Add missing kernel for CodeLlama-34B on A/H100 (no tensor parallelism) when using Multi-LoRA. (#3350)

2024-03-13 21:18:25 +02:00
parent 739c350c19
commit ae0ccb4017
2 changed files with 2 additions and 1 deletions
--- a/tests/lora/test_punica.py
+++ b/tests/lora/test_punica.py
@ -45,7 +45,7 @@ def _lora_ref_impl(
 H1 = H2 = [
    128, 256, 512, 1024, 1280, 2048, 2560, 2752, 3072, 3456, 3584, 4096, 5120,
    5504, 5632, 6144, 6912, 7168, 8192, 9216, 10240, 11008, 13824, 14336,
-    24576, 32000, 32256, 32512, 32768, 33024
+    22016, 24576, 32000, 32256, 32512, 32768, 33024
 ]
 SEED = [0xabcdabcd987]