aboutsummaryrefslogtreecommitdiff
path: root/embed_mlx.h
diff options
context:
space:
mode:
authorOnur Gündüz <ogunduz@gmail.com>2026-06-14 15:19:12 +0200
committerOnur Gündüz <ogunduz@gmail.com>2026-06-14 15:19:12 +0200
commitf463018cd700690336898d109226d489deca036d (patch)
tree4e1d83f55a5e231af519aa4d647141db5de23af7 /embed_mlx.h
parentd98ee1568c0bd88304a12e59d12104066e01e8d0 (diff)
downloadffwd-f463018cd700690336898d109226d489deca036d.tar.gz
ffwd-f463018cd700690336898d109226d489deca036d.zip
all: add support for Qwen3-Embedding + causal GQA attention
Diffstat (limited to 'embed_mlx.h')
-rw-r--r--embed_mlx.h2
1 files changed, 1 insertions, 1 deletions
diff --git a/embed_mlx.h b/embed_mlx.h
index 267c454..5f6e2c9 100644
--- a/embed_mlx.h
+++ b/embed_mlx.h
@@ -108,7 +108,7 @@ EMBED_API int embed_mlx_encode_into(embed_mlx_ctx_t *ctx, const int *token_ids,
* Compute a true padded dense batch on MLX.
*
* out_embeddings is caller-provided [batch, hidden_size].
- * Padded tokens are masked out of attention keys and mean pooling.
+ * Pooling and attention masking follow the loaded model configuration.
* Returns 0 on success, -1 on error.
*/
EMBED_API int embed_mlx_encode_batch(embed_mlx_ctx_t *ctx, const embed_input_t *inputs,