Script 'mail_helper' called by obssrc
Hello community,

here is the log from the commit of package python-vllm for openSUSE:Factory 
checked in at 2026-08-11 17:15:40
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Comparing /work/SRC/openSUSE:Factory/python-vllm (Old)
 and      /work/SRC/openSUSE:Factory/.python-vllm.new.17972 (New)
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Package is "python-vllm"

Tue Aug 11 17:15:40 2026 rev:6 rq:1370608 version:0.27.0

Changes:
--------
--- /work/SRC/openSUSE:Factory/python-vllm/python-vllm.changes  2026-08-09 
21:34:34.996633670 +0200
+++ /work/SRC/openSUSE:Factory/.python-vllm.new.17972/python-vllm.changes       
2026-08-11 17:16:51.293332010 +0200
@@ -1,0 +2,55 @@
+Tue Aug 11 05:49:32 UTC 2026 - Martin Pluskal <[email protected]>
+
+- Update to version 0.27.0:
+  * Kimi K3 support landing as a complete stack: model files and
+    kernels, Python and Rust frontends, AttnRes kernels, DeepGEMM
+    support, compressed-tensors quantized checkpoints and an option
+    to shard the shared expert instead of replicating it
+  * Further new models: Qwen3.5 text-only dense and MoE with EVS
+    video token pruning, K-EXAONE-2.0-750B-A37B, VaultGemma through
+    the Transformers modeling backend, jina-embeddings-v5-text-nano
+  * FlashAttention 4 on SM100 gains FP8 KV cache and headdim-256
+    support, backed by new JIT and runner-owned Triton warmup
+    infrastructure that removes first-request compilation stalls
+  * DeepSeek-V4 performance work: sequence parallelism, roughly 2x
+    kernel improvement from skipping empty c128 launches, workspace
+    reuse, adaptive topk width, a compact MXFP4 indexer KV cache and
+    448 MiB less GPU memory in the pipeline-parallel buffer
+  * Model Runner V2 extends to non-generative workloads: encoder-only
+    attention, sequence pooling for embedding and classification,
+    encoder token classification and embedding, BGE-M3 pooling and
+    multimodal on CPU
+  * Simplified fault tolerance framework for data-parallel and
+    expert-parallel deployments behind an external load balancer,
+    plus async preparation for elastic expert-parallel scaling
+  * Prefill/decode disaggregation for hybrid models: NIXL support for
+    hybrid MLA+SSM models, heterogeneous block sizes, and MoRIIO
+    heterogeneous read routing
+  * The Rust frontend grows a gRPC control plane covering health
+    reporting, abort control, server/model discovery and KV event
+    source discovery, and integrates vllm-bench into the vllm CLI
+  * CPU backend: INT8 fused MoE kernel for Arm CPUs, s390x oneDNN
+    INT8 GEMM, optimised GDN conv path for speculative decoding,
+    granite-4 enabled, and the CPU kernels bumped to the latest
+    version
+  * Quantization: FP4 Qutlass for compressed-tensors, CuTeDSL MoE for
+    ReLU2 NVFP4, MXFP8 linear in INC, AutoRound W4A16 MoE, and a KV
+    quant mode for TurboQuant
+  * API and frontend: Cohere chat v2, cache_salt in the Anthropic
+    Messages API, strict tool calling for GPT-OSS Harmony, a
+    per-request stream_interval sampling parameter and a standardised
+    VLLMError hierarchy
+  * Early enablement for NVIDIA Rubin (sm_107) and ROCm gfx1250
+  * Removed the Plamo2 and Ouro models, and the no longer supported
+    max_num_partial_prefills and max_long_partial_prefills arguments
+  * Upstream moved its pinned build environment to PyTorch 2.13.0;
+    this package keeps building against the 2.12.0 that Factory ships
+- Raise the python-mistral-common requirement to >= 1.11.6, matching
+  upstream's own floor for this release
+- Rebase vllm-relax-cpu-requirements.patch and
+  vllm-cpu-disable-rust-frontend.patch onto 0.27.0:
+  * vllm-relax-cpu-requirements.patch now relaxes the torch pin to
+    >= 2.12.0, tracking what Factory ships rather than upstream's
+    exact 2.13.0
+
+-------------------------------------------------------------------

Old:
----
  vllm-0.26.0.tar.gz

New:
----
  vllm-0.27.0.tar.gz

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Other differences:
------------------
++++++ python-vllm.spec ++++++
--- /var/tmp/diff_new_pack.bArvyq/_old  2026-08-11 17:16:53.493425240 +0200
+++ /var/tmp/diff_new_pack.bArvyq/_new  2026-08-11 17:16:53.497425410 +0200
@@ -47,7 +47,7 @@
 %define         onednn_x86_version 3.10
 %define         acl_version 52.6.0
 Name:           python-vllm%{psuffix}
-Version:        0.26.0
+Version:        0.27.0
 Release:        0
 Summary:        A high-throughput and memory-efficient inference and serving 
engine for LLMs
 License:        Apache-2.0
@@ -108,7 +108,7 @@
 Requires:       python-llguidance >= 1.7.0
 Requires:       python-lm-format-enforcer = 0.11.3
 Requires:       python-mcp
-Requires:       python-mistral-common >= 1.11.5
+Requires:       python-mistral-common >= 1.11.6
 Requires:       python-model-hosting-container-standards >= 0.1.14
 Requires:       python-msgspec
 Requires:       python-numba >= 0.65.0
@@ -206,7 +206,7 @@
 %endif
 
 # Use the torch already installed in the build root (2.12.0) instead of the
-# exact 2.11.0 pin, via vLLM's own helper.  Strips torch/torchvision/torchaudio
+# exact 2.13.0 pin, via vLLM's own helper.  Strips torch/torchvision/torchaudio
 # pins from requirements/*.txt and pyproject.toml.
 %python_expand $python use_existing_torch.py --prefix
 

++++++ vllm-0.26.0.tar.gz -> vllm-0.27.0.tar.gz ++++++
/work/SRC/openSUSE:Factory/python-vllm/vllm-0.26.0.tar.gz 
/work/SRC/openSUSE:Factory/.python-vllm.new.17972/vllm-0.27.0.tar.gz differ: 
char 5, line 1

++++++ vllm-cpu-disable-rust-frontend.patch ++++++
--- /var/tmp/diff_new_pack.bArvyq/_old  2026-08-11 17:16:53.601429816 +0200
+++ /var/tmp/diff_new_pack.bArvyq/_new  2026-08-11 17:16:53.605429986 +0200
@@ -1,6 +1,6 @@
---- a/setup.py 2026-07-13 08:06:59.849797035 +0200
-+++ b/setup.py 2026-07-13 08:12:53.729680498 +0200
-@@ -1231,9 +1231,16 @@
+--- a/setup.py
++++ b/setup.py
+@@ -1257,9 +1257,16 @@
  
  # Rust artifacts, built via setuptools-rust and installed into the package
  # directory alongside the Python modules.

++++++ vllm-relax-cpu-requirements.patch ++++++
--- /var/tmp/diff_new_pack.bArvyq/_old  2026-08-11 17:16:53.625430833 +0200
+++ /var/tmp/diff_new_pack.bArvyq/_new  2026-08-11 17:16:53.633431172 +0200
@@ -11,8 +11,8 @@
 +numba >= 0.65.0; platform_machine != "s390x" # Required for N-gram 
speculative decoding
  
  # Dependencies for CPUs
--torch==2.11.0+cpu; platform_machine == "x86_64" or platform_machine == 
"s390x" or platform_machine == "aarch64"
--torch==2.11.0; platform_system == "Darwin" or platform_machine == "ppc64le" 
or platform_machine == "riscv64"
+-torch==2.13.0+cpu; platform_machine == "x86_64" or platform_machine == 
"s390x" or platform_machine == "aarch64"
+-torch==2.13.0; platform_system == "Darwin" or platform_machine == "ppc64le" 
or platform_machine == "riscv64"
 -
 -# required for the image processor of minicpm-o-2_6, this must be updated 
alongside torch
 -torchaudio; platform_machine != "s390x" and platform_machine != "riscv64"
@@ -25,8 +25,8 @@
 -
 -# Intel Extension for PyTorch, only for x86_64 CPUs
 -intel-openmp==2024.2.1; platform_machine == "x86_64"
-+torch>=2.11.0; platform_machine == "x86_64" or platform_machine == "s390x" or 
platform_machine == "aarch64"
-+torch>=2.11.0; platform_system == "Darwin" or platform_machine == "ppc64le" 
or platform_machine == "riscv64"
++torch>=2.12.0; platform_machine == "x86_64" or platform_machine == "s390x" or 
platform_machine == "aarch64"
++torch>=2.12.0; platform_system == "Darwin" or platform_machine == "ppc64le" 
or platform_machine == "riscv64"
  
  # Use this to gather CPU info and optimize based on ARM Neoverse cores
  py-cpuinfo; platform_machine == "aarch64"

Reply via email to