Class: MiniEmbed

Inherits:

Object

• Object
• MiniEmbed

Defined in:

lib/mini_embed.rb,
ext/mini_embed/mini_embed.c

Instance Method Summary

• #embed(opts) ⇒ Object
• #embeddings(text:, type: :vector) ⇒ String, <Float>

  • type == :binary - binary string, type == :vector - array of floats.

• #initialize(opts) ⇒ Object constructor

Constructor Details

#initialize(opts) ⇒ Object

# File 'ext/mini_embed/mini_embed.c', line 1757

static VALUE rb_embedder_initialize(VALUE self, VALUE opts) {
    ruby_embedder *e;
    TypedData_Get_Struct(self, ruby_embedder, &ruby_embedder_type, e);

    Check_Type(opts, T_HASH);
    VALUE path = rb_hash_aref(opts, ID2SYM(rb_intern("model")));
    if (NIL_P(path)) rb_raise(rb_eArgError, "missing required key: model");
    const char *cpath = StringValueCStr(path);

    VALUE normalize = rb_hash_aref(opts, ID2SYM(rb_intern("normalize")));
    int norm_type = NORM_NONE;
    if (!NIL_P(normalize)) {
        if (SYMBOL_P(normalize)) {
            ID sym_id = SYM2ID(normalize);
            if (sym_id == rb_intern("l2") || sym_id == rb_intern("L2")) {
                norm_type = NORM_L2;
            }
        } else if (TYPE(normalize) == T_STRING) {
            const char *norm_str = StringValueCStr(normalize);
            if (strcasecmp(norm_str, "l2") == 0) {
                norm_type = NORM_L2;
            }
        }
    }

    e->model = embed_load_gguf(cpath);
    if (!e->model) rb_raise(rb_eRuntimeError, "failed to load GGUF model: %s", cpath);

    e->model->normalize = norm_type;
    return self;
}

Instance Method Details

#embed(opts) ⇒ Object

# File 'ext/mini_embed/mini_embed.c', line 1789

static VALUE rb_embed(VALUE self, VALUE opts) {
    ruby_embedder *e;
    TypedData_Get_Struct(self, ruby_embedder, &ruby_embedder_type, e);

    Check_Type(opts, T_HASH);
    VALUE text = rb_hash_aref(opts, ID2SYM(rb_intern("text")));
    if (NIL_P(text)) rb_raise(rb_eArgError, "missing required key: text");
    const char *ctext = StringValueCStr(text);

    VALUE out = rb_str_new(NULL, e->model->dim * sizeof(float));
    embed_text(e->model, ctext, (float*)RSTRING_PTR(out));
    return out;
}

#embeddings(text:, type: :vector) ⇒ String, <Float>

Returns - type == :binary - binary string, type == :vector - array of floats.

Parameters:

• text (String) —

  • text to extract embeddings from

• type (Symbol, nil) (defaults to: :vector) —

  • :binary or :vector - type of data you want to receive

Returns:

• (String, <Float>) —

  • type == :binary - binary string, type == :vector - array of floats

Raises:

• (ArgumentError)

# File 'lib/mini_embed.rb', line 9

def embeddings(text:, type: :vector)
  binary_data = embed(text: text) # call original C method

  return binary_data if type == :binary
  return binary_data.unpack('e*') if type == :vector

  raise ArgumentError, "Unsupported data type: #{type}"
end