seamless_communication/ggml/examples/starcoder/quantize.cpp

#include "ggml/ggml.h"

#include "common.h"
#include "common-ggml.h"

#include <cassert>
#include <cmath>
#include <cstdio>
#include <cstring>
#include <fstream>
#include <map>
#include <string>
#include <vector>
#include <regex>

// default hparams (GPT-2 117M)
struct starcoder_hparams {
    int32_t n_vocab = 49280;
    int32_t n_ctx   = 2048;
    int32_t n_embd  = 2048;
    int32_t n_head  = 16;
    int32_t n_layer = 24;
    int32_t ftype   = 1;
};

// quantize a model
bool starcoder_model_quantize(const std::string & fname_inp, const std::string & fname_out, ggml_ftype ftype) {
    gpt_vocab vocab;

    printf("%s: loading model from '%s'\n", __func__, fname_inp.c_str());

    auto finp = std::ifstream(fname_inp, std::ios::binary);
    if (!finp) {
        fprintf(stderr, "%s: failed to open '%s' for reading\n", __func__, fname_inp.c_str());
        return false;
    }

    auto fout = std::ofstream(fname_out, std::ios::binary);
    if (!fout) {
        fprintf(stderr, "%s: failed to open '%s' for writing\n", __func__, fname_out.c_str());
        return false;
    }

    // verify magic
    {
        uint32_t magic;
        finp.read((char *) &magic, sizeof(magic));
        if (magic != GGML_FILE_MAGIC) {
            fprintf(stderr, "%s: invalid model file '%s' (bad magic)\n", __func__, fname_inp.c_str());
            return false;
        }

        fout.write((char *) &magic, sizeof(magic));
    }

    starcoder_hparams hparams;

    // load hparams
    {
        finp.read((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
        finp.read((char *) &hparams.n_ctx,   sizeof(hparams.n_ctx));
        finp.read((char *) &hparams.n_embd,  sizeof(hparams.n_embd));
        finp.read((char *) &hparams.n_head,  sizeof(hparams.n_head));
        finp.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
        finp.read((char *) &hparams.ftype,   sizeof(hparams.ftype));

        const int32_t qntvr_src =    hparams.ftype / GGML_QNT_VERSION_FACTOR;
        const int32_t ftype_dst = GGML_QNT_VERSION * GGML_QNT_VERSION_FACTOR + ftype;

        printf("%s: n_vocab     = %d\n", __func__, hparams.n_vocab);
        printf("%s: n_ctx       = %d\n", __func__, hparams.n_ctx);
        printf("%s: n_embd      = %d\n", __func__, hparams.n_embd);
        printf("%s: n_head      = %d\n", __func__, hparams.n_head);
        printf("%s: n_layer     = %d\n", __func__, hparams.n_layer);
        printf("%s: ftype (src) = %d\n", __func__, hparams.ftype);
        printf("%s: qntvr (src) = %d\n", __func__, qntvr_src);
        printf("%s: ftype (dst) = %d\n", __func__, ftype_dst);
        printf("%s: qntvr (dst) = %d\n", __func__, GGML_QNT_VERSION);

        fout.write((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
        fout.write((char *) &hparams.n_ctx,   sizeof(hparams.n_ctx));
        fout.write((char *) &hparams.n_embd,  sizeof(hparams.n_embd));
        fout.write((char *) &hparams.n_head,  sizeof(hparams.n_head));
        fout.write((char *) &hparams.n_layer, sizeof(hparams.n_layer));
        fout.write((char *) &ftype_dst,       sizeof(ftype_dst));
    }

    // load vocab
    {
        int32_t n_vocab = 0;
        finp.read ((char *) &n_vocab, sizeof(n_vocab));
        fout.write((char *) &n_vocab, sizeof(n_vocab));

        if (n_vocab != hparams.n_vocab) {
            fprintf(stderr, "%s: invalid model file '%s' (bad vocab size %d != %d)\n",
                    __func__, fname_inp.c_str(), n_vocab, hparams.n_vocab);
            return false;
        }

        std::string word;
        for (int i = 0; i < n_vocab; i++) {
            uint32_t len;
            finp.read ((char *) &len, sizeof(len));
            fout.write((char *) &len, sizeof(len));

            word.resize(len);
            finp.read ((char *) word.data(), len);
            fout.write((char *) word.data(), len);

            vocab.token_to_id[word] = i;
            vocab.id_to_token[i] = word;
        }
    }

    // regexes of tensor names to be quantized
    const std::vector<std::string> to_quant = {
        "model/wte",
        "model/lm_head",
        "model/h.*/attn/c_attn/w",
        "model/h.*/attn/c_proj/w",
        "model/h.*/mlp/c_fc/w",
        "model/h.*/mlp/c_proj/w",
    };

    if (!ggml_common_quantize_0(finp, fout, ftype, to_quant, {})) {
        fprintf(stderr, "%s: failed to quantize model '%s'\n", __func__, fname_inp.c_str());
        return false;
    }

    finp.close();
    fout.close();

    return true;
}

// usage:
//  ./gpt-2-quantize models/gpt-2-117M/ggml-model.bin models/gpt-2-117M/ggml-model-quant.bin type
//
int main(int argc, char ** argv) {
    if (argc != 4) {
        fprintf(stderr, "usage: %s model-f32.bin model-quant.bin type\n", argv[0]);
        ggml_print_ftypes(stderr);
        return 1;
    }

    // needed to initialize f16 tables
    {
        struct ggml_init_params params = { 0, NULL, false };
        struct ggml_context * ctx = ggml_init(params);
        ggml_free(ctx);
    }

    const std::string fname_inp = argv[1];
    const std::string fname_out = argv[2];

    const ggml_ftype ftype = ggml_parse_ftype(argv[3]);

    const int64_t t_main_start_us = ggml_time_us();

    int64_t t_quantize_us = 0;

    // load the model
    {
        const int64_t t_start_us = ggml_time_us();

        if (!starcoder_model_quantize(fname_inp, fname_out, ggml_ftype(ftype))) {
            fprintf(stderr, "%s: failed to quantize model from '%s'\n", __func__, fname_inp.c_str());
            return 1;
        }

        t_quantize_us = ggml_time_us() - t_start_us;
    }

    // report timing
    {
        const int64_t t_main_end_us = ggml_time_us();

        printf("\n");
        printf("%s: quantize time = %8.2f ms\n", __func__, t_quantize_us/1000.0f);
        printf("%s:    total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us)/1000.0f);
    }

    return 0;
}