From 86d5da40ac521516bcf35335e3194731320b4c31 Mon Sep 17 00:00:00 2001 From: Daniel Bevenius Date: Mon, 4 Mar 2024 06:40:04 +0100 Subject: [PATCH] add: ggml graph encoding format This commit adds the `ggml` graph encoding format to the `graph_encoding` enum. The motivation for this is to allow the `wasi-nn` interface to support models that are encoded in the `ggml` format which is the model format used by llama.cpp. Signed-off-by: Daniel Bevenius --- ml.md | 1 + wit/wasi-nn.wit | 1 + 2 files changed, 2 insertions(+) diff --git a/ml.md b/ml.md index e625f8f..a45a3b3 100644 --- a/ml.md +++ b/ml.md @@ -214,6 +214,7 @@ backends that encode (i.e., serialize) their graph IR with different formats.

tensorflow
  • pytorch
  • tensorflowlite
  • +
  • ggml
  • autodetect
  • enum execution-target

    diff --git a/wit/wasi-nn.wit b/wit/wasi-nn.wit index 3e54249..8bea66e 100644 --- a/wit/wasi-nn.wit +++ b/wit/wasi-nn.wit @@ -78,6 +78,7 @@ interface graph { tensorflow, pytorch, tensorflowlite, + ggml, autodetect, }