diff --git a/serving/docker/deepspeed.Dockerfile b/serving/docker/deepspeed.Dockerfile index 442c1e0b6..cab7f2e6b 100644 --- a/serving/docker/deepspeed.Dockerfile +++ b/serving/docker/deepspeed.Dockerfile @@ -22,7 +22,7 @@ ARG rotary_emb_wheel="https://publish.djl.ai/flash_attn/rotary_emb-0.1-cp39-cp39 ARG flash_attn_2_wheel="https://publish.djl.ai/flash_attn/flash_attn_2-2.0.1-cp39-cp39-linux_x86_64.whl" ARG vllm_wheel="https://publish.djl.ai/vllm/vllm-0.1.1-cp39-cp39-linux_x86_64.whl" ARG lmi_dist_wheel="https://publish.djl.ai/lmi_dist/lmi_dist-nightly-py3-none-any.whl" -ARG seq_scheduler_wheel="https://publish.djl.ai/seq_scheduler/seq_scheduler-nightly-py3-none-any.whl" +ARG seq_scheduler_wheel="https://publish.djl.ai/seq_scheduler/seq_scheduler-0.1.0-py3-none-any.whl" ARG peft_wheel="https://publish.djl.ai/peft/peft-0.5.0alpha-py3-none-any.whl" ARG protobuf_version=3.20.3 ARG transformers_version=4.33.2 diff --git a/serving/docker/fastertransformer.Dockerfile b/serving/docker/fastertransformer.Dockerfile index 5b8714783..276c765b6 100644 --- a/serving/docker/fastertransformer.Dockerfile +++ b/serving/docker/fastertransformer.Dockerfile @@ -19,7 +19,7 @@ ARG torch_wheel="https://aws-pytorch-unified-cicd-binaries.s3.us-west-2.amazonaw ARG ft_wheel="https://publish.djl.ai/fastertransformer/fastertransformer-0.24.0-py3-none-any.whl" ARG tb_wheel="https://publish.djl.ai/tritonserver/r23.04/tritontoolkit-23.4-py3-none-any.whl" ARG peft_wheel="https://publish.djl.ai/peft/peft-0.5.0alpha-py3-none-any.whl" -ARG seq_scheduler_wheel="https://publish.djl.ai/seq_scheduler/seq_scheduler-nightly-py3-none-any.whl" +ARG seq_scheduler_wheel="https://publish.djl.ai/seq_scheduler/seq_scheduler-0.1.0-py3-none-any.whl" ARG ompi_version=4.1.4 ARG protobuf_version=3.20.3 ARG transformers_version=4.33.2