From 7539881054384f90ab1fc01c3f935e1768c3d463 Mon Sep 17 00:00:00 2001 From: Nicolas Patry Date: Thu, 30 Jan 2025 11:05:21 +0100 Subject: [PATCH] Fixing moe import. --- server/text_generation_server/layers/fp8.py | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/server/text_generation_server/layers/fp8.py b/server/text_generation_server/layers/fp8.py index 7b363579..f0270ff8 100644 --- a/server/text_generation_server/layers/fp8.py +++ b/server/text_generation_server/layers/fp8.py @@ -5,7 +5,6 @@ from typing import Optional, Tuple, Type, Union, List import torch from loguru import logger -from moe_kernels.fp8_utils import w8a8_block_fp8_matmul, per_token_group_quant_fp8 from text_generation_server.utils.import_utils import SYSTEM from text_generation_server.utils.weights import ( Weight, @@ -20,6 +19,12 @@ try: except ImportError: marlin_kernels = None +try: + from moe_kernels.fp8_utils import w8a8_block_fp8_matmul, per_token_group_quant_fp8 +except ImportError: + w8a8_block_fp8_matmul = None + per_token_group_quant_fp8 = None + quant_dtype: torch.dtype = ( torch.float8_e4m3fnuz if SYSTEM == "rocm" else torch.float8_e4m3fn )