Function decode_helper

tools/perplexity/perplexity.cpp:664–696

}

static bool decode_helper(llama_context * ctx, llama_batch & batch, std::vector<float> & batch_logits, int n_batch, int n_vocab) {
    int prev_outputs = 0;
    for (int i = 0; i < (int) batch.n_tokens; i += n_batch) {
        const int n_tokens = std::min<int>(n_batch, batch.n_tokens - i);

        llama_batch batch_view = {
            n_tokens,
            batch.token    + i,
            nullptr,
            batch.pos      + i,
            batch.n_seq_id + i,
            batch.seq_id   + i,
            batch.logits   + i,
        };

        const int ret = llama_decode(ctx, batch_view);
        if (ret != 0) {
            LOG_ERR("failed to decode the batch, n_batch = %d, ret = %d\n", n_batch, ret);
            return false;
        }

        int n_outputs = 0;
        for (int i = 0; i < n_tokens; ++i) {
            n_outputs += batch_view.logits[i] != 0;
        }

        memcpy(batch_logits.data() + size_t(prev_outputs)*n_vocab, llama_get_logits(ctx), size_t(n_outputs)*n_vocab*sizeof(float));

        prev_outputs += n_outputs;
    }

    return true;
}

#define K_TOKEN_CHUNK 4