llama-quant : fix --tensor-type when default qtype is overriden

[ddh0]

fix #22544 (my fault!)

Overview

Currently, when using --tensor-type "<regex>=GGML_TYPE", if the GGML_TYPE override matches the default type for the chosen output ftype, the internal heuristics in llama_tensor_get_type_impl may still take effect, rather than being locked to the specified GGML_TYPE.

This is my own mistake that I introduced in #19770.

    ggml_type new_type = default_type;

    // get more optimal quantization type based on the tensor shape, layer, etc.
    if (!params->pure && ggml_is_quantized(default_type)) {
        // if the user provided tensor types - use those
        bool manual = false;
        if (!qs.tensor_type_patterns.empty()) {
            const std::string tensor_name(tensor->name);
            for (const auto & [pattern, qtype] : qs.tensor_type_patterns) {
                if (std::regex_search(tensor_name, pattern)) {
                    if (qtype != new_type) {
                        LLAMA_LOG_WARN("%s: %-36s - applying manual override: %s -> %s\n",
                                       __func__, tensor_name.c_str(), ggml_type_name(new_type), ggml_type_name(qtype));
                        new_type = qtype;
                        manual = true; // THIS IS WRONG
                        break;         // THIS IS WRONG
                    }
                // MOVED TO HERE!
                // MOVED TO HERE!
                }
            }
        }

        // if not manual - use the standard logic for choosing the quantization type based on the selected mixture
        if (!manual) {
            new_type = llama_tensor_get_type_impl(qs, new_type, tensor, params->ftype, tm.category);
        }

Additional information

Credit to @Anai-Guo, ref #22559 - since that one was closed due to the new contributor policy I am taking the liberty of re-submitting that PR here.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: NO