Refresh llama.cpp upstream canary patch queue

third_party/llama.cpp/patches/0001-Draft-skippy-ABI-header.patch

@@ -1,7 +1,7 @@
-From 550ab15ef7e48dc913ac6dc9df75547917308dfc Mon Sep 17 00:00:00 2001
+From aaa1f7247b3ff779f321fa76ccc12b66f212d5bd Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 10:19:37 +1000
-Subject: [PATCH 01/81] Draft skippy ABI header
+Subject: [PATCH 01/82] Draft skippy ABI header
 
 ---
  include/skippy.h | 241 +++++++++++++++++++++++++++++++++++++++++++++++
@@ -10,7 +10,7 @@ Subject: [PATCH 01/81] Draft skippy ABI header
 
 diff --git a/include/skippy.h b/include/skippy.h
 new file mode 100644
-index 00000000..b0f3d227
+index 000000000..b0f3d227f
 --- /dev/null
 +++ b/include/skippy.h
 @@ -0,0 +1,241 @@
@@ -256,5 +256,5 @@ index 00000000..b0f3d227
 +
 +#endif // SKIPPY_ABI_H
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)
 

third_party/llama.cpp/patches/0002-Implement-skippy-model-info-ABI.patch

@@ -1,7 +1,7 @@
-From 9ebca2cf952918f3114fc9b332febc84660f9f73 Mon Sep 17 00:00:00 2001
+From 3581accd09ae328190aaaadb70ebc0521c24c365 Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 10:28:54 +1000
-Subject: [PATCH 02/81] Implement skippy model info ABI
+Subject: [PATCH 02/82] Implement skippy model info ABI
 
 ---
  CMakeLists.txt     |   1 +
@@ -11,10 +11,10 @@ Subject: [PATCH 02/81] Implement skippy model info ABI
  create mode 100644 src/skippy.cpp
 
 diff --git a/CMakeLists.txt b/CMakeLists.txt
-index 4f7f4eca..a08a32fb 100644
+index 9e7b1253c..d1e6b4e6d 100644
 --- a/CMakeLists.txt
 +++ b/CMakeLists.txt
-@@ -258,6 +258,7 @@ set(LLAMA_BIN_INSTALL_DIR     ${CMAKE_INSTALL_BINDIR}     CACHE PATH "Location o
+@@ -235,6 +235,7 @@ set(LLAMA_BIN_INSTALL_DIR     ${CMAKE_INSTALL_BINDIR}     CACHE PATH "Location o
 
  set(LLAMA_PUBLIC_HEADERS
      ${CMAKE_CURRENT_SOURCE_DIR}/include/llama.h
@@ -23,10 +23,10 @@ index 4f7f4eca..a08a32fb 100644
 
  set_target_properties(llama
 diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
-index 7b1fcfca..8f3aa8c1 100644
+index d15ccfd99..2526f45cd 100644
 --- a/src/CMakeLists.txt
 +++ b/src/CMakeLists.txt
-@@ -34,6 +34,7 @@ add_library(llama
+@@ -35,6 +35,7 @@ add_library(llama
              llama-model.cpp
              llama-quant.cpp
              llama-sampler.cpp
@@ -36,7 +36,7 @@ index 7b1fcfca..8f3aa8c1 100644
              unicode.cpp
 diff --git a/src/skippy.cpp b/src/skippy.cpp
 new file mode 100644
-index 00000000..bd94fa37
+index 000000000..bd94fa37f
 --- /dev/null
 +++ b/src/skippy.cpp
 @@ -0,0 +1,219 @@
@@ -260,5 +260,5 @@ index 00000000..bd94fa37
 +
 +}
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)
 

third_party/llama.cpp/patches/0003-Implement-single-stage-runtime-ABI-baseline.patch

@@ -1,14 +1,14 @@
-From 3373ba0729f0aa133194bc7f59e640e2de537668 Mon Sep 17 00:00:00 2001
+From 7f889e8044c192797c0eb9f4dfbf61b8ede502b5 Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 11:04:49 +1000
-Subject: [PATCH 03/81] Implement single-stage runtime ABI baseline
+Subject: [PATCH 03/82] Implement single-stage runtime ABI baseline
 
 ---
  src/skippy.cpp | 323 ++++++++++++++++++++++++++++++++++++++++++++++++-
  1 file changed, 322 insertions(+), 1 deletion(-)
 
 diff --git a/src/skippy.cpp b/src/skippy.cpp
-index bd94fa37..4feca3b8 100644
+index bd94fa37f..4feca3b8f 100644
 --- a/src/skippy.cpp
 +++ b/src/skippy.cpp
 @@ -2,14 +2,27 @@
@@ -371,5 +371,5 @@ index bd94fa37..4feca3b8 100644
          const char * path,
          struct skippy_model_info ** out_info,
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)
 

third_party/llama.cpp/patches/0004-Implement-runtime-slice-tensor-filtering.patch

@@ -1,7 +1,7 @@
-From 2873f2d485da528de4c389ae3ec2260f29b83065 Mon Sep 17 00:00:00 2001
+From 655e0e84af1fd7c60e12a0d25b8bbe5bf8e3bddd Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 11:57:45 +1000
-Subject: [PATCH 04/81] Implement runtime-slice tensor filtering
+Subject: [PATCH 04/82] Implement runtime-slice tensor filtering
 
 ---
  src/llama-model-loader.cpp | 44 ++++++++++++++++++++++++++++++++++++++
@@ -10,7 +10,7 @@ Subject: [PATCH 04/81] Implement runtime-slice tensor filtering
  3 files changed, 95 insertions(+), 1 deletion(-)
 
 diff --git a/src/llama-model-loader.cpp b/src/llama-model-loader.cpp
-index c645d078..7782c250 100644
+index 474cabdfc..869092306 100644
 --- a/src/llama-model-loader.cpp
 +++ b/src/llama-model-loader.cpp
 @@ -17,6 +17,16 @@ static const size_t kiB = 1024;
@@ -30,7 +30,7 @@ index c645d078..7782c250 100644
  const char * llama_file_version_name(llama_fver version) {
      switch (version) {
          case GGUF_FILE_VERSION_V1: return "GGUF V1 (support until nov 2023)";
-@@ -1098,6 +1108,40 @@ struct ggml_tensor * llama_model_loader::create_tensor(
+@@ -1100,6 +1110,40 @@ struct ggml_tensor * llama_model_loader::create_tensor(
              throw std::runtime_error(format("missing tensor info mapping for %s", tn.str().c_str()));
          }
 
@@ -72,7 +72,7 @@ index c645d078..7782c250 100644
          if (info.op == GGML_OP_NONE || (flags & TENSOR_SKIP)) {
              const size_t nbytes = ggml_nbytes(t_meta);
 diff --git a/src/llama-model-loader.h b/src/llama-model-loader.h
-index c476026d..3fb84fdd 100644
+index c476026d3..3fb84fdd2 100644
 --- a/src/llama-model-loader.h
 +++ b/src/llama-model-loader.h
 @@ -20,6 +20,17 @@ using llama_buf_map = std::unordered_map<uint32_t, ggml_backend_buffer_t>;
@@ -94,7 +94,7 @@ index c476026d..3fb84fdd 100644
      GGUF_FILE_VERSION_V1 = 1,
      GGUF_FILE_VERSION_V2 = 2,
 diff --git a/src/skippy.cpp b/src/skippy.cpp
-index 4feca3b8..10876ee5 100644
+index 4feca3b8f..10876ee50 100644
 --- a/src/skippy.cpp
 +++ b/src/skippy.cpp
 @@ -1,6 +1,7 @@
@@ -191,5 +191,5 @@ index 4feca3b8..10876ee5 100644
      *out_session = nullptr;
 
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)
 

third_party/llama.cpp/patches/0005-Add-activation-frame-runtime-ABI.patch

@@ -1,15 +1,15 @@
-From 82245852930b1698669cfb15bf0f5609353bf9c3 Mon Sep 17 00:00:00 2001
+From 1da4c2df3ba0ea3ca20f9671ab173ba21c8c67dd Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 12:15:31 +1000
-Subject: [PATCH 05/81] Add activation frame runtime ABI
+Subject: [PATCH 05/82] Add activation frame runtime ABI
 
 ---
  include/skippy.h |  50 +++++++++++++++++++++
  src/skippy.cpp   | 112 ++++++++++++++++++++++++++++++++++++++++++++++-
  2 files changed, 161 insertions(+), 1 deletion(-)
 
 diff --git a/include/skippy.h b/include/skippy.h
-index b0f3d227..cd72379d 100644
+index b0f3d227f..cd72379df 100644
 --- a/include/skippy.h
 +++ b/include/skippy.h
 @@ -36,6 +36,7 @@ enum skippy_feature {
@@ -91,7 +91,7 @@ index b0f3d227..cd72379d 100644
          struct skippy_session * session,
          int32_t layer_start,
 diff --git a/src/skippy.cpp b/src/skippy.cpp
-index 10876ee5..4aef362a 100644
+index 10876ee50..4aef362ab 100644
 --- a/src/skippy.cpp
 +++ b/src/skippy.cpp
 @@ -187,6 +187,44 @@ static llama_token skippy_greedy_sample(skippy_session * session) {
@@ -228,5 +228,5 @@ index 10876ee5..4aef362a 100644
          struct skippy_session * session,
          int32_t layer_start,
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)
 

third_party/llama.cpp/patches/0006-Execute-runtime-slice-activation-frames.patch

@@ -1,7 +1,7 @@
-From fab9a753c08883775e7a3a53784bbfeb8c8dcaba Mon Sep 17 00:00:00 2001
+From ad8f9af6588f8f1482089b9bbd0ab71ae015363d Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 12:33:30 +1000
-Subject: [PATCH 06/81] Execute runtime-slice activation frames
+Subject: [PATCH 06/82] Execute runtime-slice activation frames
 
 ---
  src/llama-graph.cpp  |  52 ++++---
@@ -11,10 +11,10 @@ Subject: [PATCH 06/81] Execute runtime-slice activation frames
  4 files changed, 365 insertions(+), 53 deletions(-)
 
 diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp
-index fc027de8..2e7ae884 100644
+index 7468bd9b7..3f6342b2a 100644
 --- a/src/llama-graph.cpp
 +++ b/src/llama-graph.cpp
-@@ -18,6 +18,20 @@
+@@ -19,6 +19,20 @@
  #include <sstream>
  #include <unordered_set>
 
@@ -35,7 +35,7 @@ index fc027de8..2e7ae884 100644
  // dedup helpers
 
  static ggml_tensor * build_attn_inp_kq_mask(
-@@ -1747,28 +1761,32 @@ ggml_tensor * llm_graph_context::build_inp_embd(ggml_tensor * tok_embd) const {
+@@ -1842,28 +1856,32 @@ ggml_tensor * llm_graph_context::build_inp_embd(ggml_tensor * tok_embd) const {
      {
          auto & cur = inps[0];
 
@@ -86,10 +86,10 @@ index fc027de8..2e7ae884 100644
      }
 
 diff --git a/src/llama-graph.h b/src/llama-graph.h
-index bf677823..f01cbabd 100644
+index cc5cfe51d..84b7124a1 100644
 --- a/src/llama-graph.h
 +++ b/src/llama-graph.h
-@@ -27,6 +27,18 @@ class llama_memory_recurrent_context;
+@@ -28,6 +28,18 @@ class llama_memory_recurrent_context;
  class llama_memory_hybrid_context;
  class llama_memory_hybrid_iswa_context;
 
@@ -109,7 +109,7 @@ index bf677823..f01cbabd 100644
  enum llm_graph_type {
      LLM_GRAPH_TYPE_DEFAULT,
 diff --git a/src/models/llama.cpp b/src/models/llama.cpp
-index cef66d05..581826b0 100644
+index 4bfebc884..23864f0dd 100644
 --- a/src/models/llama.cpp
 +++ b/src/models/llama.cpp
 @@ -105,7 +105,12 @@ llama_model_llama::graph<embed>::graph(const llama_model & model, const llm_grap
@@ -135,10 +135,10 @@ index cef66d05..581826b0 100644
 
 -    for (int il = 0; il < n_layer; ++il) {
 +    for (int il = il_start; il < il_end; ++il) {
-         ggml_tensor * inpSA = inpL;
+         res->t_layer_inp[il] = inpL;
 
-         // norm
-@@ -226,6 +231,13 @@ llama_model_llama::graph<embed>::graph(const llama_model & model, const llm_grap
+         ggml_tensor * inpSA = inpL;
+@@ -228,6 +233,13 @@ llama_model_llama::graph<embed>::graph(const llama_model & model, const llm_grap
      }
      cur = inpL;
 
@@ -153,7 +153,7 @@ index cef66d05..581826b0 100644
              model.output_norm, NULL,
              LLM_NORM_RMS, -1);
 diff --git a/src/skippy.cpp b/src/skippy.cpp
-index 4aef362a..732a1a15 100644
+index 4aef362ab..732a1a15c 100644
 --- a/src/skippy.cpp
 +++ b/src/skippy.cpp
 @@ -1,6 +1,8 @@
@@ -612,5 +612,5 @@ index 4aef362a..732a1a15 100644
 
  enum skippy_status skippy_export_state(
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)
 

third_party/llama.cpp/patches/0007-Implement-GGUF-slice-writer-ABI.patch

@@ -1,14 +1,14 @@
-From 9f7d122a97d176fa31253dd8b5ce84764c234319 Mon Sep 17 00:00:00 2001
+From 47841b44bdf2dbffc25dd15b70ffcc93a1a6364c Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 14:39:10 +1000
-Subject: [PATCH 07/81] Implement GGUF slice writer ABI
+Subject: [PATCH 07/82] Implement GGUF slice writer ABI
 
 ---
  src/skippy.cpp | 413 +++++++++++++++++++++++++++++++++++++++++++++++++
  1 file changed, 413 insertions(+)
 
 diff --git a/src/skippy.cpp b/src/skippy.cpp
-index 732a1a15..2b54a7db 100644
+index 732a1a15c..2b54a7db7 100644
 --- a/src/skippy.cpp
 +++ b/src/skippy.cpp
 @@ -11,9 +11,12 @@
@@ -470,5 +470,5 @@ index 732a1a15..2b54a7db 100644
 +
  }
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)
 

third_party/llama.cpp/patches/0008-Add-GGUF-package-part-composer-ABI.patch

@@ -1,15 +1,15 @@
-From 991a7982943373f01a161b959cad9816201a12a5 Mon Sep 17 00:00:00 2001
+From 3daa6935328e8d61c23a244be7c0327733339005 Mon Sep 17 00:00:00 2001
 From: James Dumay <jameswdumay@gmail.com>
 Date: Sat, 25 Apr 2026 14:51:18 +1000
-Subject: [PATCH 08/81] Add GGUF package part composer ABI
+Subject: [PATCH 08/82] Add GGUF package part composer ABI
 
 ---
  include/skippy.h |   6 +
  src/skippy.cpp   | 318 +++++++++++++++++++++++++++++++++--------------
  2 files changed, 230 insertions(+), 94 deletions(-)
 
 diff --git a/include/skippy.h b/include/skippy.h
-index cd72379d..7ab93d35 100644
+index cd72379df..7ab93d353 100644
 --- a/include/skippy.h
 +++ b/include/skippy.h
 @@ -284,6 +284,12 @@ LLAMA_API enum skippy_status skippy_write_slice_gguf(
@@ -26,7 +26,7 @@ index cd72379d..7ab93d35 100644
  }
  #endif
 diff --git a/src/skippy.cpp b/src/skippy.cpp
-index 2b54a7db..26169320 100644
+index 2b54a7db7..261693203 100644
 --- a/src/skippy.cpp
 +++ b/src/skippy.cpp
 @@ -14,6 +14,7 @@
@@ -434,5 +434,5 @@ index 2b54a7db..26169320 100644
 
      return skippy_success(out_error);
 -- 
-2.50.1 (Apple Git-155)
+2.54.0 (Apple Git-156)