summaryrefslogtreecommitdiff
path: root/candle-transformers/src/models/moondream.rs
diff options
context:
space:
mode:
Diffstat (limited to 'candle-transformers/src/models/moondream.rs')
-rw-r--r--candle-transformers/src/models/moondream.rs11
1 files changed, 11 insertions, 0 deletions
diff --git a/candle-transformers/src/models/moondream.rs b/candle-transformers/src/models/moondream.rs
index cde59d43..d351d7c0 100644
--- a/candle-transformers/src/models/moondream.rs
+++ b/candle-transformers/src/models/moondream.rs
@@ -1,3 +1,14 @@
+//! MoonDream Model vision-to-text
+//!
+//! The model consists of:
+//! - Vision encoder using a ViT-style architecture
+//! - Text decoder based on Microsoft's Phi model
+//! - Vision projection module to align vision and text embeddings
+//!
+//! References:
+//! - [MoonDream Original Implementation](https://github.com/vikhyat/moondream)
+//!
+
use crate::models::mixformer::{Config as PhiConfig, MixFormerSequentialForCausalLM as PhiModel};
use crate::models::with_tracing::{layer_norm, linear_b, LayerNorm, Linear};
use candle::{IndexOp, Module, Result, Tensor, D};