Add cublaslt matmul

Cargo.toml

@@ -25,8 +25,8 @@ license = "MIT"
 
 [workspace.dependencies]
 anyhow = "1.0.80"
-candle-core = { git = "https://github.com/EricLBuehler/candle.git", version = "0.7.0", rev = "20a57c4" }
-candle-nn = { git = "https://github.com/EricLBuehler/candle.git", version = "0.7.0", rev = "20a57c4" }
+candle-core = { git = "https://github.com/EricLBuehler/candle.git", version = "0.7.0", rev = "4754afe" }
+candle-nn = { git = "https://github.com/EricLBuehler/candle.git", version = "0.7.0", rev = "4754afe" }
 serde = "1.0.197"
 serde_json = "1.0.114"
 indexmap = { version = "2.2.5", features = ["serde"] }
@@ -49,3 +49,4 @@ rayon = "1.1.0"
 url = "2.5.2"
 data-url = "0.3.1"
 buildstructor = "0.5.4"
+float8 = "0.1.1"

mistralrs-core/Cargo.toml

@@ -17,7 +17,7 @@ candle-core.workspace = true
 candle-nn.workspace = true
 serde.workspace = true
 serde_json.workspace = true
-candle-flash-attn = { git = "https://github.com/EricLBuehler/candle.git", version = "0.7.0", rev = "20a57c4", optional = true }
+candle-flash-attn = { git = "https://github.com/EricLBuehler/candle.git", version = "0.7.0", rev = "4754afe", optional = true }
 dirs = "5.0.1"
 hf-hub = "0.3.2"
 thiserror = "1.0.57"
@@ -78,10 +78,11 @@ regex = "1.10.6"
 safetensors = "0.4.5"
 serde_plain = "1.0.2"
 as-any = "0.3.1"
+float8.workspace = true
 
 [features]
 pyo3_macros = ["pyo3"]
-cuda = ["candle-core/cuda", "candle-nn/cuda", "dep:bindgen_cuda", "mistralrs-quant/cuda", "dep:mistralrs-paged-attn", "mistralrs-paged-attn/cuda"]
+cuda = ["candle-core/cuda", "candle-nn/cuda", "dep:bindgen_cuda", "mistralrs-quant/cuda", "dep:mistralrs-paged-attn", "mistralrs-paged-attn/cuda", "float8/cuda"]
 cudnn = ["candle-core/cudnn"]
 metal = ["candle-core/metal", "candle-nn/metal"]
 flash-attn = ["cuda", "dep:candle-flash-attn"]