From 3b24d48cf3422271aecc03bd21895713da91519a Mon Sep 17 00:00:00 2001
From: abetlen <abetlen@gmail.com>
Date: Mon, 21 Sep 2026 12:43:59 -0700
Subject: [PATCH] feat: update llama.cpp to fb34fc262

---
 CHANGELOG.md           |  2 +-
 llama_cpp/llama_cpp.py | 31 ++++++++++++++++++++++++++++---
 vendor/llama.cpp       |  2 +-
 3 files changed, 30 insertions(+), 5 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 6e13a4b5c..be1130f95 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,7 +7,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
-- feat: update llama.cpp to ggml-org/llama.cpp@v0.4.0
+- feat: update llama.cpp to ggml-org/llama.cpp@fb34fc262
 
 ## [0.3.35]
 
diff --git a/llama_cpp/llama_cpp.py b/llama_cpp/llama_cpp.py
index 4aa9c6ec4..c7af8b9a7 100644
--- a/llama_cpp/llama_cpp.py
+++ b/llama_cpp/llama_cpp.py
@@ -213,6 +213,7 @@ def _warn_deprecated(symbol: str, hint: str) -> None:
 #     LLAMA_VOCAB_TYPE_UGM    = 4, // T5 tokenizer based on Unigram
 #     LLAMA_VOCAB_TYPE_RWKV   = 5, // RWKV tokenizer based on greedy tokenization
 #     LLAMA_VOCAB_TYPE_PLAMO2 = 6, // PLaMo-2 tokenizer based on Aho-Corasick with dynamic programming
+#     LLAMA_VOCAB_TYPE_TEST   = 7, // Dummy tokenizer for testing: rolling hash of fixed-size chunks -> tokens, tokens -> hex
 # };
 LLAMA_VOCAB_TYPE_NONE = 0
 """For models without vocab"""
@@ -228,6 +229,8 @@ def _warn_deprecated(symbol: str, hint: str) -> None:
 """RWKV tokenizer based on greedy tokenization"""
 LLAMA_VOCAB_TYPE_PLAMO2 = 6
 """PLaMo-2 tokenizer based on Aho-Corasick with dynamic programming"""
+LLAMA_VOCAB_TYPE_TEST = 7
+"""Dummy tokenizer for testing: rolling hash of fixed-size chunks -> tokens, tokens -> hex"""
 
 
 # NOTE: Deprecated and will be removed in the future. (already gone in llama.cpp)
@@ -1476,6 +1479,8 @@ def llama_model_load_from_splits(
 
 
 # // Load a model from an open FILE pointer
+# // The GGUF is read from the current position, so it can be embedded in a larger file
+# // mmap needs the GGUF data section at a file offset to be aligned to the CPU tensor alignment (32 bytes)
 # LLAMA_API struct llama_model * llama_model_load_from_file_ptr(
 #                                FILE * file,
 #           struct llama_model_params   params);
@@ -1487,7 +1492,11 @@ def llama_model_load_from_splits(
 def llama_model_load_from_file_ptr(
     file: ctypes.c_void_p, params: llama_model_params, /
 ) -> Optional[llama_model_p]:
-    """Load a model from an open FILE pointer."""
+    """Load a model from an open FILE pointer
+
+    The GGUF is read from the current position, so it can be embedded in a larger file
+    mmap needs the GGUF data section at a file offset to be aligned to the CPU tensor alignment (32 bytes)
+    """
     ...
 
 
@@ -2115,6 +2124,22 @@ def llama_adapter_lora_init(
 ) -> Optional[llama_adapter_lora_p]: ...
 
 
+# // Load a LoRA adapter from an open FILE pointer, reading from its current position
+# LLAMA_API struct llama_adapter_lora * llama_adapter_lora_init_from_file_ptr(
+#         struct llama_model * model,
+#         FILE * file);
+@ctypes_function(
+    "llama_adapter_lora_init_from_file_ptr",
+    [llama_model_p_ctypes, ctypes.c_void_p],
+    llama_adapter_lora_p_ctypes,
+)
+def llama_adapter_lora_init_from_file_ptr(
+    model: llama_model_p, file: ctypes.c_void_p, /
+) -> Optional[llama_adapter_lora_p]:
+    """Load a LoRA adapter from an open FILE pointer, reading from its current position"""
+    ...
+
+
 # // Get metadata value as a string by key name
 # LLAMA_API int32_t llama_adapter_meta_val_str(const struct llama_adapter_lora * adapter, const char * key, char * buf, size_t buf_size);
 @ctypes_function(
@@ -4519,11 +4544,11 @@ def llama_sampler_chain_get(
 ) -> llama_sampler_p: ...
 
 
-# LLAMA_API int                    llama_sampler_chain_n  (const struct llama_sampler * chain);
+# LLAMA_API int32_t                llama_sampler_chain_n  (const struct llama_sampler * chain);
 @ctypes_function(
     "llama_sampler_chain_n",
     [llama_sampler_p_ctypes],
-    ctypes.c_int,
+    ctypes.c_int32,
 )
 def llama_sampler_chain_n(chain: llama_sampler_p, /) -> int: ...
 
diff --git a/vendor/llama.cpp b/vendor/llama.cpp
index 5266f24da..fb34fc262 160000
--- a/vendor/llama.cpp
+++ b/vendor/llama.cpp
@@ -1 +1 @@
-Subproject commit 5266f24da75dc449bd56cbed7addb9c8e4a6a73e
+Subproject commit fb34fc262c1b43f1832c7472429fb2247d650493