feat:support for two long-context embedding models (Qwen3-Embedding-0.6B and EmbeddingGemma-300M) (vllm-project#453)

feat:support for two long-context embedding models (Qwen3-Embedding-0.6B and EmbeddingGemma-300M)

Signed-off-by: OneZero-Y <[email protected]>
Signed-off-by: Huamin Chen <[email protected]>

Author: OneZero-Y

candle-binding/Cargo.toml

@@ -10,14 +10,21 @@ name = "candle_semantic_router"
 crate-type = ["staticlib", "cdylib"]
 
 [features]
-default = ["cuda"]
+default = []
 cuda = ["candle-core/cuda", "candle-nn/cuda", "candle-transformers/cuda"]
+# Flash Attention 2 support (requires CUDA and compatible GPU)
+# Enable with: cargo build --features flash-attn
+# Note: Requires CUDA Compute Capability >= 8.0 (Ampere or newer)
+flash-attn = ["candle-flash-attn"]
 
 [dependencies]
 anyhow = { version = "1", features = ["backtrace"] }
 candle-core = "0.8.4"
 candle-nn = "0.8.4"
 candle-transformers = "0.8.4"
+# Flash Attention 2 (optional, requires CUDA)
+# Reference: https://github.com/huggingface/candle/tree/main/candle-flash-attn
+candle-flash-attn = { version = "0.8.4", optional = true }
 tokenizers = { version = "0.21.0", features = ["http"] }
 hf-hub = "0.4.1"
 safetensors = "0.4.1"