@fugood/llama.node 1.4.11 → 1.4.12

package/src/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h

@@ -14,10 +14,6 @@
 #include <arm_neon.h>
 #endif
 
-#if defined(__F16C__)
-#include <immintrin.h>
-#endif
-
 #if defined(__riscv_v_intrinsic)
 #include <riscv_vector.h>
 #endif

package/src/llama.cpp/include/llama.h

@@ -286,7 +286,7 @@ extern "C" {
         // NULL-terminated list of buffer types to use for tensors that match a pattern
         const struct llama_model_tensor_buft_override * tensor_buft_overrides;
 
-        int32_t n_gpu_layers; // number of layers to store in VRAM
+        int32_t n_gpu_layers; // number of layers to store in VRAM, a negative value means all layers
         enum llama_split_mode split_mode; // how to split the model across multiple GPUs
 
         // the GPU that is used for the entire model when split_mode is LLAMA_SPLIT_MODE_NONE
@@ -467,10 +467,17 @@ extern "C" {
     // Frees all allocated memory
     LLAMA_API void llama_free(struct llama_context * ctx);
 
+    enum llama_params_fit_status {
+        LLAMA_PARAMS_FIT_STATUS_SUCCESS = 0, // found allocations that are projected to fit
+        LLAMA_PARAMS_FIT_STATUS_FAILURE = 1, // could not find allocations that are projected to fit
+        LLAMA_PARAMS_FIT_STATUS_ERROR   = 2, // a hard error occured, e.g. because no model could be found at the specified path
+    };
+
     // fits mparams and cparams to free device memory (assumes system memory is unlimited)
-    // returns true if the parameters could be successfully modified to fit device memory
-    // this function is NOT thread safe because it modifies the global llama logger state
-    LLAMA_API bool llama_params_fit(
+    //   - returns true if the parameters could be successfully modified to fit device memory
+    //   - this function is NOT thread safe because it modifies the global llama logger state
+    //   - only parameters that have the same value as in llama_default_model_params are modified
+    LLAMA_API enum llama_params_fit_status llama_params_fit(
                                    const char   * path_model,
                     struct llama_model_params   * mparams,
                     struct llama_context_params * cparams,
@@ -600,6 +607,8 @@ extern "C" {
     //
 
     // Load a LoRA adapter from file
+    // The adapter is valid as long as the associated model is not freed
+    // All adapters must be loaded before context creation
     LLAMA_API struct llama_adapter_lora * llama_adapter_lora_init(
             struct llama_model * model,
             const char * path_lora);

package/src/llama.cpp/src/CMakeLists.txt

@@ -87,9 +87,12 @@ add_library(llama
             models/llada.cpp
             models/llama-iswa.cpp
             models/llama.cpp
+            models/maincoder.cpp
             models/mamba.cpp
+            models/mimo2-iswa.cpp
             models/minicpm3.cpp
             models/minimax-m2.cpp
+            models/modern-bert.cpp
             models/mpt.cpp
             models/nemotron-h.cpp
             models/nemotron.cpp
@@ -105,6 +108,7 @@ add_library(llama
             models/phi3.cpp
             models/plamo.cpp
             models/plamo2.cpp
+            models/plamo3.cpp
             models/plm.cpp
             models/qwen.cpp
             models/qwen2.cpp

package/src/llama.cpp/src/llama-adapter.cpp

@@ -146,9 +146,11 @@ llama_adapter_lora_weight * llama_adapter_lora::get_weight(ggml_tensor * w) {
     return nullptr;
 }
 
-static void llama_adapter_lora_init_impl(llama_model & model, const char * path_lora, llama_adapter_lora & adapter) {
+static void llama_adapter_lora_init_impl(const char * path_lora, llama_adapter_lora & adapter) {
     LLAMA_LOG_INFO("%s: loading lora adapter from '%s' ...\n", __func__, path_lora);
 
+    llama_model & model = adapter.model;
+
     ggml_context * ctx_init;
     gguf_init_params meta_gguf_params = {
         /* .no_alloc = */ true,
@@ -411,14 +413,17 @@ static void llama_adapter_lora_init_impl(llama_model & model, const char * path_
         }
     }
 
+    // update number of nodes used
+    model.n_lora_nodes += adapter.get_n_nodes();
+
     LLAMA_LOG_INFO("%s: loaded %zu tensors from lora file\n", __func__, adapter.ab_map.size()*2);
 }
 
 llama_adapter_lora * llama_adapter_lora_init(llama_model * model, const char * path_lora) {
-    llama_adapter_lora * adapter = new llama_adapter_lora();
+    llama_adapter_lora * adapter = new llama_adapter_lora(*model);
 
     try {
-        llama_adapter_lora_init_impl(*model, path_lora, *adapter);
+        llama_adapter_lora_init_impl(path_lora, *adapter);
         return adapter;
     } catch (const std::exception & err) {
         LLAMA_LOG_ERROR("%s: failed to apply lora adapter: %s\n", __func__, err.what());
@@ -469,6 +474,10 @@ int32_t llama_adapter_meta_val_str_by_index(const llama_adapter_lora * adapter,
 }
 
 void llama_adapter_lora_free(llama_adapter_lora * adapter) {
+    // update number of nodes used
+    GGML_ASSERT(adapter->model.n_lora_nodes >= adapter->get_n_nodes());
+    adapter->model.n_lora_nodes -= adapter->get_n_nodes();
+
     delete adapter;
 }
 

package/src/llama.cpp/src/llama-adapter.h

@@ -59,6 +59,8 @@ struct llama_adapter_lora_weight {
 };
 
 struct llama_adapter_lora {
+    llama_model & model;
+
     // map tensor name to lora_a_b
     std::unordered_map<std::string, llama_adapter_lora_weight> ab_map;
 
@@ -73,10 +75,14 @@ struct llama_adapter_lora {
     // activated lora (aLoRA)
     std::vector<llama_token> alora_invocation_tokens;
 
-    llama_adapter_lora() = default;
+    llama_adapter_lora(llama_model & model) : model(model) {}
     ~llama_adapter_lora() = default;
 
     llama_adapter_lora_weight * get_weight(ggml_tensor * w);
+
+    uint32_t get_n_nodes() const {
+        return ab_map.size() * 6u; // a, b, scale, add, 2 x mul_mat
+    }
 };
 
 using llama_adapter_loras = std::unordered_map<llama_adapter_lora *, float>;

package/src/llama.cpp/src/llama-arch.cpp

@@ -20,6 +20,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_STARCODER,        "starcoder"        },
     { LLM_ARCH_REFACT,           "refact"           },
     { LLM_ARCH_BERT,             "bert"             },
+    { LLM_ARCH_MODERN_BERT,      "modern-bert"      },
     { LLM_ARCH_NOMIC_BERT,       "nomic-bert"       },
     { LLM_ARCH_NOMIC_BERT_MOE,   "nomic-bert-moe"   },
     { LLM_ARCH_NEO_BERT,         "neo-bert"         },
@@ -41,6 +42,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_PHIMOE,           "phimoe"           },
     { LLM_ARCH_PLAMO,            "plamo"            },
     { LLM_ARCH_PLAMO2,           "plamo2"           },
+    { LLM_ARCH_PLAMO3,           "plamo3"           },
     { LLM_ARCH_CODESHELL,        "codeshell"        },
     { LLM_ARCH_ORION,            "orion"            },
     { LLM_ARCH_INTERNLM2,        "internlm2"        },
@@ -114,6 +116,9 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_RND1,             "rnd1"             },
     { LLM_ARCH_PANGU_EMBED,      "pangu-embedded"   },
     { LLM_ARCH_MISTRAL3,         "mistral3"         },
+    { LLM_ARCH_MIMO2,            "mimo2"           },
+    { LLM_ARCH_LLAMA_EMBED,      "llama-embed"      },
+    { LLM_ARCH_MAINCODER,        "maincoder"        },
     { LLM_ARCH_UNKNOWN,          "(unknown)"        },
 };
 
@@ -204,6 +209,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_ATTENTION_GATE_LORA_RANK,               "%s.attention.gate_lora_rank"               },
     { LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,       "%s.attention.relative_buckets_count"       },
     { LLM_KV_ATTENTION_SLIDING_WINDOW,               "%s.attention.sliding_window"               },
+    { LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN,       "%s.attention.sliding_window_pattern"       },
     { LLM_KV_ATTENTION_SCALE,                        "%s.attention.scale"                        },
     { LLM_KV_ATTENTION_OUTPUT_SCALE,                 "%s.attention.output_scale"                 },
     { LLM_KV_ATTENTION_TEMPERATURE_LENGTH,           "%s.attention.temperature_length"           },
@@ -214,6 +220,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_ROPE_DIMENSION_COUNT,          "%s.rope.dimension_count"                 },
     { LLM_KV_ROPE_DIMENSION_SECTIONS,       "%s.rope.dimension_sections"              },
     { LLM_KV_ROPE_FREQ_BASE,                "%s.rope.freq_base"                       },
+    { LLM_KV_ROPE_FREQ_BASE_SWA,            "%s.rope.freq_base_swa"                   },
     { LLM_KV_ROPE_SCALE_LINEAR,             "%s.rope.scale_linear"                    },
     { LLM_KV_ROPE_SCALING_TYPE,             "%s.rope.scaling.type"                    },
     { LLM_KV_ROPE_SCALING_FACTOR,           "%s.rope.scaling.factor"                  },
@@ -497,6 +504,7 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
         case LLM_ARCH_LLAMA:
         case LLM_ARCH_DECI:
         case LLM_ARCH_MISTRAL3:
+        case LLM_ARCH_LLAMA_EMBED:
             return {
                 LLM_TENSOR_TOKEN_EMBD,
                 LLM_TENSOR_OUTPUT_NORM,
@@ -778,6 +786,20 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
                 LLM_TENSOR_CLS,
                 LLM_TENSOR_CLS_OUT,
             };
+        case LLM_ARCH_MODERN_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_CLS,
+                LLM_TENSOR_CLS_OUT,
+            };
         case LLM_ARCH_JINA_BERT_V2:
             return {
                 LLM_TENSOR_TOKEN_EMBD,
@@ -1057,6 +1079,22 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
                 LLM_TENSOR_ATTN_POST_NORM,
                 LLM_TENSOR_FFN_POST_NORM,
             };
+        case LLM_ARCH_PLAMO3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
         case LLM_ARCH_CODESHELL:
             return {
                 LLM_TENSOR_TOKEN_EMBD,
@@ -2171,11 +2209,49 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
                 LLM_TENSOR_VISEXP_FFN_DOWN,
                 LLM_TENSOR_VISEXP_FFN_UP,
             };
+        case LLM_ARCH_MIMO2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_SINKS,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
         case LLM_ARCH_GPTJ:
         case LLM_ARCH_UNKNOWN:
             return {
                 LLM_TENSOR_TOKEN_EMBD,
             };
+        case LLM_ARCH_MAINCODER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
         default:
             GGML_ABORT("unknown architecture for tensor mapping");
     }

package/src/llama.cpp/src/llama-arch.h

@@ -24,6 +24,7 @@ enum llm_arch {
     LLM_ARCH_STARCODER,
     LLM_ARCH_REFACT,
     LLM_ARCH_BERT,
+    LLM_ARCH_MODERN_BERT,
     LLM_ARCH_NOMIC_BERT,
     LLM_ARCH_NOMIC_BERT_MOE,
     LLM_ARCH_NEO_BERT,
@@ -45,6 +46,7 @@ enum llm_arch {
     LLM_ARCH_PHIMOE,
     LLM_ARCH_PLAMO,
     LLM_ARCH_PLAMO2,
+    LLM_ARCH_PLAMO3,
     LLM_ARCH_CODESHELL,
     LLM_ARCH_ORION,
     LLM_ARCH_INTERNLM2,
@@ -118,6 +120,9 @@ enum llm_arch {
     LLM_ARCH_RND1,
     LLM_ARCH_PANGU_EMBED,
     LLM_ARCH_MISTRAL3,
+    LLM_ARCH_MIMO2,
+    LLM_ARCH_LLAMA_EMBED,
+    LLM_ARCH_MAINCODER,
     LLM_ARCH_UNKNOWN,
 };
 
@@ -208,6 +213,7 @@ enum llm_kv {
     LLM_KV_ATTENTION_GATE_LORA_RANK,
     LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,
     LLM_KV_ATTENTION_SLIDING_WINDOW,
+    LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN,
     LLM_KV_ATTENTION_SCALE,
     LLM_KV_ATTENTION_OUTPUT_SCALE,
     LLM_KV_ATTENTION_TEMPERATURE_LENGTH,
@@ -218,6 +224,7 @@ enum llm_kv {
     LLM_KV_ROPE_DIMENSION_COUNT,
     LLM_KV_ROPE_DIMENSION_SECTIONS,
     LLM_KV_ROPE_FREQ_BASE,
+    LLM_KV_ROPE_FREQ_BASE_SWA,
     LLM_KV_ROPE_SCALE_LINEAR,
     LLM_KV_ROPE_SCALING_TYPE,
     LLM_KV_ROPE_SCALING_FACTOR,

package/src/llama.cpp/src/llama-chat.cpp

@@ -74,6 +74,7 @@ static const std::map<std::string, llm_chat_template> LLM_CHAT_TEMPLATES = {
     { "seed_oss",          LLM_CHAT_TEMPLATE_SEED_OSS          },
     { "grok-2",            LLM_CHAT_TEMPLATE_GROK_2            },
     { "pangu-embedded",    LLM_CHAT_TEMPLATE_PANGU_EMBED       },
+    { "solar-open",        LLM_CHAT_TEMPLATE_SOLAR_OPEN        },
 };
 
 llm_chat_template llm_chat_template_from_str(const std::string & name) {
@@ -216,6 +217,8 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
         return LLM_CHAT_TEMPLATE_GROK_2;
     } else if (tmpl_contains(LU8("[unused9]系统：[unused10]"))) {
         return LLM_CHAT_TEMPLATE_PANGU_EMBED;
+    } else if (tmpl_contains("<|begin|>") && tmpl_contains("<|end|>") && tmpl_contains("<|content|>")) {
+        return LLM_CHAT_TEMPLATE_SOLAR_OPEN;
     }
     return LLM_CHAT_TEMPLATE_UNKNOWN;
 }
@@ -845,6 +848,14 @@ int32_t llm_chat_apply_template(
         if (add_ass) {
             ss << "[unused9]助手：";
         }
+    } else if (tmpl == LLM_CHAT_TEMPLATE_SOLAR_OPEN) {
+        for (auto message : chat) {
+            std::string role(message->role);
+            ss << "<|begin|>" << role << "<|content|>" << message->content << "<|end|>";
+        }
+        if (add_ass) {
+            ss << "<|begin|>assistant";
+        }
     } else {
         // template not supported
         return -1;

package/src/llama.cpp/src/llama-chat.h

@@ -54,6 +54,7 @@ enum llm_chat_template {
     LLM_CHAT_TEMPLATE_SEED_OSS,
     LLM_CHAT_TEMPLATE_GROK_2,
     LLM_CHAT_TEMPLATE_PANGU_EMBED,
+    LLM_CHAT_TEMPLATE_SOLAR_OPEN,
     LLM_CHAT_TEMPLATE_UNKNOWN,
 };
 

package/src/llama.cpp/src/llama-context.cpp

@@ -294,8 +294,8 @@ llama_context::llama_context(
         // enabling pipeline parallelism in the scheduler increases memory usage, so it is only done when necessary
         bool pipeline_parallel =
             model.n_devices() > 1 &&
-            model.params.n_gpu_layers > (int) model.hparams.n_layer &&
-            model.params.split_mode == LLAMA_SPLIT_MODE_LAYER &&
+            model.n_gpu_layers() > model.hparams.n_layer &&
+            model.split_mode() == LLAMA_SPLIT_MODE_LAYER &&
             cparams.offload_kqv &&
             !model.has_tensor_overrides();
 
@@ -459,23 +459,22 @@ llama_context::llama_context(
 }
 
 llama_context::~llama_context() {
-    // FIXME this currently results in a use-after-free bug if the model is freed before the context
-    // if (!model.hparams.no_alloc) {
-    //     for (size_t i = 0; i < backend_ptrs.size(); ++i) {
-    //         ggml_backend_t             backend = backend_ptrs[i];
-    //         ggml_backend_buffer_type_t buft    = backend_buft[i];
-
-    //         const size_t size_exp = backend_buf_exp_size[i];
-    //         const size_t size_act = ggml_backend_sched_get_buffer_size(sched.get(), backend);
-    //         if (size_exp == size_act) {
-    //             LLAMA_LOG_DEBUG("%s: %10s compute buffer size is %8.4f MiB, matches expectation of %8.4f MiB\n",
-    //                 __func__, ggml_backend_buft_name(buft), size_act / (1024.0*1024.0), size_exp / (1024.0*1024.0));
-    //         } else {
-    //             LLAMA_LOG_WARN("%s: %10s compute buffer size of %8.4f MiB, does not match expectation of %8.4f MiB\n",
-    //                 __func__, ggml_backend_buft_name(buft), size_act / (1024.0*1024.0), size_exp / (1024.0*1024.0));
-    //         }
-    //     }
-    // }
+    if (!model.hparams.no_alloc) {
+        for (size_t i = 0; i < backend_ptrs.size(); ++i) {
+            ggml_backend_t             backend = backend_ptrs[i];
+            ggml_backend_buffer_type_t buft    = backend_buft[i];
+
+            const size_t size_exp = backend_buf_exp_size[i];
+            const size_t size_act = ggml_backend_sched_get_buffer_size(sched.get(), backend);
+            if (size_exp == size_act) {
+                LLAMA_LOG_DEBUG("%s: %10s compute buffer size is %8.4f MiB, matches expectation of %8.4f MiB\n",
+                    __func__, ggml_backend_buft_name(buft), size_act / (1024.0*1024.0), size_exp / (1024.0*1024.0));
+            } else {
+                LLAMA_LOG_WARN("%s: %10s compute buffer size of %8.4f MiB, does not match expectation of %8.4f MiB\n",
+                    __func__, ggml_backend_buft_name(buft), size_act / (1024.0*1024.0), size_exp / (1024.0*1024.0));
+            }
+        }
+    }
     ggml_opt_free(opt_ctx);
 }
 
@@ -1443,7 +1442,9 @@ uint32_t llama_context::graph_max_nodes(uint32_t n_tokens) const {
     if (model.arch == LLM_ARCH_QWEN3NEXT) {
         return std::max<uint32_t>(n_tokens * 40, 32u * model.n_tensors());
     }
-    return std::max<uint32_t>(1024u, 8u*model.n_tensors());
+    uint32_t res = std::max<uint32_t>(1024u, 8u*model.n_tensors());
+    res += model.n_lora_nodes;
+    return res;
 }
 
 llm_graph_result * llama_context::get_gf_res_reserve() const {
@@ -1571,7 +1572,7 @@ llm_graph_cb llama_context::graph_get_cb() const {
 
         // norm may be automatically assigned to the backend of the previous layer, increasing data transfer between backends
         // FIXME: fix in ggml_backend_sched
-        const bool full_offload = model.params.n_gpu_layers > (int) model.hparams.n_layer;
+        const bool full_offload = model.n_gpu_layers() > model.hparams.n_layer;
         if (ubatch.n_tokens < 32 || full_offload) {
             if (il != -1 && strcmp(name, "norm") == 0) {
                 const auto & dev_layer = model.dev_layer(il);

package/src/llama.cpp/src/llama-hparams.h

@@ -123,10 +123,11 @@ struct llama_hparams {
     llama_swa_type swa_type = LLAMA_SWA_TYPE_NONE;
     // the size of the sliding window (0 - no SWA)
     uint32_t n_swa = 0;
-    // if swa_layers[il] == true, then layer il is SWA
-    // if swa_layers[il] == false, then layer il is dense (i.e. non-SWA)
+    // if swa_layers[il] == 1, then layer il is SWA
+    // if swa_layers[il] == 0, then layer il is dense (i.e. non-SWA)
     // by default, all layers are dense
-    std::array<bool, LLAMA_MAX_LAYERS> swa_layers;
+    // note: using uint32_t type for compatibility reason
+    std::array<uint32_t, LLAMA_MAX_LAYERS> swa_layers;
 
     // for State Space Models
     uint32_t ssm_d_conv  = 0;

package/src/llama.cpp/src/llama-kv-cache.h

@@ -305,7 +305,7 @@ public:
             bool do_shift,
             stream_copy_info sc_info);
 
-    // used to create a batch procesing context from a batch
+    // used to create a batch processing context from a batch
     llama_kv_cache_context(
             llama_kv_cache * kv,
             slot_info_vec_t sinfos,

package/src/llama.cpp/src/llama-mmap.cpp

@@ -240,9 +240,10 @@ struct llama_file::impl {
                 throw std::runtime_error("unexpectedly reached end of file");
             }
         } else {
-            bool successful = false;
-            while (!successful) {
-                off_t ret = read(fd, ptr, len);
+            size_t bytes_read = 0;
+            while (bytes_read < len) {
+                const size_t to_read = len - bytes_read;
+                ssize_t ret = ::read(fd, reinterpret_cast<char *>(ptr) + bytes_read, to_read);
 
                 if (ret == -1) {
                     if (errno == EINTR) {
@@ -251,10 +252,16 @@ struct llama_file::impl {
                     throw std::runtime_error(format("read error: %s", strerror(errno)));
                 }
                 if (ret == 0) {
+                    // EOF: allow if this read was only pulling alignment padding past file end
+                    off_t pos = lseek(fd, 0, SEEK_CUR);
+                    if (pos != -1 && (size_t) pos == size) {
+                        std::memset(reinterpret_cast<char *>(ptr) + bytes_read, 0, len - bytes_read);
+                        return;
+                    }
                     throw std::runtime_error("unexpectedly reached end of file");
                 }
 
-                successful = true;
+                bytes_read += (size_t) ret;
             }
         }
     }

package/src/llama.cpp/src/llama-model-loader.cpp

@@ -462,6 +462,29 @@ namespace GGUFMeta {
         return get_key_or_arr(llm_kv(kid), result, n, required);
     }
 
+    bool llama_model_loader::get_key_or_arr(enum llm_kv kid, uint32_t & result, bool required) {
+        const std::string key = llm_kv(kid);
+
+        const int id = gguf_find_key(meta.get(), key.c_str());
+
+        if (id < 0) {
+            if (required) {
+                throw std::runtime_error(format("key not found in model: %s", key.c_str()));
+            }
+            return false;
+        }
+
+        // throw and error if type is an array
+        if (gguf_get_kv_type(meta.get(), id) == GGUF_TYPE_ARRAY) {
+            if (required) {
+                throw std::runtime_error(format("expected scalar, found array for key: %s", key.c_str()));
+            }
+            return false;
+        }
+
+        return get_key(key, result, required);
+    }
+
     // TODO: this is not very clever - figure out something better
     template bool llama_model_loader::get_key_or_arr<std::array<int, 4>>(enum llm_kv kid, std::array<int, 4> & result, uint32_t n, bool required);
     template bool llama_model_loader::get_key_or_arr<std::array<uint32_t, 512>>(enum llm_kv kid, std::array<uint32_t, 512> & result, uint32_t n, bool required);

package/src/llama.cpp/src/llama-model-loader.h

@@ -131,6 +131,8 @@ struct llama_model_loader {
     template<typename T>
     bool get_key_or_arr(enum llm_kv kid, T & result, uint32_t n, bool required = true);
 
+    bool get_key_or_arr(enum llm_kv kid, uint32_t & result, bool required = true);
+
     std::string get_arch_name() const;
 
     enum llm_arch get_arch() const;