Decided to find it for you. You're welcome.

See common/fit.cpp:

    // step 3: iteratively fill the back to front with "dense" layers
    //   - for a dense model simply fill full layers, giving each device a contiguous slice of the model
    //   - for a MoE model, same as dense model but with all MoE tensors in system memory

    // utility function that returns a static C string matching the tensors for a specific layer index and layer fraction:
    auto get_overflow_pattern = [&](const size_t il, const common_layer_fraction_t lf) -> const char * {
        constexpr size_t n_strings = 1000;
        if (il >= n_strings) {
            throw std::runtime_error("at most " + std::to_string(n_strings) + " model layers are supported");
        }
        switch (lf) {
            case LAYER_FRACTION_ATTN: {
                static std::array<std::string, n_strings> patterns;
                if (patterns[il].empty()) {
                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_(gate|up|gate_up|down).*";
                }
                return patterns[il].c_str();
            }
            case LAYER_FRACTION_UP: {
                static std::array<std::string, n_strings> patterns;
                if (patterns[il].empty()) {
                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_(gate|gate_up|down).*";
                }
                return patterns[il].c_str();
            }
            case LAYER_FRACTION_GATE: {
                static std::array<std::string, n_strings> patterns;
                if (patterns[il].empty()) {
                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_down.*";
                }
                return patterns[il].c_str();
            }
            case LAYER_FRACTION_MOE: {
                static std::array<std::string, n_strings> patterns;
                if (patterns[il].empty()) {
                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_(up|down|gate_up|gate)_(ch|)exps";
                }
                return patterns[il].c_str();
            }
            default:
                GGML_ABORT("fatal error");
        }
    };

I'm not arguing. I'm telling. I wrote a tool myself to deduce optimal -ot args before replacing it with -fit after llama added it. We're not on the same level. You use tools. I make them. I don't debate people below me. You can take the info from your better or not. I'm not wasting more time on you.