MCPcopy Create free account
hub / github.com/M4THYOU/TokenDagger / encode

Method encode

src/main.cpp:50–63  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

48 std::unique_ptr<tiktoken::CoreBPE> bpe;
49
50 std::vector<int> encode(const std::string& prompt) const {
51 // std::vector<int> tokens = {MistralSpecialTokens::BOS, MistralSpecialTokens::BEGIN_INST};
52 std::vector<int> tokens = {Llama4SpecialTokens::BOS.rank};
53 tokens.reserve(prompt.size()); // some compression occurs, so shouldn't be greater than the prompt size.
54 if (bpe) {
55 // auto result = bpe->encode_ordinary(prompt);
56 auto result = bpe->encode(prompt, {Llama4SpecialTokens::BOS.content, Llama4SpecialTokens::EOS.content, Llama4SpecialTokens::FULL_EOS.content});
57 for (auto token : result.first) {
58 tokens.push_back(token);
59 }
60 }
61 // tokens.push_back(MistralSpecialTokens::END_INST);
62 return tokens;
63 }
64};
65
66NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE(SpecialToken, content);

Callers 1

TokenizeFunction · 0.45

Calls 2

reserveMethod · 0.45
sizeMethod · 0.45

Tested by

no test coverage detected