pin v2 payload: engine 2610160712001, cuda 2610160820001, cuda12 2610141455002, sbsa 2610141455003, vulkan 2610141455004, macos 2610141455005, media 2610071902001

.gitattributes

@@ -294,3 +294,7 @@ components/sbsa/2610141455003/ggml-cuda-sbsa-aarch64.so filter=lfs diff=lfs merg
 components/vulkan/2610141455004/ggml-vulkan-x86_64.so filter=lfs diff=lfs merge=lfs -text
 components/vulkan/2610141455004/ggml-vulkan-win-x86_64.dll filter=lfs diff=lfs merge=lfs -text
 components/macos/2610141455005/ggml-metal-aarch64.dylib filter=lfs diff=lfs merge=lfs -text
+components/engine/2610160712001/opencoti-0.10.5-c13-2610160712001 filter=lfs diff=lfs merge=lfs -text
+components/engine/2610160712001/opencoti-0.10.5-c13-2610160712001.exe filter=lfs diff=lfs merge=lfs -text
+components/cuda/2610160820001/ggml-cuda-x86_64.so filter=lfs diff=lfs merge=lfs -text
+components/cuda/2610160820001/ggml-cuda-win-x86_64.dll filter=lfs diff=lfs merge=lfs -text

components/cuda/2610160820001/BUILD_INFO.cuda.md

@@ -0,0 +1,34 @@
+# cuda 2610160820001
+
+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 2610160712001, 2610160746001.
+
+| file | platform | kind | sha256 | bytes |
+|---|---|---|---|---|
+| `ggml-cuda-x86_64.so` | x86_64 | cuda | `1beda05b1dbf406796975bedfdd55cdd51ad4bc07696ba28c3aa0ca0e3c8acbb` | 757231168 |
+| `ggml-cuda-win-x86_64.dll` | win-x86_64 | cuda | `68cd73b1afe125bb54d2e9d62320b07cf4f38617e725d54bf603ef5a9ccd151d` | 732354560 |
+
+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
+```
+
+**c13 r3 (2026-10-16):** the Windows DLL is rebuilt with `GGML_CUDA_FA_ALL_QUANTS` (bug-4090: every Windows DLL since c10 r3 lacked it, so flash attention with K != V had no CUDA kernel and ran on the CPU), the same six generations (read from the DLL's own fatbin), from the c13 CUDA sources (no CUDA input changed since) on bs2's content-keyed wine lane; the Linux library is the c13 bytes, unchanged. Gated with the r3 engine 2610160746001: the glibc 2.28 image (Linux), greedy identity with r2 (Linux), eleven2go RTX 3090 f16-K / q8_0-V 84.8 tok/s with no CPU flash attention (the r2 DLL: 30.9).
+
+CUDA 13. Six generations (sm_75, 80, 86, 89, 90, 120); `120` is the sm_120f kernel set. Built from the clean-room tree of
+chain 0662 on a cloud runner, from the c12 round's object cache (the interface is unchanged): the six-generation library compiled 36 of its 405 units (the KVarN decode instances 0657 changed) and reused 369 from the c12 round's cache by content; the Windows DLL compiled the same 36 units plus ggml-backend.cpp under wine on bs2 and reused the c12 release's pinned objects by content (the set the content diff of the two source trees names). The Windows DLL is
+the same six generations (wine lane). Carries the lazy KVarN stage of 0657.
+Gated with the 2610141255001 engine: bs2 (sm_120) Bonsai 2 gate 56 / 56 and the slot gate 6 / 6; the glibc 2.28 image; the Windows DLL on eleven2go (RTX 3090, sm_86): the Windows GPU gate's CUDA cases, 77.4 tok/s against the c12 control's 77.5 with the same 64 tokens, no ABI mismatch, memory on the device the counter names, no spill. sm_75 / 80 / 89 / 90 were not run on a card at this cut (sm_86 = the RTX 3090, sm_120 = bs2).

components/cuda/2610160820001/ggml-cuda-win-x86_64.dll

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68cd73b1afe125bb54d2e9d62320b07cf4f38617e725d54bf603ef5a9ccd151d
+size 732354560

components/cuda/2610160820001/ggml-cuda-x86_64.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1beda05b1dbf406796975bedfdd55cdd51ad4bc07696ba28c3aa0ca0e3c8acbb
+size 757231168

components/engine/2610160712001/BUILD_INFO.engine.md

@@ -0,0 +1,50 @@
+# engine 2610160712001
+
+Component of the opencoti pin format v2 (docs/protocols/PIN_FORMAT.md).
+
+| file | platform | kind | sha256 | bytes |
+|---|---|---|---|---|
+| `opencoti-0.10.5-c13-2610160712001` | any | bin | `b81b2952cccae11d3cd16331f49ee7628ef6c89761780d41cfc2ff0481d20e38` | 145615678 |
+| `opencoti-0.10.5-c13-2610160712001.exe` | win-x86_64 | bin | `b81b2952cccae11d3cd16331f49ee7628ef6c89761780d41cfc2ff0481d20e38` | 145615678 |
+
+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
+```
+
+ABI `media` = `b099409b3029a6e380268e31d1caefa6706569961d32e72b3543fbc73903d95c` (exported) — sha256 of this list:
+
+```
+llamafile/oc-audiocpp/audiocpp.h 6ad0b55d5f28adc9ad60a45bf9a7caa428c4f5441389a6932d6670a8e2a90d73
+llamafile/oc-codec/oc_codec_sidecar.h f2c9001d1414aa2a55533b867b79c52a52c682558e0234c5bebd30bf97b91005
+```
+
+ABI `loader` = `d6ff34022f1d3fb5af56ca0c80e3f4e552bf09f1bebdf0cb5f314a454bf38dbc` (exported) — sha256 of this list:
+
+```
+.cosmocc/4.0.2/bin/ape-m1.c 78af79e20abbb3f99a97355959b59bd58949564064adb754f0302ebabb4bf7a3
+```
+
+**c13 r3 dev snapshot — the development build of the c13 r3 release, for integration testing.** Engine 2610160712001 (`opencoti-0.10.5-c13`, dev channel) is built from the same source as the c13 r3 release engine 2610160746001 (chain 0001–0688; identical by content). The libraries are the r3 release index's: the Windows CUDA DLL is new (68cd73b1, flash attention with mixed K / V types on the GPU, bug-4090), the rest are the c13 bytes; the `ggml` / `media` / `loader` digests are c13's. This engine itself ran the glibc 2.28 route gate (Vulkan + CUDA) and the Whistle gate on bs2 166 / 166 (CUDA + CPU arms); the scoped regression gates of the release ran on 2610160746001 (see the release notes).
+
+What is new in r3 (chain 0666–0688):
+
+- **The CPU at stock llama.cpp speed** (0666–0675, 0678, 0685–0687, c13 row 41.1). Measured on the release bytes (2610160746001) against stock llama.cpp b11521, Qwen3.6-35B-A3B UD-IQ3_XXS on the CPU: EPYC 9575F 6 threads **110.2 %** (r2 92.3 %), 1 thread 107.4 % (r2 88.3 %); Mac mini (Apple M-series, Metal off) 6 threads 99.7 % (r2 81.7 %), 4 threads 97.9 % (r2 77.4 %); Ryzen 5 5600G 6 threads 102.2 % (r2 96.1 %, a noisy host); Raspberry Pi 5 (Qwen2.5-1.5B, 4 threads) 104.2 % (r2 97.3 %). The causes, each fixed with stock's own code path: cosmopolitan's `memcpy` copied backward (a forward copy replaces it); flash attention runs on the CPU as in stock (llamafile's layer turned it off); stock's tiled int8 matmul for K / IQ quant prompts; stock's Q8_0 / Q4_0 prompt kernel (with an AVX-VNNI copy); f16 V rows converted at the variant's ISA; the CPU work buffer 64-byte aligned; F16 / BF16 prompt matmuls reach stock's kernel; ARM kernels built with clang and tuned for Apple cores; IQ2_S matrix × vector on ARM through stock's kernel; no OpenMP (ggml's own threadpool, as stock's portable builds); a CPU-only engine samples on the host; on macOS a recurrent state on the CPU is f32. `--recurrent-state-type {bf16,f16,f32}` (0680).
+- **CPUSet core selection** (0671, row 41.2; a port of the owner's llama.cpp PR #6832): threads are pinned, the default pick is physical cores, no efficiency cores, one last level cache (the prompt pool gets every last level cache); Windows CPU sets across processor groups (no 64-core limit). `--cpuset on|off`, `-bco`, `-llct`, `-atc`, `--cpuset-skip-core0`. Thread placement had never run in an opencoti build (bug-4080). i5-14600K: 115.2 % of stock native, 9950X3D 103.7 %.
+- **cpp-httplib 0.60.1** (0676, 0688, row 41.3): with stock's build definitions — a request body over 8 KB without a JSON content type is accepted (HTTP 413 before, bug-4082); a realtime websocket message that arrives in pieces no longer ends the session (bug-4105).
+- **Generation notices, contract v2** (0681, 0682, 0684, row 41.4; feature `generation_notice_v2`): the request's `notices` object (steps by `at_percent` or `left`, templates with `{left}` / `{total}`), `--notice-templates FILE`, `--print-notice-config`, `/props` `generation_notice` (the templates and the spans a client hides), `timings.notices_planted`, `POST /v1/notice {key, kind: "message_waiting"}` (through the router by key). A notice is planted only at a prose boundary, never inside code, and only while 200 or more tokens of its phase are left.
+- **`num_ctx_hold`** (0683, row 41.6; feature `elastic_window_unheld_v1`): `false` sizes the window from `num_ctx` / `num_ctx_min` and releases it at the end of the request, with a `session_id` too.
+- **Tool calls keep Windows path backslashes** (0679, bug-4092).
+- **The Windows CUDA library runs flash attention with mixed K / V types on the GPU** (bug-4090): every Windows DLL since c10 r3 was built without `GGML_CUDA_FA_ALL_QUANTS`, so an f16-K / q8_0-V cache ran attention on the CPU. New cuda component 2610160820001.

components/engine/2610160712001/opencoti-0.10.5-c13-2610160712001

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b81b2952cccae11d3cd16331f49ee7628ef6c89761780d41cfc2ff0481d20e38
+size 145615678

components/engine/2610160712001/opencoti-0.10.5-c13-2610160712001.exe

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b81b2952cccae11d3cd16331f49ee7628ef6c89761780d41cfc2ff0481d20e38
+size 145615678

components/vulkan/2610141455004/BUILD_INFO.vulkan.md

@@ -2,7 +2,7 @@
 
 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 2610141212001, 2610141255001.
+Built from the engine source of build 2610141255001; gated floor (engine-min) 2610141212001; gated with 2610141255001, 2610141633001, 2610160746001, 2610160712001.
 
 | file | platform | kind | sha256 | bytes |
 |---|---|---|---|---|