Expose to/from fp8 in Python and don't auto-convert fp8 when loading from safetensors (#2985)

2026-01-13 15:48:21 -08:00
parent 8654b8281d
commit 099dcc0f4c
6 changed files with 63 additions and 38 deletions
@@ -1,5 +1,4 @@
 cuda_skip = {
-    "TestLoad.test_load_f8_e4m3",
    "TestLayers.test_quantized_embedding",
    # Block masked matmul NYI
    "TestBlas.test_block_masked_matmul",