to LLM applications. Pie ships an MCP client. Inferlets discover and call MCP servers without writing transport code. Read this after [Tool-call parser](../decoder/tool-calling).
Copy file name to clipboardExpand all lines: website/docs/reference/bakery.mdx
+4Lines changed: 4 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -5,6 +5,10 @@ description: Reference for the bakery CLI (scaffold, build, publish inferlets)
5
5
6
6
# Bakery
7
7
8
+
:::warning[Experimental]
9
+
Bakery is a prototype. Interfaces, defaults, and capabilities change frequently.
10
+
:::
11
+
8
12
`bakery` is the developer-side CLI for inferlets. It wraps the Cargo / npm build, talks to the Pie registry, and produces packaged WASM artifacts.
9
13
10
14
For the end-to-end workflow narrative, see [Your first inferlet](../guide/first-inferlet) (build) and [Deploy and publish](../guide/deploy/serve) (publish). This page is the command reference.
Copy file name to clipboardExpand all lines: website/docs/reference/drivers/sglang.mdx
+4Lines changed: 4 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -6,6 +6,10 @@ sidebar_position: 5
6
6
7
7
# SGLang
8
8
9
+
:::warning[Experimental]
10
+
The SGLang driver is a prototype. Interfaces, defaults, and capabilities change frequently.
11
+
:::
12
+
9
13
`type = "sglang"`. A subprocess Python driver that delegates the forward pass to SGLang. Pie wraps SGLang's `ServerArgs`; field names match SGLang exactly, so values pass through unchanged.
10
14
11
15
Use this driver to run a model that the embedded [`cuda_native` and `dev`](./cuda) drivers do not yet implement, or to run against SGLang's attention kernels and its n-gram speculative decoding.
Copy file name to clipboardExpand all lines: website/docs/reference/drivers/vllm.mdx
+4Lines changed: 4 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -6,6 +6,10 @@ sidebar_position: 4
6
6
7
7
# vLLM
8
8
9
+
:::warning[Experimental]
10
+
The vLLM driver is a prototype. Interfaces, defaults, and capabilities change frequently.
11
+
:::
12
+
9
13
`type = "vllm"`. A subprocess Python driver that delegates the forward pass to vLLM. Pie wraps vLLM's `EngineArgs`; field names match vLLM exactly, so values pass through unchanged.
10
14
11
15
Use this driver to run a model that the embedded [`cuda_native` and `dev`](./cuda) drivers do not yet implement, or to run against vLLM's attention kernels.
0 commit comments