summaryrefslogtreecommitdiff
path: root/candle-metal-kernels
diff options
context:
space:
mode:
authorLaurent Mazare <laurent.mazare@gmail.com>2024-05-04 10:14:57 +0200
committerGitHub <noreply@github.com>2024-05-04 10:14:57 +0200
commitb13a82a4387a55df07bec4e2eb6f7a8ebd0b98a2 (patch)
treeaed5a019e7e053900ffa5be57ddfd20bdfad8582 /candle-metal-kernels
parent59b18d974ec3cad6963b774aa245e23f8c80414f (diff)
downloadcandle-b13a82a4387a55df07bec4e2eb6f7a8ebd0b98a2.tar.gz
candle-b13a82a4387a55df07bec4e2eb6f7a8ebd0b98a2.tar.bz2
candle-b13a82a4387a55df07bec4e2eb6f7a8ebd0b98a2.zip
Separate quantized phi-3 implementation. (#2157)
* Separate quantized phi-3 implementation. * Integrate the quantized phi3 model.= * Small fixes, get the generation to work properly. * Keep the old llama implementation around. * Change the default.
Diffstat (limited to 'candle-metal-kernels')
-rw-r--r--candle-metal-kernels/src/lib.rs2
1 files changed, 1 insertions, 1 deletions
diff --git a/candle-metal-kernels/src/lib.rs b/candle-metal-kernels/src/lib.rs
index c08e44fe..814ca0b9 100644
--- a/candle-metal-kernels/src/lib.rs
+++ b/candle-metal-kernels/src/lib.rs
@@ -350,7 +350,7 @@ pub fn call_unary_contiguous_tiled(
let pipeline = kernels.load_pipeline(device, Source::Unary, kernel_name.0)?;
let encoder = command_buffer.new_compute_command_encoder();
let tile_size = 2;
- let tiles = length.div_ceil(tile_size);
+ let tiles = (length + tile_size - 1) / tile_size;
encoder.set_compute_pipeline_state(&pipeline);