whispercpp 1.3.3 → 1.3.4

data/ext/sources/examples/talk-llama/llama-vocab.h

@@ -6,6 +6,50 @@
 #include <vector>
 #include <memory>
 
+// pre-tokenization types
+enum llama_vocab_pre_type {
+    LLAMA_VOCAB_PRE_TYPE_DEFAULT        = 0,
+    LLAMA_VOCAB_PRE_TYPE_LLAMA3         = 1,
+    LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM   = 2,
+    LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODER = 3,
+    LLAMA_VOCAB_PRE_TYPE_FALCON         = 4,
+    LLAMA_VOCAB_PRE_TYPE_MPT            = 5,
+    LLAMA_VOCAB_PRE_TYPE_STARCODER      = 6,
+    LLAMA_VOCAB_PRE_TYPE_GPT2           = 7,
+    LLAMA_VOCAB_PRE_TYPE_REFACT         = 8,
+    LLAMA_VOCAB_PRE_TYPE_COMMAND_R      = 9,
+    LLAMA_VOCAB_PRE_TYPE_STABLELM2      = 10,
+    LLAMA_VOCAB_PRE_TYPE_QWEN2          = 11,
+    LLAMA_VOCAB_PRE_TYPE_OLMO           = 12,
+    LLAMA_VOCAB_PRE_TYPE_DBRX           = 13,
+    LLAMA_VOCAB_PRE_TYPE_SMAUG          = 14,
+    LLAMA_VOCAB_PRE_TYPE_PORO           = 15,
+    LLAMA_VOCAB_PRE_TYPE_CHATGLM3       = 16,
+    LLAMA_VOCAB_PRE_TYPE_CHATGLM4       = 17,
+    LLAMA_VOCAB_PRE_TYPE_VIKING         = 18,
+    LLAMA_VOCAB_PRE_TYPE_JAIS           = 19,
+    LLAMA_VOCAB_PRE_TYPE_TEKKEN         = 20,
+    LLAMA_VOCAB_PRE_TYPE_SMOLLM         = 21,
+    LLAMA_VOCAB_PRE_TYPE_CODESHELL      = 22,
+    LLAMA_VOCAB_PRE_TYPE_BLOOM          = 23,
+    LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH   = 24,
+    LLAMA_VOCAB_PRE_TYPE_EXAONE         = 25,
+    LLAMA_VOCAB_PRE_TYPE_CHAMELEON      = 26,
+    LLAMA_VOCAB_PRE_TYPE_MINERVA        = 27,
+    LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM  = 28,
+    LLAMA_VOCAB_PRE_TYPE_GPT4O          = 29,
+    LLAMA_VOCAB_PRE_TYPE_SUPERBPE       = 30,
+    LLAMA_VOCAB_PRE_TYPE_TRILLION       = 31,
+    LLAMA_VOCAB_PRE_TYPE_BAILINGMOE     = 32,
+    LLAMA_VOCAB_PRE_TYPE_LLAMA4         = 33,
+    LLAMA_VOCAB_PRE_TYPE_PIXTRAL        = 34,
+    LLAMA_VOCAB_PRE_TYPE_SEED_CODER     = 35,
+    LLAMA_VOCAB_PRE_TYPE_HUNYUAN        = 36,
+    LLAMA_VOCAB_PRE_TYPE_KIMI_K2        = 37,
+    LLAMA_VOCAB_PRE_TYPE_HUNYUAN_DENSE  = 38,
+    LLAMA_VOCAB_PRE_TYPE_GROK_2         = 39,
+};
+
 struct LLM_KV;
 struct llama_model_loader;
 
@@ -59,6 +103,7 @@ struct llama_vocab {
     llama_token token_sep() const;
     llama_token token_nl () const;
     llama_token token_pad() const;
+    llama_token token_mask() const;
 
     llama_token token_prefix() const;
     llama_token token_middle() const;

data/ext/sources/examples/talk-llama/llama.cpp

@@ -25,6 +25,18 @@
 // interface implementation
 //
 
+const char * llama_flash_attn_type_name(enum llama_flash_attn_type flash_attn_type) {
+    switch (flash_attn_type) {
+        case LLAMA_FLASH_ATTN_TYPE_AUTO:
+            return "auto";
+        case LLAMA_FLASH_ATTN_TYPE_DISABLED:
+            return "disabled";
+        case LLAMA_FLASH_ATTN_TYPE_ENABLED:
+            return "enabled";
+    }
+    GGML_ABORT("fatal error");
+}
+
 struct llama_sampler_chain_params llama_sampler_chain_default_params() {
     struct llama_sampler_chain_params result = {
         /*.no_perf                     =*/ true,
@@ -47,6 +59,7 @@ bool llama_supports_mlock(void) {
 
 bool llama_supports_gpu_offload(void) {
     return ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_GPU) != nullptr ||
+           ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_IGPU) != nullptr ||
            llama_supports_rpc();
 }
 
@@ -71,7 +84,9 @@ void llama_numa_init(enum ggml_numa_strategy numa) {
         GGML_ASSERT(dev && "CPU backend is not loaded");
         auto * reg = ggml_backend_dev_backend_reg(dev);
         auto * numa_init_fn = (decltype(ggml_numa_init) *) ggml_backend_reg_get_proc_address(reg, "ggml_backend_cpu_numa_init");
-        numa_init_fn(numa);
+        if (numa_init_fn) {
+            numa_init_fn(numa);
+        }
     }
 }
 
@@ -170,8 +185,13 @@ static struct llama_model * llama_model_load_from_file_impl(
             model->devices.push_back(*dev);
         }
     } else {
+        // default device selection
+
+        // build list of available devices
+        std::vector<ggml_backend_dev_t> gpus;
+        std::vector<ggml_backend_dev_t> igpus;
         std::vector<ggml_backend_dev_t> rpc_servers;
-        // use all available devices
+
         for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
             ggml_backend_dev_t dev = ggml_backend_dev_get(i);
             switch (ggml_backend_dev_type(dev)) {
@@ -180,19 +200,51 @@ static struct llama_model * llama_model_load_from_file_impl(
                     // skip CPU backends since they are handled separately
                     break;
 
-                case GGML_BACKEND_DEVICE_TYPE_GPU:
+                case GGML_BACKEND_DEVICE_TYPE_GPU: {
                     ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev);
                     if (ggml_backend_reg_name(reg) == std::string("RPC")) {
                         rpc_servers.push_back(dev);
                     } else {
-                        model->devices.push_back(dev);
+                        // check if there is already a GPU with the same device id
+                        ggml_backend_dev_props props;
+                        ggml_backend_dev_get_props(dev, &props);
+                        auto it = std::find_if(gpus.begin(), gpus.end(), [&props](ggml_backend_dev_t d) {
+                            ggml_backend_dev_props d_props;
+                            ggml_backend_dev_get_props(d, &d_props);
+                            if (props.device_id && d_props.device_id) {
+                                return strcmp(props.device_id, d_props.device_id) == 0;
+                            }
+                            return false;
+                        });
+
+                        if (it != gpus.end()) {
+                            LLAMA_LOG_INFO("%s: skipping device %s (%s) with id %s - already using device %s (%s) with the same id\n",
+                                    __func__,
+                                    ggml_backend_dev_name(dev), ggml_backend_dev_description(dev),
+                                    props.device_id ? props.device_id : "unknown id",
+                                    ggml_backend_dev_name(*it), ggml_backend_dev_description(*it));
+                        } else {
+                            gpus.push_back(dev);
+                        }
                     }
                     break;
+                }
+
+                case GGML_BACKEND_DEVICE_TYPE_IGPU:
+                    igpus.push_back(dev);
+                    break;
             }
         }
-        // add RPC servers at the front of the list
-        if (!rpc_servers.empty()) {
-            model->devices.insert(model->devices.begin(), rpc_servers.begin(), rpc_servers.end());
+
+        // add RPC servers at the front of the list to minimize network transfers
+        model->devices.insert(model->devices.begin(), rpc_servers.begin(), rpc_servers.end());
+
+        // add GPUs
+        model->devices.insert(model->devices.end(), gpus.begin(), gpus.end());
+
+        // add integrated GPUs only if no other devices were found
+        if (model->devices.empty()) {
+            model->devices.insert(model->devices.end(), igpus.begin(), igpus.end());
         }
     }
 
@@ -213,9 +265,12 @@ static struct llama_model * llama_model_load_from_file_impl(
     }
 
     for (auto * dev : model->devices) {
-        size_t free, total; // NOLINT
-        ggml_backend_dev_memory(dev, &free, &total);
-        LLAMA_LOG_INFO("%s: using device %s (%s) - %zu MiB free\n", __func__, ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), free/1024/1024);
+        ggml_backend_dev_props props;
+        ggml_backend_dev_get_props(dev, &props);
+        LLAMA_LOG_INFO("%s: using device %s (%s) (%s) - %zu MiB free\n", __func__,
+                ggml_backend_dev_name(dev), ggml_backend_dev_description(dev),
+                props.device_id ? props.device_id : "unknown id",
+                props.memory_free/1024/1024);
     }
 
     const int status = llama_model_load(path_model, splits, *model, params);

data/ext/sources/examples/talk-llama/llama.h

@@ -64,59 +64,18 @@ extern "C" {
 
     typedef struct llama_memory_i * llama_memory_t;
 
-    struct llama_kv_cache; // DEPRECATED (use llama_memory instead)
-
     typedef int32_t llama_pos;
     typedef int32_t llama_token;
     typedef int32_t llama_seq_id;
 
     enum llama_vocab_type {
-        LLAMA_VOCAB_TYPE_NONE = 0, // For models without vocab
-        LLAMA_VOCAB_TYPE_SPM  = 1, // LLaMA tokenizer based on byte-level BPE with byte fallback
-        LLAMA_VOCAB_TYPE_BPE  = 2, // GPT-2 tokenizer based on byte-level BPE
-        LLAMA_VOCAB_TYPE_WPM  = 3, // BERT tokenizer based on WordPiece
-        LLAMA_VOCAB_TYPE_UGM  = 4, // T5 tokenizer based on Unigram
-        LLAMA_VOCAB_TYPE_RWKV = 5, // RWKV tokenizer based on greedy tokenization
-    };
-
-    // pre-tokenization types
-    enum llama_vocab_pre_type {
-        LLAMA_VOCAB_PRE_TYPE_DEFAULT        = 0,
-        LLAMA_VOCAB_PRE_TYPE_LLAMA3         = 1,
-        LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM   = 2,
-        LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODER = 3,
-        LLAMA_VOCAB_PRE_TYPE_FALCON         = 4,
-        LLAMA_VOCAB_PRE_TYPE_MPT            = 5,
-        LLAMA_VOCAB_PRE_TYPE_STARCODER      = 6,
-        LLAMA_VOCAB_PRE_TYPE_GPT2           = 7,
-        LLAMA_VOCAB_PRE_TYPE_REFACT         = 8,
-        LLAMA_VOCAB_PRE_TYPE_COMMAND_R      = 9,
-        LLAMA_VOCAB_PRE_TYPE_STABLELM2      = 10,
-        LLAMA_VOCAB_PRE_TYPE_QWEN2          = 11,
-        LLAMA_VOCAB_PRE_TYPE_OLMO           = 12,
-        LLAMA_VOCAB_PRE_TYPE_DBRX           = 13,
-        LLAMA_VOCAB_PRE_TYPE_SMAUG          = 14,
-        LLAMA_VOCAB_PRE_TYPE_PORO           = 15,
-        LLAMA_VOCAB_PRE_TYPE_CHATGLM3       = 16,
-        LLAMA_VOCAB_PRE_TYPE_CHATGLM4       = 17,
-        LLAMA_VOCAB_PRE_TYPE_VIKING         = 18,
-        LLAMA_VOCAB_PRE_TYPE_JAIS           = 19,
-        LLAMA_VOCAB_PRE_TYPE_TEKKEN         = 20,
-        LLAMA_VOCAB_PRE_TYPE_SMOLLM         = 21,
-        LLAMA_VOCAB_PRE_TYPE_CODESHELL      = 22,
-        LLAMA_VOCAB_PRE_TYPE_BLOOM          = 23,
-        LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH   = 24,
-        LLAMA_VOCAB_PRE_TYPE_EXAONE         = 25,
-        LLAMA_VOCAB_PRE_TYPE_CHAMELEON      = 26,
-        LLAMA_VOCAB_PRE_TYPE_MINERVA        = 27,
-        LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM  = 28,
-        LLAMA_VOCAB_PRE_TYPE_GPT4O          = 29,
-        LLAMA_VOCAB_PRE_TYPE_SUPERBPE       = 30,
-        LLAMA_VOCAB_PRE_TYPE_TRILLION       = 31,
-        LLAMA_VOCAB_PRE_TYPE_BAILINGMOE     = 32,
-        LLAMA_VOCAB_PRE_TYPE_LLAMA4         = 33,
-        LLAMA_VOCAB_PRE_TYPE_PIXTRAL        = 34,
-        LLAMA_VOCAB_PRE_TYPE_SEED_CODER     = 35,
+        LLAMA_VOCAB_TYPE_NONE   = 0, // For models without vocab
+        LLAMA_VOCAB_TYPE_SPM    = 1, // LLaMA tokenizer based on byte-level BPE with byte fallback
+        LLAMA_VOCAB_TYPE_BPE    = 2, // GPT-2 tokenizer based on byte-level BPE
+        LLAMA_VOCAB_TYPE_WPM    = 3, // BERT tokenizer based on WordPiece
+        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
     };
 
     enum llama_rope_type {
@@ -191,6 +150,7 @@ extern "C" {
         //LLAMA_FTYPE_MOSTLY_Q4_0_8_8      = 35, // removed from gguf files, use Q4_0 and runtime repack
         LLAMA_FTYPE_MOSTLY_TQ1_0         = 36, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_TQ2_0         = 37, // except 1d tensors
+        LLAMA_FTYPE_MOSTLY_MXFP4_MOE     = 38, // except 1d tensors
 
         LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
     };
@@ -219,6 +179,14 @@ extern "C" {
         LLAMA_ATTENTION_TYPE_NON_CAUSAL  = 1,
     };
 
+    enum llama_flash_attn_type {
+        LLAMA_FLASH_ATTN_TYPE_AUTO     = -1,
+        LLAMA_FLASH_ATTN_TYPE_DISABLED = 0,
+        LLAMA_FLASH_ATTN_TYPE_ENABLED  = 1,
+    };
+
+    LLAMA_API const char * llama_flash_attn_type_name(enum llama_flash_attn_type flash_attn_type);
+
     enum llama_split_mode {
         LLAMA_SPLIT_MODE_NONE  = 0, // single GPU
         LLAMA_SPLIT_MODE_LAYER = 1, // split layers and KV across GPUs
@@ -238,7 +206,7 @@ extern "C" {
         llama_token_data * data;
         size_t size;
         int64_t selected; // this is the index in the data array (i.e. not the token id)
-        bool sorted;
+        bool sorted;      // note: do not assume the data is sorted - always check this flag
     } llama_token_data_array;
 
     typedef bool (*llama_progress_callback)(float progress, void * user_data);
@@ -323,10 +291,11 @@ extern "C" {
         const struct llama_model_kv_override * kv_overrides;
 
         // Keep the booleans together to avoid misalignment during copy-by-value.
-        bool vocab_only;    // only load the vocabulary, no weights
-        bool use_mmap;      // use mmap if possible
-        bool use_mlock;     // force system to keep model in RAM
-        bool check_tensors; // validate model tensor data
+        bool vocab_only;      // only load the vocabulary, no weights
+        bool use_mmap;        // use mmap if possible
+        bool use_mlock;       // force system to keep model in RAM
+        bool check_tensors;   // validate model tensor data
+        bool use_extra_bufts; // use extra buffer types (used for weight repacking)
     };
 
     // NOTE: changing the default values of parameters marked as [EXPERIMENTAL] may cause crashes or incorrect results in certain configurations
@@ -342,6 +311,7 @@ extern "C" {
         enum llama_rope_scaling_type rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
         enum llama_pooling_type      pooling_type;      // whether to pool (sum) embedding results by sequence id
         enum llama_attention_type    attention_type;    // attention type to use for embeddings
+        enum llama_flash_attn_type   flash_attn_type;   // when to enable Flash Attention
 
         // ref: https://github.com/ggml-org/llama.cpp/pull/2054
         float    rope_freq_base;   // RoPE base frequency, 0 = from model
@@ -351,7 +321,7 @@ extern "C" {
         float    yarn_beta_fast;   // YaRN low correction dim
         float    yarn_beta_slow;   // YaRN high correction dim
         uint32_t yarn_orig_ctx;    // YaRN original context size
-        float    defrag_thold;     // defragment the KV cache if holes/size > thold, <= 0 disabled (default)
+        float    defrag_thold;     // [DEPRECATED] defragment the KV cache if holes/size > thold, <= 0 disabled (default)
 
         ggml_backend_sched_eval_callback cb_eval;
         void * cb_eval_user_data;
@@ -368,12 +338,14 @@ extern "C" {
         // Keep the booleans together and at the end of the struct to avoid misalignment during copy-by-value.
         bool embeddings;  // if true, extract embeddings (together with logits)
         bool offload_kqv; // offload the KQV ops (including the KV cache) to GPU
-        bool flash_attn;  // use flash attention [EXPERIMENTAL]
         bool no_perf;     // measure performance timings
         bool op_offload;  // offload host tensor operations to device
         bool swa_full;    // use full-size SWA cache (https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
                           // NOTE: setting to false when n_seq_max > 1 can cause bad performance in some cases
                           //       ref: https://github.com/ggml-org/llama.cpp/pull/13845#issuecomment-2924800573
+        bool kv_unified;  // use a unified buffer across the input sequences when computing the attention
+                          // try to disable when n_seq_max > 1 for improved performance when the sequences do not share a large prefix
+                          // ref: https://github.com/ggml-org/llama.cpp/pull/14363
     };
 
     // model quantization parameters
@@ -503,8 +475,6 @@ extern "C" {
     LLAMA_API           llama_memory_t   llama_get_memory  (const struct llama_context * ctx);
     LLAMA_API  enum llama_pooling_type   llama_pooling_type(const struct llama_context * ctx); // TODO: rename to llama_get_pooling_type
 
-    DEPRECATED(LLAMA_API struct llama_kv_cache * llama_get_kv_self(struct llama_context * ctx), "use llama_get_memory instead");
-
     LLAMA_API const struct llama_vocab * llama_model_get_vocab(const struct llama_model * model);
     LLAMA_API enum llama_rope_type       llama_model_rope_type(const struct llama_model * model);
 
@@ -573,6 +543,9 @@ extern "C" {
     // Returns true if the model is recurrent (like Mamba, RWKV, etc.)
     LLAMA_API bool llama_model_is_recurrent(const struct llama_model * model);
 
+    // Returns true if the model is diffusion-based (like LLaDA, Dream, etc.)
+    LLAMA_API bool llama_model_is_diffusion(const struct llama_model * model);
+
     // Returns 0 on success
     LLAMA_API uint32_t llama_model_quantize(
             const char * fname_inp,
@@ -588,10 +561,32 @@ extern "C" {
             struct llama_model * model,
             const char * path_lora);
 
+    // Functions to access the adapter's GGUF metadata scalar values
+    // - The functions return the length of the string on success, or -1 on failure
+    // - The output string is always null-terminated and cleared on failure
+    // - When retrieving a string, an extra byte must be allocated to account for the null terminator
+    // - GGUF array values are not supported by these functions
+
+    // 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);
+
+    // Get the number of metadata key/value pairs
+    LLAMA_API int32_t llama_adapter_meta_count(const struct llama_adapter_lora * adapter);
+
+    // Get metadata key name by index
+    LLAMA_API int32_t llama_adapter_meta_key_by_index(const struct llama_adapter_lora * adapter, int32_t i, char * buf, size_t buf_size);
+
+    // Get metadata value as a string by index
+    LLAMA_API int32_t llama_adapter_meta_val_str_by_index(const struct llama_adapter_lora * adapter, int32_t i, char * buf, size_t buf_size);
+
     // Manually free a LoRA adapter
     // Note: loaded adapters will be free when the associated model is deleted
     LLAMA_API void llama_adapter_lora_free(struct llama_adapter_lora * adapter);
 
+    // Get the invocation tokens if the current lora is an alora
+    LLAMA_API uint64_t            llama_adapter_get_alora_n_invocation_tokens(const struct llama_adapter_lora * adapter);
+    LLAMA_API const llama_token * llama_adapter_get_alora_invocation_tokens  (const struct llama_adapter_lora * adapter);
+
     // The following functions operate on a llama_context, hence the naming: llama_verb_...
 
     // Add a loaded LoRA adapter to given context
@@ -698,111 +693,6 @@ extern "C" {
     // Check if the memory supports shifting
     LLAMA_API bool llama_memory_can_shift(llama_memory_t mem);
 
-    //
-    // KV cache for self-attention (TODO: deprecate in favor of llama_memory)
-    //
-
-    // Returns the number of tokens in the KV cache (slow, use only for debug)
-    // If a KV cell has multiple sequences assigned to it, it will be counted multiple times
-    DEPRECATED(LLAMA_API int32_t llama_kv_self_n_tokens(const struct llama_context * ctx),
-               "Use llama_kv_self_seq_pos_max() and llama_kv_self_seq_pos_min() instead (https://github.com/ggml-org/llama.cpp/issues/13793)");
-
-    // Returns the number of used KV cells (i.e. have at least one sequence assigned to them)
-    DEPRECATED(LLAMA_API int32_t llama_kv_self_used_cells(const struct llama_context * ctx),
-               "Use llama_kv_self_seq_pos_max() and llama_kv_self_seq_pos_min() instead (https://github.com/ggml-org/llama.cpp/issues/13793)");
-
-    // Clear the KV cache - both cell info is erased and KV data is zeroed
-    DEPRECATED(LLAMA_API void llama_kv_self_clear(
-                struct llama_context * ctx),
-            "Use llama_memory_clear() instead");
-
-    // Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
-    // Returns false if a partial sequence cannot be removed. Removing a whole sequence never fails
-    // seq_id < 0 : match any sequence
-    // p0 < 0     : [0,  p1]
-    // p1 < 0     : [p0, inf)
-    DEPRECATED(LLAMA_API bool llama_kv_self_seq_rm(
-            struct llama_context * ctx,
-                    llama_seq_id   seq_id,
-                       llama_pos   p0,
-                       llama_pos   p1),
-            "Use llama_memory_seq_rm() instead");
-
-    // Copy all tokens that belong to the specified sequence to another sequence
-    // Note that this does not allocate extra KV cache memory - it simply assigns the tokens to the new sequence
-    // p0 < 0 : [0,  p1]
-    // p1 < 0 : [p0, inf)
-    DEPRECATED(LLAMA_API void llama_kv_self_seq_cp(
-            struct llama_context * ctx,
-                    llama_seq_id   seq_id_src,
-                    llama_seq_id   seq_id_dst,
-                       llama_pos   p0,
-                       llama_pos   p1),
-            "Use llama_memory_seq_cp() instead");
-
-    // Removes all tokens that do not belong to the specified sequence
-    DEPRECATED(LLAMA_API void llama_kv_self_seq_keep(
-            struct llama_context * ctx,
-                    llama_seq_id   seq_id),
-            "Use llama_memory_seq_keep() instead");
-
-    // Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1)
-    // If the KV cache is RoPEd, the KV data is updated accordingly:
-    //   - lazily on next llama_decode()
-    // p0 < 0 : [0,  p1]
-    // p1 < 0 : [p0, inf)
-    DEPRECATED(LLAMA_API void llama_kv_self_seq_add(
-            struct llama_context * ctx,
-                    llama_seq_id   seq_id,
-                       llama_pos   p0,
-                       llama_pos   p1,
-                       llama_pos   delta),
-            "Use llama_memory_seq_add() instead");
-
-    // Integer division of the positions by factor of `d > 1`
-    // If the KV cache is RoPEd, the KV data is updated accordingly:
-    //   - lazily on next llama_decode()
-    // p0 < 0 : [0,  p1]
-    // p1 < 0 : [p0, inf)
-    DEPRECATED(void llama_kv_self_seq_div(
-            struct llama_context * ctx,
-                    llama_seq_id   seq_id,
-                       llama_pos   p0,
-                       llama_pos   p1,
-                             int   d),
-            "Use llama_memory_seq_div() instead");
-
-    // Returns the smallest position present in the KV cache for the specified sequence
-    // This is typically non-zero only for SWA caches
-    // Note that all positions in the range [pos_min, pos_max] are guaranteed to be present in the KV cache
-    // Return -1 if the sequence is empty
-    DEPRECATED(LLAMA_API llama_pos llama_kv_self_seq_pos_min(
-            struct llama_context * ctx,
-                    llama_seq_id   seq_id),
-            "Use llama_memory_seq_pos_min() instead");
-
-    // Returns the largest position present in the KV cache for the specified sequence
-    // Note that all positions in the range [pos_min, pos_max] are guaranteed to be present in the KV cache
-    // Return -1 if the sequence is empty
-    DEPRECATED(LLAMA_API llama_pos llama_kv_self_seq_pos_max(
-            struct llama_context * ctx,
-                    llama_seq_id   seq_id),
-            "Use llama_memory_seq_pos_max() instead");
-
-    // Defragment the KV cache
-    // This will be applied:
-    //   - lazily on next llama_decode()
-    DEPRECATED(LLAMA_API void llama_kv_self_defrag(struct llama_context * ctx),
-            "simply remove this call, the context will automatically decide when to do a defragmentation based on 'defrag_thold'");
-
-    // Check if the context supports KV cache shifting
-    DEPRECATED(LLAMA_API bool llama_kv_self_can_shift(const struct llama_context * ctx),
-            "use llama_memory_can_shift() instead");
-
-    // Apply the KV cache updates (such as K-shifts, defragmentation, etc.)
-    DEPRECATED(LLAMA_API void llama_kv_self_update(struct llama_context * ctx),
-            "simply remove this call, updates are applied lazily on the next llama_decode()");
-
     //
     // State / sessions
     //
@@ -901,6 +791,29 @@ extern "C" {
                           size_t   n_token_capacity,
                           size_t * n_token_count_out);
 
+#define LLAMA_STATE_SEQ_FLAGS_SWA_ONLY 1
+
+    typedef uint32_t llama_state_seq_flags;
+
+    LLAMA_API size_t llama_state_seq_get_size_ext(
+            struct llama_context * ctx,
+                    llama_seq_id   seq_id,
+           llama_state_seq_flags   flags);
+
+    LLAMA_API size_t llama_state_seq_get_data_ext(
+            struct llama_context * ctx,
+                         uint8_t * dst,
+                          size_t   size,
+                    llama_seq_id   seq_id,
+           llama_state_seq_flags   flags);
+
+    LLAMA_API size_t llama_state_seq_set_data_ext(
+            struct llama_context * ctx,
+                   const uint8_t * src,
+                          size_t   size,
+                    llama_seq_id   dest_seq_id,
+           llama_state_seq_flags   flags);
+
     //
     // Decoding
     //
@@ -992,6 +905,7 @@ extern "C" {
     // in the order they have appeared in the batch.
     // Rows: number of tokens for which llama_batch.logits[i] != 0
     // Cols: n_vocab
+    // TODO: deprecate in favor of llama_get_logits_ith() (ref: https://github.com/ggml-org/llama.cpp/pull/14853#issuecomment-3113143522)
     LLAMA_API float * llama_get_logits(struct llama_context * ctx);
 
     // Logits for the ith token. For positive indices, Equivalent to:
@@ -1006,6 +920,7 @@ extern "C" {
     // in the order they have appeared in the batch.
     // shape: [n_outputs*n_embd]
     // Otherwise, returns NULL.
+    // TODO: deprecate in favor of llama_get_embeddings_ith() (ref: https://github.com/ggml-org/llama.cpp/pull/14853#issuecomment-3113143522)
     LLAMA_API float * llama_get_embeddings(struct llama_context * ctx);
 
     // Get the embeddings for the ith token. For positive indices, Equivalent to:
@@ -1044,6 +959,7 @@ extern "C" {
     LLAMA_API llama_token llama_vocab_sep(const struct llama_vocab * vocab); // sentence separator
     LLAMA_API llama_token llama_vocab_nl (const struct llama_vocab * vocab); // next-line
     LLAMA_API llama_token llama_vocab_pad(const struct llama_vocab * vocab); // padding
+    LLAMA_API llama_token llama_vocab_mask(const struct llama_vocab * vocab); // mask
 
     LLAMA_API bool llama_vocab_get_add_bos(const struct llama_vocab * vocab);
     LLAMA_API bool llama_vocab_get_add_eos(const struct llama_vocab * vocab);
@@ -1244,11 +1160,6 @@ extern "C" {
     LLAMA_API struct llama_sampler * llama_sampler_init_greedy(void);
     LLAMA_API struct llama_sampler * llama_sampler_init_dist  (uint32_t seed);
 
-    /// @details Sorts candidate tokens by their logits in descending order and calculate probabilities based on logits.
-    /// NOTE: Avoid using on the full vocabulary as the sorting can become slow. For example, apply top-k or top-p sampling first.
-    DEPRECATED(LLAMA_API struct llama_sampler * llama_sampler_init_softmax    (void),
-        "will be removed in the future (see https://github.com/ggml-org/llama.cpp/pull/9896#discussion_r1800920915)");
-
     /// @details Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
     /// Setting k <= 0 makes this a noop
     LLAMA_API struct llama_sampler * llama_sampler_init_top_k      (int32_t k);
@@ -1418,23 +1329,25 @@ extern "C" {
     //
     // Performance utils
     //
-    // NOTE: Used by llama.cpp examples, avoid using in third-party apps. Instead, do your own performance measurements.
+    // NOTE: Used by llama.cpp examples/tools, avoid using in third-party apps. Instead, do your own performance measurements.
     //
 
     struct llama_perf_context_data {
-        double t_start_ms;
-        double t_load_ms;
-        double t_p_eval_ms;
-        double t_eval_ms;
-
-        int32_t n_p_eval;
-        int32_t n_eval;
+        // ms == milliseconds
+        double t_start_ms;  // absolute start time
+        double t_load_ms;   // time needed for loading the model
+        double t_p_eval_ms; // time needed for processing the prompt
+        double t_eval_ms;   // time needed for generating tokens
+
+        int32_t n_p_eval;   // number of prompt tokens
+        int32_t n_eval;     // number of generated tokens
+        int32_t n_reused;   // number of times a ggml compute graph had been reused
     };
 
     struct llama_perf_sampler_data {
-        double t_sample_ms;
+        double t_sample_ms; // time needed for sampling in ms
 
-        int32_t n_sample;
+        int32_t n_sample;   // number of sampled tokens
     };
 
     LLAMA_API struct llama_perf_context_data llama_perf_context      (const struct llama_context * ctx);
@@ -1446,6 +1359,9 @@ extern "C" {
     LLAMA_API void                           llama_perf_sampler_print(const struct llama_sampler * chain);
     LLAMA_API void                           llama_perf_sampler_reset(      struct llama_sampler * chain);
 
+    // print a breakdown of per-device memory use via LLAMA_LOG:
+    LLAMA_API void llama_memory_breakdown_print(const struct llama_context * ctx);
+
     //
     // training
     //
@@ -1464,6 +1380,8 @@ extern "C" {
 
         ggml_opt_get_optimizer_params get_opt_pars; // callback for calculating optimizer parameters
         void * get_opt_pars_ud;                     // userdata for calculating optimizer parameters
+
+        enum ggml_opt_optimizer_type optimizer_type;
     };
 
     LLAMA_API void llama_opt_init(struct llama_context * lctx, struct llama_model * model, struct llama_opt_params lopt_params);

data/ext/sources/examples/talk-llama/talk-llama.cpp

@@ -66,7 +66,7 @@ struct whisper_params {
     float top_p = 0.80f;
     float min_p = 0.01f;
     float temp  = 0.30f;
-    
+
     float vad_thold  = 0.6f;
     float freq_thold = 100.0f;
 
@@ -76,7 +76,7 @@ struct whisper_params {
     bool no_timestamps  = true;
     bool verbose_prompt = false;
     bool use_gpu        = true;
-    bool flash_attn     = false;
+    bool flash_attn     = true;
 
     std::string person      = "Georgi";
     std::string bot_name    = "LLaMA";
@@ -122,6 +122,7 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
         else if (arg == "-vp"  || arg == "--verbose-prompt") { params.verbose_prompt = true; }
         else if (arg == "-ng"  || arg == "--no-gpu")         { params.use_gpu        = false; }
         else if (arg == "-fa"  || arg == "--flash-attn")     { params.flash_attn     = true; }
+        else if (arg == "-nfa" || arg == "--no-flash-attn")  { params.flash_attn     = false; }
         else if (arg == "-p"   || arg == "--person")         { params.person         = argv[++i]; }
         else if (arg == "-bn"   || arg == "--bot-name")      { params.bot_name       = argv[++i]; }
         else if (arg == "--session")                         { params.path_session   = argv[++i]; }
@@ -175,7 +176,8 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -pe,      --print-energy   [%-7s] print sound energy (for debugging)\n",          params.print_energy ? "true" : "false");
     fprintf(stderr, "  -vp,      --verbose-prompt [%-7s] print prompt at start\n",                       params.verbose_prompt ? "true" : "false");
     fprintf(stderr, "  -ng,      --no-gpu         [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
-    fprintf(stderr, "  -fa,      --flash-attn     [%-7s] flash attention\n",                             params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -fa,      --flash-attn     [%-7s] enable flash attention\n",                      params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -nfa,     --no-flash-attn  [%-7s] disable flash attention\n",                     params.flash_attn ? "false" : "true");
     fprintf(stderr, "  -p NAME,  --person NAME    [%-7s] person name (for prompt selection)\n",          params.person.c_str());
     fprintf(stderr, "  -bn NAME, --bot-name NAME  [%-7s] bot name (to display)\n",                       params.bot_name.c_str());
     fprintf(stderr, "  -w TEXT,  --wake-command T [%-7s] wake-up command to listen for\n",               params.wake_cmd.c_str());
@@ -340,9 +342,10 @@ int main(int argc, char ** argv) {
     llama_context_params lcparams = llama_context_default_params();
 
     // tune these to your liking
-    lcparams.n_ctx      = 2048;
-    lcparams.n_threads  = params.n_threads;
-    lcparams.flash_attn = params.flash_attn;
+    lcparams.n_ctx     = 2048;
+    lcparams.n_threads = params.n_threads;
+
+    lcparams.flash_attn_type = params.flash_attn ? LLAMA_FLASH_ATTN_TYPE_AUTO : LLAMA_FLASH_ATTN_TYPE_DISABLED;
 
     struct llama_context * ctx_llama = llama_init_from_model(model_llama, lcparams);