From dc58c339fcf362a4e7c1e0481520acf9b28e8900 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Danie=CC=88l=20de=20Kok?= Date: Fri, 21 Jun 2024 14:26:33 +0200 Subject: [PATCH] Fix `text-generation-server quantize` The subcommand did not work due to some broken imports. --- server/text_generation_server/cli.py | 2 +- server/text_generation_server/layers/gptq/quantize.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/server/text_generation_server/cli.py b/server/text_generation_server/cli.py index 430323bc..5d25bfc5 100644 --- a/server/text_generation_server/cli.py +++ b/server/text_generation_server/cli.py @@ -316,7 +316,7 @@ def quantize( logger_level=logger_level, json_output=json_output, ) - from text_generation_server.utils.gptq.quantize import quantize + from text_generation_server.layers.gptq.quantize import quantize quantize( model_id=model_id, diff --git a/server/text_generation_server/layers/gptq/quantize.py b/server/text_generation_server/layers/gptq/quantize.py index ca113d8f..8d029817 100644 --- a/server/text_generation_server/layers/gptq/quantize.py +++ b/server/text_generation_server/layers/gptq/quantize.py @@ -12,7 +12,7 @@ from huggingface_hub import HfApi from accelerate import init_empty_weights from text_generation_server.utils import initialize_torch_distributed, Weights from text_generation_server.utils.hub import weight_files -from text_generation_server.utils.gptq.quant_linear import QuantLinear +from text_generation_server.layers.gptq.quant_linear import QuantLinear from loguru import logger from typing import Optional