lmcache 0.4.8.dev21__tar.gz → 0.4.8.dev24__tar.gz

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/.buildkite/k3_tests/multiprocess/pipeline.yml

@@ -74,8 +74,8 @@ steps:
         timeout_in_minutes: 60
         env:
           MODEL: "google/gemma-4-31B-it"
-          # Require an exact score match between the two runs.
-          SCORE_TOLERANCE: "0"
+          # Allow a small score difference between the two runs.
+          SCORE_TOLERANCE: "0.03"
           ATTENTION_BACKEND: "auto"
           GPU_MEMORY_UTILIZATION: "0.85"
           # 31B load + CUDA-graph capture is slow; raise the readiness timeout

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/.github/workflows/build_cli_artifacts.yml

@@ -22,9 +22,9 @@ defaults:
     run:
         shell: bash
 
-# Cancel stale runs per ref; PR and dev-push refs differ, so workflow_call is unaffected.
+# Cancel older in-flight runs on the same PR / branch.
 concurrency:
-    group: ${{ github.workflow }}-${{ github.ref }}
+    group: build-cli-${{ github.workflow }}-${{ github.ref }}
     cancel-in-progress: true
 
 permissions:

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/.github/workflows/code_quality_checks.yml

@@ -4,10 +4,9 @@ on:
   workflow_call:
   pull_request:
 
-# Cancel older in-flight runs on the same PR / branch. github.ref is
-# refs/pull/<N>/merge on pull_request events, so each PR gets its own group.
+# Cancel older in-flight runs on the same PR / branch.
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: code-quality-${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
 permissions:

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/.github/workflows/publish.yml

@@ -18,6 +18,11 @@ on:
 env:
     LC_ALL: en_US.UTF-8
 
+# Cancel superseded runs on the same ref, but never abort a tag/release publish.
+concurrency:
+    group: publish-${{ github.ref }}
+    cancel-in-progress: ${{ !startsWith(github.ref, 'refs/tags/') && github.event_name != 'release' }}
+
 defaults:
     run:
         shell: bash
@@ -103,8 +108,12 @@ jobs:
     # - a commit lands on dev (PR merged)
     publish-test-pypi:
         name: Publish packages to test.pypi.org
+        # Require upstream jobs to have succeeded so a cancelled/skipped build
+        # doesn't trigger a publish that fails on the missing artifact.
         if: |
-            !failure() && !cancelled() &&
+            needs.build-main.result == 'success' &&
+            needs.test.result == 'success' &&
+            needs.code-quality.result == 'success' &&
             needs.changes.outputs.lmcache == 'true' &&
             github.repository_owner == 'LMCache' &&
             github.event_name == 'push' && github.ref == 'refs/heads/dev'
@@ -137,7 +146,9 @@ jobs:
     publish-cli-test-pypi:
         name: Publish lmcache-cli packages to test.pypi.org
         if: |
-            !failure() && !cancelled() &&
+            needs.build-cli.result == 'success' &&
+            needs.test.result == 'success' &&
+            needs.code-quality.result == 'success' &&
             needs.changes.outputs.lmcache == 'true' &&
             github.repository_owner == 'LMCache' &&
             github.event_name == 'push' && github.ref == 'refs/heads/dev'
@@ -171,7 +182,9 @@ jobs:
     publish-pypi:
         name: Publish release to pypi.org
         if: |
-            !failure() && !cancelled() &&
+            needs.build-main.result == 'success' &&
+            needs.test.result == 'success' &&
+            needs.code-quality.result == 'success' &&
             needs.changes.outputs.lmcache == 'true' &&
             github.repository_owner == 'LMCache' && github.event.action == 'published'
         permissions:
@@ -210,7 +223,9 @@ jobs:
     publish-cli-pypi:
         name: Publish lmcache-cli release to pypi.org
         if: |
-            !failure() && !cancelled() &&
+            needs.build-cli.result == 'success' &&
+            needs.test.result == 'success' &&
+            needs.code-quality.result == 'success' &&
             needs.changes.outputs.lmcache == 'true' &&
             github.repository_owner == 'LMCache' && github.event.action == 'published'
         permissions:
@@ -247,7 +262,9 @@ jobs:
     publish-cu129-github-release:
         name: Publish cu129 wheel to GitHub Release
         if: |
-            !failure() && !cancelled() &&
+            needs.build-cu129.result == 'success' &&
+            needs.test.result == 'success' &&
+            needs.code-quality.result == 'success' &&
             needs.changes.outputs.lmcache == 'true' &&
             github.repository_owner == 'LMCache' && github.event.action == 'published'
         permissions:

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/.github/workflows/test.yml

@@ -16,10 +16,9 @@ defaults:
   run:
     shell: bash
 
-# Cancel older in-flight runs on the same PR / branch. github.ref is
-# refs/pull/<N>/merge on pull_request events, so each PR gets its own group.
+# Cancel older in-flight runs on the same PR / branch.
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: test-${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
 permissions:

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lmcache
-Version: 0.4.8.dev21
+Version: 0.4.8.dev24
 Summary: A LLM serving engine extension to reduce TTFT and increase throughput, especially under long-context scenarios.
 Author-email: LMCache Team <lmcacheteam@gmail.com>
 License-Expression: Apache-2.0

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/docker/Dockerfile

@@ -138,7 +138,8 @@ RUN . /opt/venv/bin/activate && \
     VLLM_PRECOMPILED_WHEEL_VARIANT=${CUDA_TAG} uv pip install --prerelease=allow \
         vllm[runai,tensorizer,flashinfer] \
         --extra-index-url https://wheels.vllm.ai/nightly/${CUDA_TAG} \
-        --extra-index-url https://download.pytorch.org/whl/${CUDA_TAG} && \
+        --extra-index-url https://download.pytorch.org/whl/${CUDA_TAG} \
+        --index-strategy unsafe-first-match && \
     python3 -c 'import torch; print("TORCH=", torch.__version__)' && \
     uv pip install lmcache==${VER} \
         --extra-index-url https://download.pytorch.org/whl/${CUDA_TAG} \

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/lmcache/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.4.8.dev21'
-__version_tuple__ = version_tuple = (0, 4, 8, 'dev21')
+__version__ = version = '0.4.8.dev24'
+__version_tuple__ = version_tuple = (0, 4, 8, 'dev24')
 
-__commit_id__ = commit_id = 'g5d621e1e'
+__commit_id__ = commit_id = 'gf52b174c'

{lmcache-0.4.8.dev21 → lmcache-0.4.8.dev24}/lmcache.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lmcache
-Version: 0.4.8.dev21
+Version: 0.4.8.dev24
 Summary: A LLM serving engine extension to reduce TTFT and increase throughput, especially under long-context scenarios.
 Author-email: LMCache Team <lmcacheteam@gmail.com>
 License-Expression: Apache-2.0