Function llama_sampler_init_dry

smallthinker/src/llama-sampling.cpp:2138–2183

};

struct llama_sampler * llama_sampler_init_dry(const struct llama_vocab * vocab, int32_t context_size, float dry_multiplier, float dry_base, int32_t dry_allowed_length, int32_t dry_penalty_last_n, const char** seq_breakers, size_t num_breakers) {
    int32_t effective_dry_penalty_last_n = (dry_penalty_last_n == -1) ? context_size : std::max(dry_penalty_last_n, 0);
    std::unordered_multimap<llama_token, std::vector<llama_token>> processed_breakers;
    const int MAX_CHAR_LEN = 40;
    const int MAX_SEQ_LEN = 20;

    const bool dry_enabled = (dry_multiplier != 0.0f && dry_base >= 1.0f && dry_penalty_last_n != 0);

    if (dry_enabled && seq_breakers != nullptr && num_breakers > 0) {
        // Process sequence breakers
        for (size_t i = 0; i < num_breakers; ++i) {
            if (seq_breakers[i] == nullptr || std::strlen(seq_breakers[i]) == 0) {
                LLAMA_LOG_WARN("skipping null or empty DRY sequence breaker at index %zu\n", i);
                continue;
            }

            std::string sequence_break(seq_breakers[i]);
            if (sequence_break.empty()) {
                LLAMA_LOG_WARN("skipping empty DRY sequence breaker\n");
                continue;
            }

            if (sequence_break.size() > MAX_CHAR_LEN) {
                LLAMA_LOG_WARN("truncating DRY sequence breaker to %d characters\n", MAX_CHAR_LEN);
                sequence_break.resize(MAX_CHAR_LEN);
            }

            get_overlapping_token_sequences(*vocab, sequence_break, processed_breakers, MAX_SEQ_LEN);
        }
    }

    return llama_sampler_init(
        /* .iface = */ &llama_sampler_dry_i,
        /* .ctx   = */ new llama_sampler_dry {
            /* .total_context_size     = */ context_size,
            /* .dry_multiplier         = */ dry_multiplier,
            /* .dry_base               = */ dry_base,
            /* .dry_allowed_length     = */ dry_allowed_length,
            /* .dry_penalty_last_n     = */ dry_penalty_last_n,
            /* .dry_processed_breakers = */ std::move(processed_breakers),
            /* .dry_repeat_count       = */ dry_enabled ? std::vector<int>(effective_dry_penalty_last_n, 0) : std::vector<int>{},
            /* .dry_max_token_repeat   = */ {},
            /* .last_tokens            = */ dry_enabled ? ring_buffer<llama_token>(effective_dry_penalty_last_n) : ring_buffer<llama_token>(0),
        }
    );
}

// wrapper for test-sampling.cpp
struct llama_sampler * llama_sampler_init_dry_testing(int32_t context_size, float dry_multiplier, float dry_base, int32_t dry_allowed_length, int32_t dry_penalty_last_n, const std::vector<std::vector<llama_token>>& seq_breakers) {