Script 'mail_helper' called by obssrc
Hello community,

here is the log from the commit of package python-comfy-kitchen for 
openSUSE:Factory checked in at 2026-10-05 12:56:54
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Comparing /work/SRC/openSUSE:Factory/python-comfy-kitchen (Old)
 and      /work/SRC/openSUSE:Factory/.python-comfy-kitchen.new.3698736 (New)
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Package is "python-comfy-kitchen"

Mon Oct  5 12:56:54 2026 rev:7 rq:1382318 version:0.2.37

Changes:
--------
--- 
/work/SRC/openSUSE:Factory/python-comfy-kitchen/python-comfy-kitchen.changes    
    2026-09-29 17:55:39.953918041 +0200
+++ 
/work/SRC/openSUSE:Factory/.python-comfy-kitchen.new.3698736/python-comfy-kitchen.changes
   2026-10-05 12:58:01.019752962 +0200
@@ -1,0 +2,21 @@
+Sun Oct  4 06:35:44 UTC 2026 - Martin Pluskal <[email protected]>
+
+- Update to 0.2.37:
+  * Eager backend: quantize_signed_int4_rowwise is now a thin wrapper
+    over a new _quantize_signed_int4_rowwise_unpacked, which returns
+    signed INT4 codes in INT8 storage without packing nibbles. The
+    public function keeps its name, signature and packed return value
+  * Ascend convrot_w4a4_linear keeps the activation's INT4 codes in
+    INT8 containers instead of packing them and immediately unpacking
+    them again, cutting a pack/unpack round-trip off the NPU path.
+    Ascend RoPE coverage is now documented as requiring
+    npu_rotary_mul with rotary_mode (RMS-RoPE also npu_rms_norm)
+  * CUDA W4A8 GEMM now sizes its weight chunks from the device L2
+    cache: below 16 MiB of L2 a chunk cannot stay resident, so the
+    decode runs full-width instead. Upstream measures up to 20%
+    faster at M=512 on a 3080
+  * Remainder of the release is CUDA and HIP kernel work (dlpack
+    bindings, CUTLASS int8 GEMM, W4A8 requant, int8 attention), not
+    compiled in this CPU-only build
+
+-------------------------------------------------------------------

Old:
----
  comfy-kitchen-0.2.36.tar.gz

New:
----
  comfy-kitchen-0.2.37.tar.gz

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

Other differences:
------------------
++++++ python-comfy-kitchen.spec ++++++
--- /var/tmp/diff_new_pack.qdXTpE/_old  2026-10-05 12:58:02.174801411 +0200
+++ /var/tmp/diff_new_pack.qdXTpE/_new  2026-10-05 12:58:02.178801579 +0200
@@ -18,7 +18,7 @@
 
 %{?sle15_python_module_pythons}
 Name:           python-comfy-kitchen
-Version:        0.2.36
+Version:        0.2.37
 Release:        0
 Summary:        Fast kernel library for ComfyUI (CPU/eager backend)
 # Legal-Review-Notice: NOTICE names the NVIDIA CUDA runtime EULA for a

++++++ comfy-kitchen-0.2.36.tar.gz -> comfy-kitchen-0.2.37.tar.gz ++++++
++++ 3342 lines of diff (skipped)

Reply via email to