pin v2 payload: engine 2610182149001, cuda 2610160820001, cuda12 2610270912001, sbsa 2610141455003, vulkan 2610182214001, macos 2610182214002, media 2610071902001

.gitattributes

@@ -307,3 +307,5 @@ components/engine/2610182149001/opencoti-0.10.5-c13-2610182149001.exe filter=lfs
 components/vulkan/2610182214001/ggml-vulkan-x86_64.so filter=lfs diff=lfs merge=lfs -text
 components/vulkan/2610182214001/ggml-vulkan-win-x86_64.dll filter=lfs diff=lfs merge=lfs -text
 components/macos/2610182214002/ggml-metal-aarch64.dylib filter=lfs diff=lfs merge=lfs -text
+components/cuda12/2610270912001/ggml-cuda-cu12-x86_64.so filter=lfs diff=lfs merge=lfs -text
+components/cuda12/2610270912001/ggml-cuda-cu12-x86_64.dll filter=lfs diff=lfs merge=lfs -text

components/cuda12/2610270912001/BUILD_INFO.cuda12.md

@@ -0,0 +1,34 @@
+# cuda12 2610270912001
+
+Component of the opencoti pin format v2 (docs/protocols/PIN_FORMAT.md).
+
+Built from the engine source of build 2610141255001; gated floor (engine-min) 2610141212001; gated with 2610182159001.
+
+| file | platform | kind | sha256 | bytes |
+|---|---|---|---|---|
+| `ggml-cuda-cu12-x86_64.so` | x86_64 | cuda12 | `6e9f87bbff798cf4b33f2504aab819da5b33b91c334dc2003a7c30d3f9ae32e5` | 261655040 |
+| `ggml-cuda-cu12-x86_64.dll` | win-x86_64 | cuda12 | `3a7700971d89bf14f3810a4750bf5b3b18ef1e40f38723e921c1207696e6c9f8` | 244886528 |
+
+ABI `ggml` = `bac5050ddc8c58c51f0b059a7f9753b33af0ed79d355090e7c9a17ffaaaf340a` (exported) — sha256 of this list:
+
+```
+llama.cpp/ggml/include/ggml-alloc.h 94e4cd069b9313b2ceb35dacec901981e0bb478d8bb31035b7126be091998c23
+llama.cpp/ggml/include/ggml-backend.h 764d1640ad62579513c4837539708f89e8603c9dbeae053a13d1cbd84a82c865
+llama.cpp/ggml/include/ggml-cuda.h 674ea064021cf5b9970e78f1209e513dd23350b94a44a705a27eec1423fe4b0a
+llama.cpp/ggml/include/ggml-metal.h 322f36cd30f3e9e7aad7b5b9bc63078012fd0b7706ac4e24f5721b604f3d8980
+llama.cpp/ggml/include/ggml-neo-pipeline.h 190f39e183542d16588a3104d8cb4359bf52fc7a62afe8d8cc4c822f4085086d
+llama.cpp/ggml/include/ggml-oc-kvarn-range.h 76635fdd8978632f670640ddf5125a666441e3d47c680bb8b5c0a099e4861202
+llama.cpp/ggml/include/ggml-oc-ledger.h c1be11d69a57a1e721f1ae8720b87887d239ac0183d4277fbaa07f1be257d20c
+llama.cpp/ggml/include/ggml-vulkan.h f9df4f7161a84ce7600d53c3fac8a62837d5716676c9f38cd340699ef1c070d9
+llama.cpp/ggml/include/ggml.h ecfbf114f9e9151d22e30ef6d5139570c5c41b595843ca4f99125bd0196709aa
+llama.cpp/ggml/include/opencoti-devtype.h 94dea946173449bb76bffd7f733ab822fe658fe7198992de750ac012d29308ec
+llama.cpp/ggml/src/ggml-backend-impl.h f9fdfe6bae61d1387593b40fc0f90ceee490c6a339ec6d7388c51e4fca318d8f
+llamafile/gpu_backend.h 4929c36c034b6589e88653bcb7a5c5a225743a07320f64c0857d92bdaf048208
+```
+
+**2610270912001 (2026-10-27, emergency, bug-4246):** adds the **Windows** CUDA 12 library `ggml-cuda-cu12-x86_64.dll` (every earlier cuda12 component was Linux-only, so a pre-Turing NVIDIA GPU on Windows had no CUDA code). sm_52 / 61 / 70, CUDA 12.8.1 Windows toolkit (nvcc 12.8.93), MSVC 14.44 under wine on bs2, built from the c13 r6 release pin's source tree; ABI ggml bac5050d (the c13 engine's). TinyBLAS with the CUTLASS sm_70 GEMM (CUTLASS v4.8.0, 098de2a652cf), the same tinyblas.cu as the Linux .so; `OPENCOTI_TINYBLAS_CUTLASS=0` turns it off. The Linux .so is unchanged (the 2610141455002 bytes).
+Gated on a GTX 1070 (cc 6.1, Windows 10, driver 582.66) with the c13 engine 2610182159001 and gemma-4 E2B QAT Q4_K_XL: the engine picks this library with no override, every per-token weight on the GPU, 512 tokens per arm (plain 68.4 tok/s, FA + q8_0 K/V 62.9 tok/s; Vulkan on the same card 83.3 tok/s), the published-name file alone loads and generates. RTX 3090 (cc 8.6): not taken without an override; with OPENCOTI_CUDA_LEGACY=1 it loads and finds no sm_86 code (expected). The CUTLASS sm_70 kernel itself ran on no card in this cut (no Volta GPU on Windows); its Linux twin was validated on a V100 (0454).
+The new single file `opencoti-llamafile-0.10.5-c13-win-x86_64-gpu-cuda12.llamafile.exe` (the bare engine + this DLL + the c13 Vulkan DLL) run alone: GTX 1070 auto and --gpu nvidia extract and load this DLL (67.7 tok/s, 512 tokens), --gpu vulkan loads the Vulkan DLL; on the RTX 3090 it takes Vulkan (no CUDA 13 library inside).
+
+CUDA 12.8, for GPUs older than sm_75 (sm_52, 61, 70). Chain 0662, from the c12 object cache. **These bytes have not been run
+on a card** (no such GPU reachable at the cut).

components/cuda12/2610270912001/ggml-cuda-cu12-x86_64.dll

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a7700971d89bf14f3810a4750bf5b3b18ef1e40f38723e921c1207696e6c9f8
+size 244886528

components/cuda12/2610270912001/ggml-cuda-cu12-x86_64.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e9f87bbff798cf4b33f2504aab819da5b33b91c334dc2003a7c30d3f9ae32e5
+size 261655040