Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Replace load with load-by-name #74

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Replace load with load-by-name #74

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0f721a8
Replace `load` with `load-by-name`
abrown Jul 20, 2024
2fc6261
Fixup `ml.md`
abrown Jul 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
64 changes: 9 additions & 55 deletions ml.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,10 +102,6 @@ containing a single value, use <code>[1]</code> for the tensor dimensions.</p>
<p>Caller module passed an invalid argument.
</li>
<li>
<p><a name="error_code.invalid_encoding"></a><code>invalid-encoding</code></p>
<p>Invalid encoding.
</li>
<li>
<p><a name="error_code.timeout"></a><code>timeout</code></p>
<p>The operation timed out.
</li>
Expand Down Expand Up @@ -231,73 +227,31 @@ framework (e.g., TensorFlow):</p>
<h4><a name="error"></a><code>type error</code></h4>
<p><a href="#error"><a href="#error"><code>error</code></a></a></p>
<p>
#### <a name="tensor"></a>`type tensor`
[`tensor`](#tensor)
<p>
#### <a name="graph_execution_context"></a>`type graph-execution-context`
[`graph-execution-context`](#graph_execution_context)
<p>
#### <a name="graph"></a>`resource graph`
<p>An execution graph for performing inference (i.e., a model).</p>
<h4><a name="graph_encoding"></a><code>enum graph-encoding</code></h4>
<p>Describes the encoding of the graph. This allows the API to be implemented by various
backends that encode (i.e., serialize) their graph IR with different formats.</p>
<h5>Enum Cases</h5>
<ul>
<li><a name="graph_encoding.openvino"></a><code>openvino</code></li>
<li><a name="graph_encoding.onnx"></a><code>onnx</code></li>
<li><a name="graph_encoding.tensorflow"></a><code>tensorflow</code></li>
<li><a name="graph_encoding.pytorch"></a><code>pytorch</code></li>
<li><a name="graph_encoding.tensorflowlite"></a><code>tensorflowlite</code></li>
<li><a name="graph_encoding.ggml"></a><code>ggml</code></li>
<li><a name="graph_encoding.autodetect"></a><code>autodetect</code></li>
</ul>
<h4><a name="execution_target"></a><code>enum execution-target</code></h4>
<p>Define where the graph should be executed.</p>
<h5>Enum Cases</h5>
<ul>
<li><a name="execution_target.cpu"></a><code>cpu</code></li>
<li><a name="execution_target.gpu"></a><code>gpu</code></li>
<li><a name="execution_target.tpu"></a><code>tpu</code></li>
</ul>
<h4><a name="graph_builder"></a><code>type graph-builder</code></h4>
<p><a href="#graph_builder"><a href="#graph_builder"><code>graph-builder</code></a></a></p>
<p>The graph initialization data.
<p>This gets bundled up into an array of buffers because implementing backends may encode their
graph IR in parts (e.g., OpenVINO stores its IR and weights separately).</p>
<hr />
<h2>An execution graph for performing inference (i.e., a model).</h2>
<h3>Functions</h3>
<h4><a name="method_graph_init_execution_context"></a><code>[method]graph.init-execution-context: func</code></h4>
<h5>Params</h5>
<ul>
<li><a name="method_graph_init_execution_context.self"></a><code>self</code>: borrow&lt;<a href="#graph"><a href="#graph"><code>graph</code></a></a>&gt;</li>
</ul>
<h5>Return values</h5>
<ul>
<li><a name="method_graph_init_execution_context.0"></a> result&lt;own&lt;<a href="#graph_execution_context"><a href="#graph_execution_context"><code>graph-execution-context</code></a></a>&gt;, own&lt;<a href="#error"><a href="#error"><code>error</code></a></a>&gt;&gt;</li>
</ul>
<h4><a name="load"></a><code>load: func</code></h4>
<p>Load a <a href="#graph"><code>graph</code></a> from an opaque sequence of bytes to use for inference.</p>
<p>Load a <a href="#graph"><code>graph</code></a> by name.</p>
<p>How the host expects the names to be passed and how it stores the graphs for retrieval via
this function is <strong>implementation-specific</strong>. This allows hosts to choose name schemes that
range from simple to complex (e.g., URLs?) and caching mechanisms of various kinds.</p>
<h5>Params</h5>
<ul>
<li><a name="load.builder"></a><code>builder</code>: list&lt;<a href="#graph_builder"><a href="#graph_builder"><code>graph-builder</code></a></a>&gt;</li>
<li><a name="load.encoding"></a><code>encoding</code>: <a href="#graph_encoding"><a href="#graph_encoding"><code>graph-encoding</code></a></a></li>
<li><a name="load.target"></a><code>target</code>: <a href="#execution_target"><a href="#execution_target"><code>execution-target</code></a></a></li>
<li><a name="load.name"></a><code>name</code>: <code>string</code></li>
</ul>
<h5>Return values</h5>
<ul>
<li><a name="load.0"></a> result&lt;own&lt;<a href="#graph"><a href="#graph"><code>graph</code></a></a>&gt;, own&lt;<a href="#error"><a href="#error"><code>error</code></a></a>&gt;&gt;</li>
</ul>
<h4><a name="load_by_name"></a><code>load-by-name: func</code></h4>
<p>Load a <a href="#graph"><code>graph</code></a> by name.</p>
<p>How the host expects the names to be passed and how it stores the graphs for retrieval via
this function is <strong>implementation-specific</strong>. This allows hosts to choose name schemes that
range from simple to complex (e.g., URLs?) and caching mechanisms of various kinds.</p>
<h4><a name="method_graph_init_execution_context"></a><code>[method]graph.init-execution-context: func</code></h4>
<h5>Params</h5>
<ul>
<li><a name="load_by_name.name"></a><code>name</code>: <code>string</code></li>
<li><a name="method_graph_init_execution_context.self"></a><code>self</code>: borrow&lt;<a href="#graph"><a href="#graph"><code>graph</code></a></a>&gt;</li>
</ul>
<h5>Return values</h5>
<ul>
<li><a name="load_by_name.0"></a> result&lt;own&lt;<a href="#graph"><a href="#graph"><code>graph</code></a></a>&gt;, own&lt;<a href="#error"><a href="#error"><code>error</code></a></a>&gt;&gt;</li>
<li><a name="method_graph_init_execution_context.0"></a> result&lt;own&lt;<a href="#graph_execution_context"><a href="#graph_execution_context"><code>graph-execution-context</code></a></a>&gt;, own&lt;<a href="#error"><a href="#error"><code>error</code></a></a>&gt;&gt;</li>
</ul>
43 changes: 6 additions & 37 deletions wit/wasi-nn.wit
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,48 +62,19 @@ interface tensor {
/// framework (e.g., TensorFlow):
interface graph {
use errors.{error};
use tensor.{tensor};
use inference.{graph-execution-context};

/// An execution graph for performing inference (i.e., a model).
resource graph {
init-execution-context: func() -> result<graph-execution-context, error>;
}

/// Describes the encoding of the graph. This allows the API to be implemented by various
/// backends that encode (i.e., serialize) their graph IR with different formats.
enum graph-encoding {
openvino,
onnx,
tensorflow,
pytorch,
tensorflowlite,
ggml,
autodetect,
}

/// Define where the graph should be executed.
enum execution-target {
cpu,
gpu,
tpu
}

/// The graph initialization data.
///
/// This gets bundled up into an array of buffers because implementing backends may encode their
/// graph IR in parts (e.g., OpenVINO stores its IR and weights separately).
type graph-builder = list<u8>;

/// Load a `graph` from an opaque sequence of bytes to use for inference.
load: func(builder: list<graph-builder>, encoding: graph-encoding, target: execution-target) -> result<graph, error>;

/// Load a `graph` by name.
///
/// How the host expects the names to be passed and how it stores the graphs for retrieval via
/// this function is **implementation-specific**. This allows hosts to choose name schemes that
/// range from simple to complex (e.g., URLs?) and caching mechanisms of various kinds.
load-by-name: func(name: string) -> result<graph, error>;
load: func(name: string) -> result<graph, error>;

/// An execution graph for performing inference (i.e., a model).
resource graph {
init-execution-context: func() -> result<graph-execution-context, error>;
}
}

/// An inference "session" is encapsulated by a `graph-execution-context`. This structure binds a
Expand Down Expand Up @@ -137,8 +108,6 @@ interface errors {
enum error-code {
// Caller module passed an invalid argument.
invalid-argument,
// Invalid encoding.
invalid-encoding,
// The operation timed out.
timeout,
// Runtime Error.
Expand Down