From 273d978bedc89bc8cb1a5c4d57d9ea04aea2cc9c Mon Sep 17 00:00:00 2001 From: Baizhou Zhang Date: Wed, 12 Aug 2026 16:21:40 -0700 Subject: [PATCH] [CI] Use default installer for B300 tests (#34635) --- scripts/ci/cuda/ci_install_kimi_k3.sh | 58 --------------------------- scripts/ci/runner_configs.yml | 3 +- 2 files changed, 1 insertion(+), 60 deletions(-) delete mode 100755 scripts/ci/cuda/ci_install_kimi_k3.sh diff --git a/scripts/ci/cuda/ci_install_kimi_k3.sh b/scripts/ci/cuda/ci_install_kimi_k3.sh deleted file mode 100755 index f66d4af9e..000000000 --- a/scripts/ci/cuda/ci_install_kimi_k3.sh +++ /dev/null @@ -1,58 +0,0 @@ -#!/bin/bash -# Install the standard CUDA CI dependencies plus Kimi-K3's Transformers fix. -set -euxo pipefail - -SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" - -# Source (not bash) so the generic install's Python/venv selection remains -# active while applying the Transformers compatibility fix. -# shellcheck disable=SC1091 -source "${SCRIPT_DIR}/ci_install_dependency.sh" "$@" - -apply_transformers_symlink_patch() { - # transformers 5.12.1 resolves custom-code symlinks out of the HF snapshot - # and into blobs/, then looks for relative imports by filename in blobs/. - # Mirror the hot patch in docker/rocm.Dockerfile until upstream PR #46618 - # reaches a transformers release. - python3 - <<'PY' -import importlib -import pathlib -import tempfile - -import transformers.dynamic_module_utils as dynamic_module_utils - -marks = ["Path(resolved_module_file).resolve()", "Path(source_file).resolve()"] -path = pathlib.Path(dynamic_module_utils.__file__) -src = path.read_text() -if not any(mark in src for mark in marks): - print("transformers dynamic_module_utils already fixed; no patch needed") -else: - patched = ( - src.replace( - "Path(resolved_module_file).resolve()", "Path(resolved_module_file)" - ).replace("Path(source_file).resolve()", "Path(source_file)") - ) - assert patched != src, "FATAL: transformers symlink patch matched nothing" - path.write_text(patched) - print("patched transformers dynamic_module_utils.py (symlink hash fix)") - -# Exercise the exact HF-cache layout that failed in the B300 Kimi-K3 run. -dynamic_module_utils = importlib.reload(dynamic_module_utils) -with tempfile.TemporaryDirectory() as tmp: - cache = pathlib.Path(tmp) / "models--org--model" - blobs = cache / "blobs" - snapshot = cache / "snapshots" / "revision" - blobs.mkdir(parents=True) - snapshot.mkdir(parents=True) - (blobs / "modeling-blob").write_text("from .media_utils import VALUE\n") - (blobs / "media-blob").write_text("VALUE = 1\n") - (snapshot / "modeling.py").symlink_to("../../blobs/modeling-blob") - (snapshot / "media_utils.py").symlink_to("../../blobs/media-blob") - source_hash = dynamic_module_utils._compute_local_source_files_hash( - snapshot, snapshot / "modeling.py" - ) - assert len(source_hash) == 16, source_hash -PY -} - -apply_transformers_symlink_patch diff --git a/scripts/ci/runner_configs.yml b/scripts/ci/runner_configs.yml index 70e3045bb..b9c032b1b 100644 --- a/scripts/ci/runner_configs.yml +++ b/scripts/ci/runner_configs.yml @@ -16,7 +16,6 @@ _anchors: default_install: &default scripts/ci/cuda/ci_install_dependency.sh - kimi_k3_install: &kimi_k3 scripts/ci/cuda/ci_install_kimi_k3.sh runner_configs: 1-gpu-small: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 1-gpu-5090 } @@ -27,5 +26,5 @@ runner_configs: 4-gpu-h100: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 4-gpu-h100 } 8-gpu-h200: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 8-gpu-h200 } 8-gpu-b200: { install: *default, artifact_version: v6, install_timeout: "20", runs_on: 8-gpu-b200 } - 8-gpu-b300: { install: *kimi_k3, artifact_version: v6, install_timeout: "20", runs_on: 8-gpu-b300 } + 8-gpu-b300: { install: *default, artifact_version: v6, install_timeout: "20", runs_on: 8-gpu-b300 } 8-gpu-h20: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 8-gpu-h20, rdma_devices: "mlx5_1,mlx5_2,mlx5_3,mlx5_4" }