cui-llama.rn 1.1.2 → 1.1.5

package/cpp/llama-grammar.h

@@ -2,11 +2,115 @@
 
 #include "llama-impl.h"
 
+#include <map>
+
 struct llama_vocab;
-struct llama_sampling;
+
+// grammar element type
+enum llama_gretype {
+    // end of rule definition
+    LLAMA_GRETYPE_END            = 0,
+
+    // start of alternate definition for rule
+    LLAMA_GRETYPE_ALT            = 1,
+
+    // non-terminal element: reference to rule
+    LLAMA_GRETYPE_RULE_REF       = 2,
+
+    // terminal element: character (code point)
+    LLAMA_GRETYPE_CHAR           = 3,
+
+    // inverse char(s) ([^a], [^a-b] [^abc])
+    LLAMA_GRETYPE_CHAR_NOT       = 4,
+
+    // modifies a preceding LLAMA_GRETYPE_CHAR or LLAMA_GRETYPE_CHAR_ALT to
+    // be an inclusive range ([a-z])
+    LLAMA_GRETYPE_CHAR_RNG_UPPER = 5,
+
+    // modifies a preceding LLAMA_GRETYPE_CHAR or
+    // LLAMA_GRETYPE_CHAR_RNG_UPPER to add an alternate char to match ([ab], [a-zA])
+    LLAMA_GRETYPE_CHAR_ALT       = 6,
+
+    // any character (.)
+    LLAMA_GRETYPE_CHAR_ANY       = 7,
+};
+
+typedef struct llama_grammar_element {
+    enum llama_gretype type;
+    uint32_t           value; // Unicode code point or rule ID
+} llama_grammar_element;
+
+struct llama_partial_utf8 {
+    uint32_t value;    // bit value so far (unshifted)
+    int      n_remain; // num bytes remaining; -1 indicates invalid sequence
+};
+
+struct llama_grammar_candidate {
+    size_t               index;
+    const uint32_t     * code_points;
+    llama_partial_utf8   partial_utf8;
+};
+
+using llama_grammar_rule  = std::vector<      llama_grammar_element>;
+using llama_grammar_stack = std::vector<const llama_grammar_element *>;
+
+using llama_grammar_rules      = std::vector<llama_grammar_rule>;
+using llama_grammar_stacks     = std::vector<llama_grammar_stack>;
+using llama_grammar_candidates = std::vector<llama_grammar_candidate>;
+
+const llama_grammar_rules  & llama_grammar_get_rules (const struct llama_grammar * grammar);
+      llama_grammar_stacks & llama_grammar_get_stacks(      struct llama_grammar * grammar);
+
+// takes a set of possible pushdown stacks on a grammar, which are required to
+// be positioned at a character range (see `llama_grammar_advance_stack`), and
+// produces the N possible stacks if the given char is accepted at those
+// positions
+void llama_grammar_accept(
+        const llama_grammar_rules  & rules,
+        const llama_grammar_stacks & stacks,
+                          uint32_t   chr,
+              llama_grammar_stacks & stacks_new);
+
+std::vector<llama_grammar_candidate> llama_grammar_reject_candidates_for_stack(
+        const llama_grammar_rules      & rules,
+        const llama_grammar_stack      & stack,
+        const llama_grammar_candidates & candidates);
+
+struct llama_grammar_parser {
+    std::map<std::string, uint32_t> symbol_ids;
+
+    llama_grammar_rules rules;
+
+    llama_grammar_stack c_rules() const;
+
+    uint32_t get_symbol_id(const char * src, size_t len);
+    uint32_t generate_symbol_id(const std::string & base_name);
+
+    void add_rule(uint32_t rule_id, const llama_grammar_rule & rule);
+
+    const char * parse_alternates(
+            const char        * src,
+            const std::string & rule_name,
+            uint32_t            rule_id,
+            bool                is_nested);
+
+    const char * parse_sequence(
+            const char         * src,
+            const std::string  & rule_name,
+            llama_grammar_rule & rule,
+            bool               is_nested);
+
+    const char * parse_rule(const char * src);
+
+    bool parse(const char * src);
+    void print(FILE * file);
+};
 
 struct llama_grammar {
-    const llama_grammar_rules  rules;
+    // note: allow null vocab for testing (not great)
+    const llama_vocab * vocab;
+
+    const llama_grammar_rules  rules;  // TODO: shared ptr
           llama_grammar_stacks stacks;
 
     // buffer for partially generated UTF-8 sequence from accepted tokens
@@ -17,23 +121,24 @@ struct llama_grammar {
 // internal API
 //
 
+// note: needed for tests (not great)
 struct llama_grammar * llama_grammar_init_impl(
-            const llama_grammar_element ** rules,
-                                 size_t    n_rules,
-                                 size_t    start_rule_index);
+        const struct llama_vocab * vocab,
+        const llama_grammar_element ** rules,
+        size_t n_rules,
+        size_t start_rule_index);
+
+struct llama_grammar * llama_grammar_init_impl(const struct llama_vocab * vocab, const char * grammar_str, const char * grammar_root);
 
 void llama_grammar_free_impl(struct llama_grammar * grammar);
 
-struct llama_grammar * llama_grammar_copy_impl(const struct llama_grammar * grammar);
+struct llama_grammar * llama_grammar_clone_impl(const struct llama_grammar & grammar);
 
-void llama_grammar_sample_impl(
-        const struct llama_grammar * grammar,
-          const struct llama_vocab * vocab,
-       const struct llama_sampling * smpl,
-            llama_token_data_array * candidates);
+// TODO: move the API below as member functions of llama_grammar
+void llama_grammar_apply_impl(
+        const struct llama_grammar & grammar,
+            llama_token_data_array * cur_p);
 
-void llama_grammar_accept_token_impl(
-              struct llama_grammar * grammar,
-          const struct llama_vocab * vocab,
-       const struct llama_sampling * smpl,
+void llama_grammar_accept_impl(
+              struct llama_grammar & grammar,
                        llama_token   token);

package/cpp/llama-impl.h

@@ -1,8 +1,11 @@
 #pragma once
 
-#define LLAMA_API_INTERNAL
 #include "llama.h"
 
+#include <string>
+#include <vector>
+#include <stdexcept>
+
 #ifdef __GNUC__
 #ifdef __MINGW32__
 #define LLAMA_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
@@ -29,6 +32,20 @@ void llama_log_callback_default(lm_ggml_log_level level, const char * text, void
 // helpers
 //
 
+struct time_meas {
+    time_meas(int64_t & t_acc, bool disable = false) : t_start_us(disable ? -1 : lm_ggml_time_us()), t_acc(t_acc) {}
+
+    ~time_meas() {
+        if (t_start_us >= 0) {
+            t_acc += lm_ggml_time_us() - t_start_us;
+        }
+    }
+
+    const int64_t t_start_us;
+
+    int64_t & t_acc;
+};
+
 static void replace_all(std::string & s, const std::string & search, const std::string & replace) {
     if (search.empty()) {
         return;
@@ -45,3 +62,117 @@ static void replace_all(std::string & s, const std::string & search, const std::
     builder.append(s, last_pos, std::string::npos);
     s = std::move(builder);
 }
+
+const std::vector<std::pair<std::string, struct lm_ggml_tensor *>> & llama_internal_get_tensor_map(
+    struct llama_context * ctx
+);
+
+// the ring buffer works similarly to std::deque, but with a fixed capacity
+template<typename T>
+struct ring_buffer {
+    ring_buffer(size_t cap) : capacity(cap), data(cap) {}
+
+    T & front() {
+        if (sz == 0) {
+            throw std::runtime_error("ring buffer is empty");
+        }
+        return data[first];
+    }
+
+    const T & front() const {
+        if (sz == 0) {
+            throw std::runtime_error("ring buffer is empty");
+        }
+        return data[first];
+    }
+
+    T & back() {
+        if (sz == 0) {
+            throw std::runtime_error("ring buffer is empty");
+        }
+        return data[pos];
+    }
+
+    const T & back() const {
+        if (sz == 0) {
+            throw std::runtime_error("ring buffer is empty");
+        }
+        return data[pos];
+    }
+
+    void push_back(const T & value) {
+        if (capacity == 0) {
+            throw std::runtime_error("ring buffer: capacity is zero");
+        }
+
+        if (sz == capacity) {
+            // advance the start when buffer is full
+            first = (first + 1) % capacity;
+        } else {
+            sz++;
+        }
+        data[pos] = value;
+        pos = (pos + 1) % capacity;
+    }
+
+    T pop_front() {
+        if (sz == 0) {
+            throw std::runtime_error("ring buffer is empty");
+        }
+        T value = data[first];
+        first = (first + 1) % capacity;
+        sz--;
+        return value;
+    }
+
+    //T & operator[](size_t i) {
+    //    if (i >= sz) {
+    //        throw std::runtime_error("ring buffer: index out of bounds");
+    //    }
+    //    return data[(first + i) % capacity];
+    //}
+
+    //const T & at(size_t i) const {
+    //    if (i >= sz) {
+    //        throw std::runtime_error("ring buffer: index out of bounds");
+    //    }
+    //    return data[(first + i) % capacity];
+    //}
+
+    const T & rat(size_t i) const {
+        if (i >= sz) {
+            throw std::runtime_error("ring buffer: index out of bounds");
+        }
+        return data[(first + sz - i - 1) % capacity];
+    }
+
+    std::vector<T> to_vector() const {
+        std::vector<T> result;
+        result.reserve(sz);
+        for (size_t i = 0; i < sz; i++) {
+            result.push_back(data[(first + i) % capacity]);
+        }
+        return result;
+    }
+
+    void clear() {
+        // here only reset the status of the buffer
+        sz = 0;
+        first = 0;
+        pos = 0;
+    }
+
+    bool empty() const {
+        return sz == 0;
+    }
+
+    size_t size() const {
+        return sz;
+    }
+
+    size_t capacity = 0;
+    size_t sz = 0;
+    size_t first = 0;
+    size_t pos = 0;
+    std::vector<T> data;
+};