cui-llama.rn 1.4.6 → 1.6.0

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/speculative.h

@@ -0,0 +1,28 @@
+#pragma once
+
+#include "llama.h"
+#include "common.h"
+
+struct common_speculative;
+
+struct common_speculative_params {
+    int n_draft = 16;  // max drafted tokens
+    int n_reuse = 256;
+
+    float p_min = 0.75f; // min probability required to accept a token in the draft
+};
+
+struct common_speculative * common_speculative_init(struct llama_context * ctx_dft);
+
+void common_speculative_free(struct common_speculative * spec);
+
+bool common_speculative_are_compatible(
+        const struct llama_context * ctx_tgt,
+        const struct llama_context * ctx_dft);
+
+// sample up to n_draft tokens and add them to the batch using the draft model
+llama_tokens common_speculative_gen_draft(
+               struct common_speculative * spec,
+        struct common_speculative_params   params,
+                      const llama_tokens & prompt,
+                             llama_token   id_last);

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/unary-ops.h

@@ -0,0 +1,28 @@
+#pragma once
+
+#include "cpu-common.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void lm_ggml_compute_forward_abs(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_sgn(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_neg(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_step(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_tanh(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_elu(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_relu(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_sigmoid(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_hardsigmoid(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_exp(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_hardswish(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_sqr(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_sqrt(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_sin(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_cos(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+void lm_ggml_compute_forward_log(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
+
+#ifdef __cplusplus
+}
+#endif

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/unicode-data.h

@@ -0,0 +1,20 @@
+#pragma once
+
+#include <cstdint>
+#include <vector>
+#include <unordered_map>
+#include <unordered_set>
+
+struct range_nfd {
+    uint32_t first;
+    uint32_t last;
+    uint32_t nfd;
+};
+
+static const uint32_t MAX_CODEPOINTS = 0x110000;
+
+extern const std::initializer_list<std::pair<uint32_t, uint16_t>> unicode_ranges_flags;
+extern const std::unordered_set<uint32_t> unicode_set_whitespace;
+extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase;
+extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase;
+extern const std::initializer_list<range_nfd> unicode_ranges_nfd;

package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/unicode.h

@@ -0,0 +1,66 @@
+#pragma once
+
+#include <cstdint>
+#include <string>
+#include <vector>
+
+struct unicode_cpt_flags {
+    enum {
+        UNDEFINED       = 0x0001,
+        NUMBER          = 0x0002,  // regex: \p{N}
+        LETTER          = 0x0004,  // regex: \p{L}
+        SEPARATOR       = 0x0008,  // regex: \p{Z}
+        ACCENT_MARK     = 0x0010,  // regex: \p{M}
+        PUNCTUATION     = 0x0020,  // regex: \p{P}
+        SYMBOL          = 0x0040,  // regex: \p{S}
+        CONTROL         = 0x0080,  // regex: \p{C}
+        MASK_CATEGORIES = 0x00FF,
+    };
+
+    // codepoint type
+    uint16_t is_undefined   : 1;
+    uint16_t is_number      : 1;  // regex: \p{N}
+    uint16_t is_letter      : 1;  // regex: \p{L}
+    uint16_t is_separator   : 1;  // regex: \p{Z}
+    uint16_t is_accent_mark : 1;  // regex: \p{M}
+    uint16_t is_punctuation : 1;  // regex: \p{P}
+    uint16_t is_symbol      : 1;  // regex: \p{S}
+    uint16_t is_control     : 1;  // regex: \p{C}
+    // helper flags
+    uint16_t is_whitespace  : 1;  // regex: \s
+    uint16_t is_lowercase   : 1;
+    uint16_t is_uppercase   : 1;
+    uint16_t is_nfd         : 1;
+
+    // decode from uint16
+    inline unicode_cpt_flags(const uint16_t flags = 0) {
+        *reinterpret_cast<uint16_t*>(this) = flags;
+    }
+
+    inline uint16_t as_uint() const {
+        return *reinterpret_cast<const uint16_t*>(this);
+    }
+
+    inline uint16_t category_flag() const {
+        return this->as_uint() & MASK_CATEGORIES;
+    }
+};
+
+size_t unicode_len_utf8(char src);
+
+std::string unicode_cpt_to_utf8  (uint32_t cpt);
+uint32_t    unicode_cpt_from_utf8(const std::string & utf8, size_t & offset);
+
+std::vector<uint32_t> unicode_cpts_from_utf8(const std::string & utf8);
+
+std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & cpts);
+
+unicode_cpt_flags unicode_cpt_flags_from_cpt (uint32_t cpt);
+unicode_cpt_flags unicode_cpt_flags_from_utf8(const std::string & utf8);
+
+std::string unicode_byte_to_utf8(uint8_t byte);
+uint8_t     unicode_utf8_to_byte(const std::string & utf8);
+
+uint32_t unicode_tolower(uint32_t cpt);
+
+std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs);