diff --git a/backends/gaudi/Makefile b/backends/gaudi/Makefile index 77581517..c153a5ff 100644 --- a/backends/gaudi/Makefile +++ b/backends/gaudi/Makefile @@ -2,7 +2,7 @@ mkfile_path := $(abspath $(lastword $(MAKEFILE_LIST))) mkfile_dir := $(dir $(mkfile_path)) root_dir := ${mkfile_dir}/../.. -HABANA_VERSION := 1.21.0 +HABANA_VERSION := 1.20.0 PYTORCH_VERSION := 2.6.0 .PHONY: image run-local-dev-container install-dependencies install-server install-router install-launcher local-dev-install diff --git a/backends/gaudi/server/text_generation_server/models/custom_modeling/qwen2_5_vl.py b/backends/gaudi/server/text_generation_server/models/custom_modeling/qwen2_5_vl.py index 5f180458..441b0016 100644 --- a/backends/gaudi/server/text_generation_server/models/custom_modeling/qwen2_5_vl.py +++ b/backends/gaudi/server/text_generation_server/models/custom_modeling/qwen2_5_vl.py @@ -49,8 +49,7 @@ from text_generation_server.models.custom_modeling.flash_qwen2_modeling import ( # Copied from: https://github.com/huggingface/transformers/blob/main/src/transformers/models/qwen2_5_vl/processing_qwen2_5_vl.py from typing import Union from transformers.feature_extraction_utils import BatchFeature -from transformers.image_utils import ImageInput -from transformers.video_utils import VideoInput +from transformers.image_utils import ImageInput, VideoInput from transformers.processing_utils import ( ProcessingKwargs, ProcessorMixin, diff --git a/backends/gaudi/tgi-entrypoint.sh b/backends/gaudi/tgi-entrypoint.sh index 80ff1720..d787ea8e 100644 --- a/backends/gaudi/tgi-entrypoint.sh +++ b/backends/gaudi/tgi-entrypoint.sh @@ -12,7 +12,7 @@ if [[ "$ATTENTION" == "paged" ]]; then # Check if Llama-4 is in the command line arguments if [[ "$*" == *"Llama-4"* || "$*" == *"Qwen3"* ]]; then echo 'ATTENTION=paged and Llama-4 or Qwen3 detected' - pip install transformers==4.52.2 + pip install git+https://github.com/huggingface/transformers.git@29338949 fi fi