cui-llama.rn 1.5.0 → 1.6.1

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-adapter.h

@@ -0,0 +1,76 @@
+#pragma once
+
+#include "llama.h"
+
+#include "ggml-cpp.h"
+
+#include <string>
+#include <unordered_map>
+#include <vector>
+
+// TODO: pimpl
+
+//
+// llama_adapter_cvec
+//
+
+struct llama_adapter_cvec {
+    lm_ggml_tensor * tensor_for(int il) const;
+
+    lm_ggml_tensor * apply_to(lm_ggml_context * ctx, lm_ggml_tensor * cur, int  il) const;
+
+    bool apply(
+            const llama_model & model,
+            const float * data,
+            size_t len,
+            int32_t n_embd,
+            int32_t il_start,
+            int32_t il_end);
+
+private:
+    bool init(const llama_model & model);
+
+    int32_t layer_start = -1;
+    int32_t layer_end   = -1;
+
+    std::vector<lm_ggml_context_ptr> ctxs;
+    std::vector<lm_ggml_backend_buffer_ptr> bufs;
+
+    std::vector<lm_ggml_tensor *> tensors; // per layer
+};
+
+//
+// llama_adapter_lora
+//
+
+struct llama_adapter_lora_weight {
+    lm_ggml_tensor * a = nullptr;
+    lm_ggml_tensor * b = nullptr;
+
+    // get actual scale based on rank and alpha
+    float get_scale(float alpha, float adapter_scale) const {
+        const float rank  = (float) b->ne[0];
+        const float scale = alpha ? adapter_scale * alpha / rank : adapter_scale;
+        return scale;
+    }
+
+    llama_adapter_lora_weight() = default;
+    llama_adapter_lora_weight(lm_ggml_tensor * a, lm_ggml_tensor * b) : a(a), b(b) {}
+};
+
+struct llama_adapter_lora {
+    // map tensor name to lora_a_b
+    std::unordered_map<std::string, llama_adapter_lora_weight> ab_map;
+
+    std::vector<lm_ggml_context_ptr> ctxs;
+    std::vector<lm_ggml_backend_buffer_ptr> bufs;
+
+    float alpha;
+
+    llama_adapter_lora() = default;
+    ~llama_adapter_lora() = default;
+
+    llama_adapter_lora_weight * get_weight(lm_ggml_tensor * w);
+};
+
+using llama_adapter_loras = std::unordered_map<llama_adapter_lora *, float>;

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-arch.h

@@ -0,0 +1,437 @@
+#pragma once
+
+#include "ggml.h" // lm_ggml_op
+
+#include <string>
+
+//
+// gguf constants (sync with gguf.py)
+//
+
+enum llm_arch {
+    LLM_ARCH_LLAMA,
+    LLM_ARCH_LLAMA4,
+    LLM_ARCH_DECI,
+    LLM_ARCH_FALCON,
+    LLM_ARCH_BAICHUAN,
+    LLM_ARCH_GROK,
+    LLM_ARCH_GPT2,
+    LLM_ARCH_GPTJ,
+    LLM_ARCH_GPTNEOX,
+    LLM_ARCH_MPT,
+    LLM_ARCH_STARCODER,
+    LLM_ARCH_REFACT,
+    LLM_ARCH_BERT,
+    LLM_ARCH_NOMIC_BERT,
+    LLM_ARCH_NOMIC_BERT_MOE,
+    LLM_ARCH_JINA_BERT_V2,
+    LLM_ARCH_BLOOM,
+    LLM_ARCH_STABLELM,
+    LLM_ARCH_QWEN,
+    LLM_ARCH_QWEN2,
+    LLM_ARCH_QWEN2MOE,
+    LLM_ARCH_QWEN2VL,
+    LLM_ARCH_QWEN3,
+    LLM_ARCH_QWEN3MOE,
+    LLM_ARCH_PHI2,
+    LLM_ARCH_PHI3,
+    LLM_ARCH_PHIMOE,
+    LLM_ARCH_PLAMO,
+    LLM_ARCH_CODESHELL,
+    LLM_ARCH_ORION,
+    LLM_ARCH_INTERNLM2,
+    LLM_ARCH_MINICPM,
+    LLM_ARCH_MINICPM3,
+    LLM_ARCH_GEMMA,
+    LLM_ARCH_GEMMA2,
+    LLM_ARCH_GEMMA3,
+    LLM_ARCH_STARCODER2,
+    LLM_ARCH_MAMBA,
+    LLM_ARCH_XVERSE,
+    LLM_ARCH_COMMAND_R,
+    LLM_ARCH_COHERE2,
+    LLM_ARCH_DBRX,
+    LLM_ARCH_OLMO,
+    LLM_ARCH_OLMO2,
+    LLM_ARCH_OLMOE,
+    LLM_ARCH_OPENELM,
+    LLM_ARCH_ARCTIC,
+    LLM_ARCH_DEEPSEEK,
+    LLM_ARCH_DEEPSEEK2,
+    LLM_ARCH_CHATGLM,
+    LLM_ARCH_GLM4,
+    LLM_ARCH_BITNET,
+    LLM_ARCH_T5,
+    LLM_ARCH_T5ENCODER,
+    LLM_ARCH_JAIS,
+    LLM_ARCH_NEMOTRON,
+    LLM_ARCH_EXAONE,
+    LLM_ARCH_RWKV6,
+    LLM_ARCH_RWKV6QWEN2,
+    LLM_ARCH_RWKV7,
+    LLM_ARCH_ARWKV7,
+    LLM_ARCH_GRANITE,
+    LLM_ARCH_GRANITE_MOE,
+    LLM_ARCH_CHAMELEON,
+    LLM_ARCH_WAVTOKENIZER_DEC,
+    LLM_ARCH_PLM,
+    LLM_ARCH_BAILINGMOE,
+    LLM_ARCH_UNKNOWN,
+};
+
+enum llm_kv {
+    LLM_KV_GENERAL_TYPE,
+    LLM_KV_GENERAL_ARCHITECTURE,
+    LLM_KV_GENERAL_QUANTIZATION_VERSION,
+    LLM_KV_GENERAL_ALIGNMENT,
+    LLM_KV_GENERAL_FILE_TYPE,
+    LLM_KV_GENERAL_NAME,
+    LLM_KV_GENERAL_AUTHOR,
+    LLM_KV_GENERAL_VERSION,
+    LLM_KV_GENERAL_URL,
+    LLM_KV_GENERAL_DESCRIPTION,
+    LLM_KV_GENERAL_LICENSE,
+    LLM_KV_GENERAL_SOURCE_URL,
+    LLM_KV_GENERAL_SOURCE_HF_REPO,
+
+    LLM_KV_VOCAB_SIZE,
+    LLM_KV_CONTEXT_LENGTH,
+    LLM_KV_EMBEDDING_LENGTH,
+    LLM_KV_FEATURES_LENGTH,
+    LLM_KV_BLOCK_COUNT,
+    LLM_KV_LEADING_DENSE_BLOCK_COUNT,
+    LLM_KV_FEED_FORWARD_LENGTH,
+    LLM_KV_EXPERT_FEED_FORWARD_LENGTH,
+    LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH,
+    LLM_KV_USE_PARALLEL_RESIDUAL,
+    LLM_KV_TENSOR_DATA_LAYOUT,
+    LLM_KV_EXPERT_COUNT,
+    LLM_KV_EXPERT_USED_COUNT,
+    LLM_KV_EXPERT_SHARED_COUNT,
+    LLM_KV_EXPERT_WEIGHTS_SCALE,
+    LLM_KV_EXPERT_WEIGHTS_NORM,
+    LLM_KV_EXPERT_GATING_FUNC,
+    LLM_KV_MOE_EVERY_N_LAYERS,
+    LLM_KV_POOLING_TYPE,
+    LLM_KV_LOGIT_SCALE,
+    LLM_KV_DECODER_START_TOKEN_ID,
+    LLM_KV_ATTN_LOGIT_SOFTCAPPING,
+    LLM_KV_FINAL_LOGIT_SOFTCAPPING,
+    LLM_KV_SWIN_NORM,
+    LLM_KV_RESCALE_EVERY_N_LAYERS,
+    LLM_KV_TIME_MIX_EXTRA_DIM,
+    LLM_KV_TIME_DECAY_EXTRA_DIM,
+    LLM_KV_RESIDUAL_SCALE,
+    LLM_KV_EMBEDDING_SCALE,
+    LLM_KV_TOKEN_SHIFT_COUNT,
+    LLM_KV_INTERLEAVE_MOE_LAYER_STEP,
+
+    LLM_KV_ATTENTION_HEAD_COUNT,
+    LLM_KV_ATTENTION_HEAD_COUNT_KV,
+    LLM_KV_ATTENTION_MAX_ALIBI_BIAS,
+    LLM_KV_ATTENTION_CLAMP_KQV,
+    LLM_KV_ATTENTION_KEY_LENGTH,
+    LLM_KV_ATTENTION_VALUE_LENGTH,
+    LLM_KV_ATTENTION_LAYERNORM_EPS,
+    LLM_KV_ATTENTION_LAYERNORM_RMS_EPS,
+    LLM_KV_ATTENTION_GROUPNORM_EPS,
+    LLM_KV_ATTENTION_GROUPNORM_GROUPS,
+    LLM_KV_ATTENTION_CAUSAL,
+    LLM_KV_ATTENTION_Q_LORA_RANK,
+    LLM_KV_ATTENTION_KV_LORA_RANK,
+    LLM_KV_ATTENTION_DECAY_LORA_RANK,
+    LLM_KV_ATTENTION_ICLR_LORA_RANK,
+    LLM_KV_ATTENTION_VALUE_RESIDUAL_MIX_LORA_RANK,
+    LLM_KV_ATTENTION_GATE_LORA_RANK,
+    LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,
+    LLM_KV_ATTENTION_SLIDING_WINDOW,
+    LLM_KV_ATTENTION_SCALE,
+    LLM_KV_ATTENTION_KEY_LENGTH_MLA,
+    LLM_KV_ATTENTION_VALUE_LENGTH_MLA,
+
+    LLM_KV_ROPE_DIMENSION_COUNT,
+    LLM_KV_ROPE_DIMENSION_SECTIONS,
+    LLM_KV_ROPE_FREQ_BASE,
+    LLM_KV_ROPE_SCALE_LINEAR,
+    LLM_KV_ROPE_SCALING_TYPE,
+    LLM_KV_ROPE_SCALING_FACTOR,
+    LLM_KV_ROPE_SCALING_ATTN_FACTOR,
+    LLM_KV_ROPE_SCALING_ORIG_CTX_LEN,
+    LLM_KV_ROPE_SCALING_FINETUNED,
+    LLM_KV_ROPE_SCALING_YARN_LOG_MUL,
+
+    LLM_KV_SPLIT_NO,
+    LLM_KV_SPLIT_COUNT,
+    LLM_KV_SPLIT_TENSORS_COUNT,
+
+    LLM_KV_SSM_INNER_SIZE,
+    LLM_KV_SSM_CONV_KERNEL,
+    LLM_KV_SSM_STATE_SIZE,
+    LLM_KV_SSM_TIME_STEP_RANK,
+    LLM_KV_SSM_DT_B_C_RMS,
+
+    LLM_KV_WKV_HEAD_SIZE,
+
+    LLM_KV_TOKENIZER_MODEL,
+    LLM_KV_TOKENIZER_PRE,
+    LLM_KV_TOKENIZER_LIST,
+    LLM_KV_TOKENIZER_TOKEN_TYPE,
+    LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT,
+    LLM_KV_TOKENIZER_SCORES,
+    LLM_KV_TOKENIZER_MERGES,
+    LLM_KV_TOKENIZER_BOS_ID,
+    LLM_KV_TOKENIZER_EOS_ID,
+    LLM_KV_TOKENIZER_EOT_ID,
+    LLM_KV_TOKENIZER_EOM_ID,
+    LLM_KV_TOKENIZER_UNK_ID,
+    LLM_KV_TOKENIZER_SEP_ID,
+    LLM_KV_TOKENIZER_PAD_ID,
+    LLM_KV_TOKENIZER_CLS_ID,
+    LLM_KV_TOKENIZER_MASK_ID,
+    LLM_KV_TOKENIZER_ADD_BOS,
+    LLM_KV_TOKENIZER_ADD_EOS,
+    LLM_KV_TOKENIZER_ADD_PREFIX,
+    LLM_KV_TOKENIZER_REMOVE_EXTRA_WS,
+    LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP,
+    LLM_KV_TOKENIZER_HF_JSON,
+    LLM_KV_TOKENIZER_RWKV,
+    LLM_KV_TOKENIZER_CHAT_TEMPLATE,
+    LLM_KV_TOKENIZER_CHAT_TEMPLATE_N,
+    LLM_KV_TOKENIZER_FIM_PRE_ID,
+    LLM_KV_TOKENIZER_FIM_SUF_ID,
+    LLM_KV_TOKENIZER_FIM_MID_ID,
+    LLM_KV_TOKENIZER_FIM_PAD_ID,
+    LLM_KV_TOKENIZER_FIM_REP_ID,
+    LLM_KV_TOKENIZER_FIM_SEP_ID,
+
+    LLM_KV_ADAPTER_TYPE,
+    LLM_KV_ADAPTER_LORA_ALPHA,
+
+    LLM_KV_POSNET_EMBEDDING_LENGTH,
+    LLM_KV_POSNET_BLOCK_COUNT,
+
+    LLM_KV_CONVNEXT_EMBEDDING_LENGTH,
+    LLM_KV_CONVNEXT_BLOCK_COUNT,
+
+    // deprecated:
+    LLM_KV_TOKENIZER_PREFIX_ID,
+    LLM_KV_TOKENIZER_SUFFIX_ID,
+    LLM_KV_TOKENIZER_MIDDLE_ID,
+};
+
+enum llm_tensor {
+    LLM_TENSOR_TOKEN_EMBD,
+    LLM_TENSOR_TOKEN_EMBD_NORM,
+    LLM_TENSOR_TOKEN_TYPES,
+    LLM_TENSOR_POS_EMBD,
+    LLM_TENSOR_OUTPUT,
+    LLM_TENSOR_OUTPUT_NORM,
+    LLM_TENSOR_ROPE_FREQS,
+    LLM_TENSOR_ROPE_FACTORS_LONG,
+    LLM_TENSOR_ROPE_FACTORS_SHORT,
+    LLM_TENSOR_ATTN_Q,
+    LLM_TENSOR_ATTN_K,
+    LLM_TENSOR_ATTN_V,
+    LLM_TENSOR_ATTN_QKV,
+    LLM_TENSOR_ATTN_OUT,
+    LLM_TENSOR_ATTN_NORM,
+    LLM_TENSOR_ATTN_NORM_2,
+    LLM_TENSOR_ATTN_OUT_NORM,
+    LLM_TENSOR_ATTN_POST_NORM,
+    LLM_TENSOR_ATTN_ROT_EMBD,
+    LLM_TENSOR_FFN_GATE_INP,
+    LLM_TENSOR_FFN_GATE_INP_SHEXP,
+    LLM_TENSOR_FFN_NORM,
+    LLM_TENSOR_FFN_POST_NORM,
+    LLM_TENSOR_FFN_GATE,
+    LLM_TENSOR_FFN_DOWN,
+    LLM_TENSOR_FFN_UP,
+    LLM_TENSOR_FFN_ACT,
+    LLM_TENSOR_FFN_DOWN_EXP,  // split experts for backward compatibility
+    LLM_TENSOR_FFN_GATE_EXP,
+    LLM_TENSOR_FFN_UP_EXP,
+    LLM_TENSOR_FFN_NORM_EXPS,
+    LLM_TENSOR_FFN_DOWN_EXPS, // merged experts
+    LLM_TENSOR_FFN_GATE_EXPS,
+    LLM_TENSOR_FFN_UP_EXPS,
+    LLM_TENSOR_FFN_DOWN_SHEXP,
+    LLM_TENSOR_FFN_GATE_SHEXP,
+    LLM_TENSOR_FFN_UP_SHEXP,
+    LLM_TENSOR_FFN_EXP_PROBS_B,
+    LLM_TENSOR_ATTN_Q_NORM,
+    LLM_TENSOR_ATTN_K_NORM,
+    LLM_TENSOR_LAYER_OUT_NORM,
+    LLM_TENSOR_POST_ATTN_NORM,
+    LLM_TENSOR_POST_MLP_NORM,
+    LLM_TENSOR_SSM_IN,
+    LLM_TENSOR_SSM_CONV1D,
+    LLM_TENSOR_SSM_X,
+    LLM_TENSOR_SSM_DT,
+    LLM_TENSOR_SSM_A,
+    LLM_TENSOR_SSM_D,
+    LLM_TENSOR_SSM_OUT,
+    LLM_TENSOR_TIME_MIX_W0,
+    LLM_TENSOR_TIME_MIX_W1,
+    LLM_TENSOR_TIME_MIX_W2,
+    LLM_TENSOR_TIME_MIX_A0,
+    LLM_TENSOR_TIME_MIX_A1,
+    LLM_TENSOR_TIME_MIX_A2,
+    LLM_TENSOR_TIME_MIX_V0,
+    LLM_TENSOR_TIME_MIX_V1,
+    LLM_TENSOR_TIME_MIX_V2,
+    LLM_TENSOR_TIME_MIX_G1,
+    LLM_TENSOR_TIME_MIX_G2,
+    LLM_TENSOR_TIME_MIX_K_K,
+    LLM_TENSOR_TIME_MIX_K_A,
+    LLM_TENSOR_TIME_MIX_R_K,
+    LLM_TENSOR_TIME_MIX_LERP_X,
+    LLM_TENSOR_TIME_MIX_LERP_W,
+    LLM_TENSOR_TIME_MIX_LERP_K,
+    LLM_TENSOR_TIME_MIX_LERP_V,
+    LLM_TENSOR_TIME_MIX_LERP_R,
+    LLM_TENSOR_TIME_MIX_LERP_G,
+    LLM_TENSOR_TIME_MIX_LERP_FUSED,
+    LLM_TENSOR_TIME_MIX_FIRST,
+    LLM_TENSOR_TIME_MIX_DECAY,
+    LLM_TENSOR_TIME_MIX_DECAY_W1,
+    LLM_TENSOR_TIME_MIX_DECAY_W2,
+    LLM_TENSOR_TIME_MIX_KEY,
+    LLM_TENSOR_TIME_MIX_VALUE,
+    LLM_TENSOR_TIME_MIX_RECEPTANCE,
+    LLM_TENSOR_TIME_MIX_GATE,
+    LLM_TENSOR_TIME_MIX_LN,
+    LLM_TENSOR_TIME_MIX_OUTPUT,
+    LLM_TENSOR_CHANNEL_MIX_LERP_K,
+    LLM_TENSOR_CHANNEL_MIX_LERP_R,
+    LLM_TENSOR_CHANNEL_MIX_KEY,
+    LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,
+    LLM_TENSOR_CHANNEL_MIX_VALUE,
+    LLM_TENSOR_ATTN_Q_A,
+    LLM_TENSOR_ATTN_Q_B,
+    LLM_TENSOR_ATTN_KV_A_MQA,
+    LLM_TENSOR_ATTN_KV_B,
+    LLM_TENSOR_ATTN_K_B,
+    LLM_TENSOR_ATTN_V_B,
+    LLM_TENSOR_ATTN_Q_A_NORM,
+    LLM_TENSOR_ATTN_KV_A_NORM,
+    LLM_TENSOR_ATTN_SUB_NORM,
+    LLM_TENSOR_FFN_SUB_NORM,
+    LLM_TENSOR_DEC_ATTN_NORM,
+    LLM_TENSOR_DEC_ATTN_Q,
+    LLM_TENSOR_DEC_ATTN_K,
+    LLM_TENSOR_DEC_ATTN_V,
+    LLM_TENSOR_DEC_ATTN_OUT,
+    LLM_TENSOR_DEC_ATTN_REL_B,
+    LLM_TENSOR_DEC_CROSS_ATTN_NORM,
+    LLM_TENSOR_DEC_CROSS_ATTN_Q,
+    LLM_TENSOR_DEC_CROSS_ATTN_K,
+    LLM_TENSOR_DEC_CROSS_ATTN_V,
+    LLM_TENSOR_DEC_CROSS_ATTN_OUT,
+    LLM_TENSOR_DEC_CROSS_ATTN_REL_B,
+    LLM_TENSOR_DEC_FFN_NORM,
+    LLM_TENSOR_DEC_FFN_GATE,
+    LLM_TENSOR_DEC_FFN_DOWN,
+    LLM_TENSOR_DEC_FFN_UP,
+    LLM_TENSOR_DEC_OUTPUT_NORM,
+    LLM_TENSOR_ENC_ATTN_NORM,
+    LLM_TENSOR_ENC_ATTN_Q,
+    LLM_TENSOR_ENC_ATTN_K,
+    LLM_TENSOR_ENC_ATTN_V,
+    LLM_TENSOR_ENC_ATTN_OUT,
+    LLM_TENSOR_ENC_ATTN_REL_B,
+    LLM_TENSOR_ENC_FFN_NORM,
+    LLM_TENSOR_ENC_FFN_GATE,
+    LLM_TENSOR_ENC_FFN_DOWN,
+    LLM_TENSOR_ENC_FFN_UP,
+    LLM_TENSOR_ENC_OUTPUT_NORM,
+    LLM_TENSOR_CLS,
+    LLM_TENSOR_CLS_OUT,
+    LLM_TENSOR_CONV1D,
+    LLM_TENSOR_CONVNEXT_DW,
+    LLM_TENSOR_CONVNEXT_NORM,
+    LLM_TENSOR_CONVNEXT_PW1,
+    LLM_TENSOR_CONVNEXT_PW2,
+    LLM_TENSOR_CONVNEXT_GAMMA,
+    LLM_TENSOR_POS_NET_CONV1,
+    LLM_TENSOR_POS_NET_CONV2,
+    LLM_TENSOR_POS_NET_NORM,
+    LLM_TENSOR_POS_NET_NORM1,
+    LLM_TENSOR_POS_NET_NORM2,
+    LLM_TENSOR_POS_NET_ATTN_NORM,
+    LLM_TENSOR_POS_NET_ATTN_Q,
+    LLM_TENSOR_POS_NET_ATTN_K,
+    LLM_TENSOR_POS_NET_ATTN_V,
+    LLM_TENSOR_POS_NET_ATTN_OUT,
+};
+
+enum llm_tensor_layer {
+    LLM_TENSOR_LAYER_INPUT,
+    LLM_TENSOR_LAYER_REPEATING,
+    LLM_TENSOR_LAYER_OUTPUT,
+};
+
+struct LLM_KV {
+    LLM_KV(llm_arch arch, const char * suffix = nullptr);
+
+    llm_arch arch;
+    const char * suffix;
+
+    std::string operator()(llm_kv kv) const;
+};
+
+// helper to handle gguf constants
+// usage:
+//
+//   const auto tn = LLM_TN(LLM_ARCH_LLAMA);
+//
+//   std::string name = tn(LLM_TENSOR_OUTPUT);                     -> "output"
+//   std::string name = tn(LLM_TENSOR_TOKEN_EMBD, "bias");         -> "token_embd.bias"
+//   std::string name = tn(LLM_TENSOR_ATTN_NORM, "weight", 3);     -> "blk.3.attn_norm.weight"
+//
+struct LLM_TN_IMPL {
+    const llm_arch arch;
+    const llm_tensor tensor;
+    const char * const suffix;
+    const int bid;
+    const int xid;
+
+    std::string str() const;
+
+    operator std::string() const {
+        return str();
+    }
+
+    friend bool operator==(const std::string & str, const LLM_TN_IMPL & tn) {
+        return str == tn.str();
+    }
+
+    friend bool operator!=(const std::string & str, const LLM_TN_IMPL & tn) {
+        return str != tn.str();
+    }
+};
+
+struct LLM_TN {
+    LLM_TN(llm_arch arch) : arch(arch) {}
+
+    llm_arch arch;
+
+    LLM_TN_IMPL operator()(llm_tensor tensor, const char * suffix, int bid = -1, int xid = -1) const {
+        return { arch, tensor, suffix, bid, xid };
+    }
+
+    LLM_TN_IMPL operator()(llm_tensor tensor, int bid = -1, int xid = -1) const {
+        return { arch, tensor, nullptr, bid, xid };
+    }
+};
+
+
+struct llm_tensor_info {
+    llm_tensor_layer layer;
+    lm_ggml_op op;
+};
+
+const char * llm_arch_name(llm_arch arch);
+
+llm_arch llm_arch_from_string(const std::string & name);
+
+const llm_tensor_info & llm_tensor_info_for(llm_tensor tensor);

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-batch.h

@@ -0,0 +1,89 @@
+#pragma once
+
+#include "llama.h"
+
+#include <array>
+#include <vector>
+
+// very similar to llama_batch,
+// but has more metadata about sequences
+struct llama_ubatch {
+    bool equal_seqs;
+    // TODO: whole_seqs for embeddings?
+
+    uint32_t n_tokens; // total tokens (n_seq_tokens * n_seqs)
+    uint32_t n_seq_tokens; // tokens per sequence
+    uint32_t n_seqs;
+
+    llama_token  *  token;    // [n_tokens]
+    float        *  embd;     // [n_embd, n_tokens]
+    llama_pos    *  pos;      // [n_tokens]
+    int32_t      *  n_seq_id; // [n_seqs]
+    llama_seq_id ** seq_id;   // [n_seqs]
+    int8_t       *  output;   // [n_tokens]
+};
+
+struct llama_sbatch_seq {
+    int32_t n_seq_id;
+
+    llama_seq_id * seq_id;
+
+    size_t offset;
+    size_t length;
+};
+
+// sequence-length-aware batch splitting
+struct llama_sbatch {
+    // tokens left in this batch
+    size_t n_tokens;
+
+    size_t n_embd;
+
+    bool logits_all; // TODO: remove once lctx.logits_all is removed too
+
+    // sorted indices into the batch
+    std::vector<int64_t> ids;
+    // batch indices of the output
+    std::vector<int64_t> out_ids;
+    std::vector<llama_sbatch_seq> seq;
+
+    const llama_batch * batch = nullptr;
+
+    // buffers for the ubatch
+    std::vector<llama_token>    ubatch_token;
+    std::vector<float>          ubatch_embd;
+    std::vector<llama_pos>      ubatch_pos;
+    std::vector<int32_t>        ubatch_n_seq_id;
+    std::vector<llama_seq_id *> ubatch_seq_id;
+    std::vector<int8_t>         ubatch_output;
+
+    llama_ubatch reserve_ubatch(size_t n_ubatch, bool has_embd = false);
+
+    void add_seq_to_ubatch(llama_ubatch & ubatch, llama_sbatch_seq & seq, size_t length);
+
+    // simple split, unknown number of sequences of unequal lengths
+    llama_ubatch split_simple(size_t n_ubatch);
+
+    // make batches of equal-length sequences
+    llama_ubatch split_equal(size_t n_ubatch);
+
+    // sequence-wise split
+    llama_ubatch split_seq(size_t n_ubatch);
+
+    llama_sbatch() = default;
+    llama_sbatch(const llama_batch & batch, size_t n_embd, bool simple_split = false, bool logits_all = false);
+};
+
+// temporary allocate memory for the input batch if needed
+struct llama_batch_allocr {
+    struct llama_batch batch;
+
+    std::array<llama_seq_id, 1> seq_id_0 = { 0 }; // default sequence id
+    std::vector<llama_pos>      pos;
+    std::vector<int32_t>        n_seq_id;
+    std::vector<llama_seq_id *> seq_id;
+    std::vector<int8_t>         logits;
+
+    // optionally fulfill the batch returned by llama_batch_get_one
+    llama_batch_allocr(struct llama_batch in_batch, llama_pos p0);
+};

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-chat.h

@@ -0,0 +1,57 @@
+#pragma once
+
+#include <string>
+#include <vector>
+#include <cstdint>
+
+enum llm_chat_template {
+    LLM_CHAT_TEMPLATE_CHATML,
+    LLM_CHAT_TEMPLATE_LLAMA_2,
+    LLM_CHAT_TEMPLATE_LLAMA_2_SYS,
+    LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS,
+    LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP,
+    LLM_CHAT_TEMPLATE_MISTRAL_V1,
+    LLM_CHAT_TEMPLATE_MISTRAL_V3,
+    LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN,
+    LLM_CHAT_TEMPLATE_MISTRAL_V7,
+    LLM_CHAT_TEMPLATE_PHI_3,
+    LLM_CHAT_TEMPLATE_PHI_4,
+    LLM_CHAT_TEMPLATE_FALCON_3,
+    LLM_CHAT_TEMPLATE_ZEPHYR,
+    LLM_CHAT_TEMPLATE_MONARCH,
+    LLM_CHAT_TEMPLATE_GEMMA,
+    LLM_CHAT_TEMPLATE_ORION,
+    LLM_CHAT_TEMPLATE_OPENCHAT,
+    LLM_CHAT_TEMPLATE_VICUNA,
+    LLM_CHAT_TEMPLATE_VICUNA_ORCA,
+    LLM_CHAT_TEMPLATE_DEEPSEEK,
+    LLM_CHAT_TEMPLATE_DEEPSEEK_2,
+    LLM_CHAT_TEMPLATE_DEEPSEEK_3,
+    LLM_CHAT_TEMPLATE_COMMAND_R,
+    LLM_CHAT_TEMPLATE_LLAMA_3,
+    LLM_CHAT_TEMPLATE_CHATGLM_3,
+    LLM_CHAT_TEMPLATE_CHATGLM_4,
+    LLM_CHAT_TEMPLATE_GLMEDGE,
+    LLM_CHAT_TEMPLATE_MINICPM,
+    LLM_CHAT_TEMPLATE_EXAONE_3,
+    LLM_CHAT_TEMPLATE_RWKV_WORLD,
+    LLM_CHAT_TEMPLATE_GRANITE,
+    LLM_CHAT_TEMPLATE_GIGACHAT,
+    LLM_CHAT_TEMPLATE_MEGREZ,
+    LLM_CHAT_TEMPLATE_YANDEX,
+    LLM_CHAT_TEMPLATE_BAILING,
+    LLM_CHAT_TEMPLATE_LLAMA4,
+    LLM_CHAT_TEMPLATE_SMOLVLM,
+    LLM_CHAT_TEMPLATE_UNKNOWN,
+};
+
+struct llama_chat_message;
+
+llm_chat_template llm_chat_template_from_str(const std::string & name);
+
+llm_chat_template llm_chat_detect_template(const std::string & tmpl);
+
+int32_t llm_chat_apply_template(
+    llm_chat_template tmpl,
+    const std::vector<const llama_chat_message *> & chat,
+    std::string & dest, bool add_ass);