Add flash attn v3

Author: EricLBuehler

Cargo.toml

@@ -13,6 +13,7 @@ members = [
 exclude = [
    "candle-book",
    "candle-flash-attn",
+   "candle-flash-attn-v3",
    "candle-kernels",
    "candle-metal-kernels",
    "candle-onnx",
@@ -36,6 +37,7 @@ byteorder = "1.4.3"
 candle = { path = "./candle-core", package = "candle-core", version = "0.9.1" }
 candle-datasets = { path = "./candle-datasets", version = "0.9.1" }
 candle-flash-attn = { path = "./candle-flash-attn", version = "0.9.1" }
+candle-flash-attn-v3 = { path = "./candle-flash-attn-v3", version = "0.9.1" }
 candle-kernels = { path = "./candle-kernels", version = "0.9.1" }
 candle-metal-kernels = { path = "./candle-metal-kernels", version = "0.9.1" }
 candle-nn = { path = "./candle-nn", version = "0.9.1" }

candle-flash-attn-v3/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "candle-flash-attn-v3"
-version = "0.8.0"
+version = "0.9.1"
 edition = "2021"
 
 description = "Flash attention V3 layer for the candle ML framework."