Get the next input sequence for the model
| 694 | |
| 695 | // Get the next input sequence for the model |
| 696 | matrix<int, 0, 1> get_input_sequence(size_t desired_length) const { |
| 697 | if (desired_length < min_prompt_size_) |
| 698 | throw std::invalid_argument("Requested length is smaller than minimum prompt size"); |
| 699 | matrix<int, 0, 1> input_sequence(desired_length, 1); |
| 700 | |
| 701 | // Determine how many tokens we'll copy from context |
| 702 | size_t tokens_to_copy = std::min(current_context_.size(), desired_length); |
| 703 | size_t start_pos = current_context_.size() > desired_length ? |
| 704 | current_context_.size() - desired_length : 0; |
| 705 | |
| 706 | // Fill the matrix with tokens from context |
| 707 | for (size_t i = 0; i < tokens_to_copy; ++i) |
| 708 | input_sequence(i, 0) = current_context_[start_pos + i]; |
| 709 | |
| 710 | // Fill remaining positions with padding token if needed |
| 711 | for (size_t i = tokens_to_copy; i < desired_length; ++i) |
| 712 | input_sequence(i, 0) = padding_token_; |
| 713 | |
| 714 | return input_sequence; |
| 715 | } |
| 716 | |
| 717 | // Calculate maximum output tokens that can be generated |
| 718 | size_t get_max_output_tokens() const { |