From 2923fe6ea5977d4838a9dbc6e5690b2feebefcf0 Mon Sep 17 00:00:00 2001 From: Eugene Rakhmatulin Date: Mon, 9 Feb 2026 10:21:14 -0800 Subject: [PATCH] Removed temp fastsafetensors patch --- Dockerfile | 4 ++-- Dockerfile.wheels | 4 ++-- README.md | 6 ++++++ 3 files changed, 10 insertions(+), 4 deletions(-) diff --git a/Dockerfile b/Dockerfile index a9d3f02..935a48e 100644 --- a/Dockerfile +++ b/Dockerfile @@ -216,8 +216,8 @@ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \ # Apply Patches # TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/foundation-model-stack/fastsafetensors/issues/36 -COPY fastsafetensors.patch . -RUN patch -p1 < fastsafetensors.patch +# COPY fastsafetensors.patch . +# RUN patch -p1 < fastsafetensors.patch # Final Compilation # We mount the ccache directory here. Ideally, map this to a host volume for persistence diff --git a/Dockerfile.wheels b/Dockerfile.wheels index e287f1c..9edc7c9 100644 --- a/Dockerfile.wheels +++ b/Dockerfile.wheels @@ -41,7 +41,7 @@ RUN mkdir -p tiktoken_encodings && \ wget -O tiktoken_encodings/cl100k_base.tiktoken "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken" # Cache TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/foundation-model-stack/fastsafetensors/issues/36 -COPY fastsafetensors.patch . +# COPY fastsafetensors.patch . # Install fastsafetensors RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \ @@ -68,7 +68,7 @@ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \ # Apply TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/foundation-model-stack/fastsafetensors/issues/36 # Apply in site-packages -RUN patch -p1 -d /usr/local/lib/python3.12/dist-packages < ${VLLM_BASE_DIR}/fastsafetensors.patch +# RUN patch -p1 -d /usr/local/lib/python3.12/dist-packages < ${VLLM_BASE_DIR}/fastsafetensors.patch ARG FLASHINFER_PRE="" diff --git a/README.md b/README.md index 66b06fb..7a2040d 100644 --- a/README.md +++ b/README.md @@ -159,6 +159,12 @@ Don't do it every time you rebuild, because it will slow down compilation times. For periodic maintenance, I recommend using a filter: `docker builder prune --filter until=72h` +### 2026-02-09 + +- Migrated to a new base image with PyTorch 2.10 compiled with Spark support. +- Triton 3.6.0 is now default. +- Removed temporary fastsafetensors patch, as proper fix is now merged into vLLM main branch. + ### 2026-02-04 #### Recipes support