Skip to content

Commit

Permalink
ggml : add GGML_QNT_VERSION for tracking changes to the quantization …
Browse files Browse the repository at this point in the history 
…format

ref #150
  • Loading branch information
@ggerganov
ggerganov committed May 14, 2023
1 parent fa8665f commit effcfa6
Show file tree
Hide file tree
Showing 13 changed files with 115 additions and 46 deletions.
5 changes: 5 additions & 0 deletions examples/dolly-v2/main.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -114,13 +114,18 @@ bool dollyv2_model_load(const std::string & fname, dollyv2_model & model, gpt_vo
fin.read((char *) &hparams.n_rot, sizeof(hparams.n_rot));
fin.read((char *) &hparams.ftype, sizeof(hparams.ftype));

const int32_t qntvr = hparams.ftype / GGML_QNT_VERSION_FACTOR;

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: n_rot = %d\n", __func__, hparams.n_rot);
printf("%s: ftype = %d\n", __func__, hparams.ftype);
printf("%s: qntvr = %d\n", __func__, qntvr);

hparams.ftype %= GGML_QNT_VERSION_FACTOR;
}

// load vocab
Expand Down
20 changes: 13 additions & 7 deletions examples/dolly-v2/quantize.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -66,20 +66,26 @@ bool stablelm_model_quantize(const std::string & fname_inp, const std::string &
finp.read((char *) &hparams.n_rot, sizeof(hparams.n_rot));
finp.read((char *) &hparams.ftype, sizeof(hparams.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 = %d\n", __func__, 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 *) &hparams.n_rot, sizeof(hparams.n_rot));
fout.write((char *) &ftype, sizeof(hparams.ftype));
fout.write((char *) &ftype_dst, sizeof(ftype_dst));
}

// load vocab
Expand Down
5 changes: 5 additions & 0 deletions examples/gpt-2/main.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -100,12 +100,17 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
fin.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
fin.read((char *) &hparams.ftype, sizeof(hparams.ftype));

const int32_t qntvr = hparams.ftype / GGML_QNT_VERSION_FACTOR;

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 = %d\n", __func__, hparams.ftype);
printf("%s: qntvr = %d\n", __func__, qntvr);

hparams.ftype %= GGML_QNT_VERSION_FACTOR;
}

// load vocab
Expand Down
24 changes: 15 additions & 9 deletions examples/gpt-2/quantize.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,7 @@ struct gpt2_hparams {
int32_t n_embd = 768;
int32_t n_head = 12;
int32_t n_layer = 12;
int32_t f16 = 1;
int32_t ftype = 1;
};

// quantize a model
Expand Down Expand Up @@ -62,21 +62,27 @@ bool gpt2_model_quantize(const std::string & fname_inp, const std::string & fnam
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.f16, sizeof(hparams.f16));
finp.read((char *) &hparams.ftype, sizeof(hparams.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: f16 = %d\n", __func__, hparams.f16);
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, sizeof(hparams.f16));
fout.write((char *) &ftype_dst, sizeof(ftype_dst));
}

// load vocab
Expand Down
5 changes: 5 additions & 0 deletions examples/gpt-j/main.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -100,13 +100,18 @@ bool gptj_model_load(const std::string & fname, gptj_model & model, gpt_vocab &
fin.read((char *) &hparams.n_rot, sizeof(hparams.n_rot));
fin.read((char *) &hparams.ftype, sizeof(hparams.ftype));

const int32_t qntvr = hparams.ftype / GGML_QNT_VERSION_FACTOR;

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: n_rot = %d\n", __func__, hparams.n_rot);
printf("%s: ftype = %d\n", __func__, hparams.ftype);
printf("%s: qntvr = %d\n", __func__, qntvr);

hparams.ftype %= GGML_QNT_VERSION_FACTOR;
}

// load vocab
Expand Down
24 changes: 15 additions & 9 deletions examples/gpt-j/quantize.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@ struct gptj_hparams {
int32_t n_head = 16;
int32_t n_layer = 28;
int32_t n_rot = 64;
int32_t f16 = 1;
int32_t ftype = 1;
};

// quantize a model
Expand Down Expand Up @@ -64,22 +64,28 @@ bool gptj_model_quantize(const std::string & fname_inp, const std::string & fnam
finp.read((char *) &hparams.n_head, sizeof(hparams.n_head));
finp.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
finp.read((char *) &hparams.n_rot, sizeof(hparams.n_rot));
finp.read((char *) &hparams.f16, sizeof(hparams.f16));
finp.read((char *) &hparams.ftype, sizeof(hparams.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: f16 = %d\n", __func__, hparams.f16);
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 *) &hparams.n_rot, sizeof(hparams.n_rot));
fout.write((char *) &ftype, sizeof(hparams.f16));
fout.write((char *) &ftype_dst, sizeof(ftype_dst));
}

// load vocab
Expand Down
5 changes: 5 additions & 0 deletions examples/gpt-neox/main.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,8 @@ bool gpt_neox_model_load(const std::string & fname, gpt_neox_model & model, gpt_
fin.read((char *) &hparams.par_res, sizeof(hparams.par_res));
fin.read((char *) &hparams.ftype, sizeof(hparams.ftype));

const int32_t qntvr = hparams.ftype / GGML_QNT_VERSION_FACTOR;

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);
Expand All @@ -114,6 +116,9 @@ bool gpt_neox_model_load(const std::string & fname, gpt_neox_model & model, gpt_
printf("%s: n_rot = %d\n", __func__, hparams.n_rot);
printf("%s: par_res = %d\n", __func__, hparams.par_res);
printf("%s: ftype = %d\n", __func__, hparams.ftype);
printf("%s: qntvr = %d\n", __func__, qntvr);

hparams.ftype %= GGML_QNT_VERSION_FACTOR;
}

// load vocab
Expand Down
22 changes: 14 additions & 8 deletions examples/gpt-neox/quantize.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -68,13 +68,19 @@ bool gpt_neox_model_quantize(const std::string & fname_inp, const std::string &
finp.read((char *) &hparams.par_res, sizeof(hparams.par_res));
finp.read((char *) &hparams.ftype, sizeof(hparams.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: par_res = %d\n", __func__, hparams.par_res);
printf("%s: ftype = %d\n", __func__, 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: par_res = %d\n", __func__, hparams.par_res);
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));
Expand All @@ -83,7 +89,7 @@ bool gpt_neox_model_quantize(const std::string & fname_inp, const std::string &
fout.write((char *) &hparams.n_layer, sizeof(hparams.n_layer));
fout.write((char *) &hparams.n_rot, sizeof(hparams.n_rot));
fout.write((char *) &hparams.par_res, sizeof(hparams.par_res));
fout.write((char *) &ftype, sizeof(hparams.ftype));
fout.write((char *) &ftype_dst, sizeof(ftype_dst));
}

// load vocab
Expand Down
5 changes: 5 additions & 0 deletions examples/starcoder/main.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -101,12 +101,17 @@ bool starcoder_model_load(const std::string & fname, starcoder_model & model, gp
fin.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
fin.read((char *) &hparams.ftype, sizeof(hparams.ftype));

const int32_t qntvr = hparams.ftype / GGML_QNT_VERSION_FACTOR;

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 = %d\n", __func__, hparams.ftype);
printf("%s: qntvr = %d\n", __func__, qntvr);

hparams.ftype %= GGML_QNT_VERSION_FACTOR;
}

// load vocab
Expand Down
24 changes: 15 additions & 9 deletions examples/starcoder/quantize.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,7 @@ struct starcoder_hparams {
int32_t n_embd = 2048;
int32_t n_head = 16;
int32_t n_layer = 24;
int32_t f16 = 1;
int32_t ftype = 1;
};

// quantize a model
Expand Down Expand Up @@ -62,21 +62,27 @@ bool starcoder_model_quantize(const std::string & fname_inp, const std::string &
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.f16, sizeof(hparams.f16));
finp.read((char *) &hparams.ftype, sizeof(hparams.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: f16 = %d\n", __func__, hparams.f16);
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, sizeof(hparams.f16));
fout.write((char *) &ftype_dst, sizeof(ftype_dst));
}

// load vocab
Expand Down
14 changes: 10 additions & 4 deletions examples/whisper/quantize.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ struct whisper_hparams {
int32_t n_text_head = 6;
int32_t n_text_layer = 4;
int32_t n_mels = 80;
int32_t f16 = 1;
int32_t ftype = 1;
};

struct whisper_filters {
Expand Down Expand Up @@ -79,7 +79,10 @@ bool whisper_model_quantize(const std::string & fname_inp, const std::string & f
finp.read((char *) &hparams.n_text_head, sizeof(hparams.n_text_head));
finp.read((char *) &hparams.n_text_layer, sizeof(hparams.n_text_layer));
finp.read((char *) &hparams.n_mels, sizeof(hparams.n_mels));
finp.read((char *) &hparams.f16, sizeof(hparams.f16));
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;

fprintf(stderr, "%s: n_vocab = %d\n", __func__, hparams.n_vocab);
fprintf(stderr, "%s: n_audio_ctx = %d\n", __func__, hparams.n_audio_ctx);
Expand All @@ -91,7 +94,10 @@ bool whisper_model_quantize(const std::string & fname_inp, const std::string & f
fprintf(stderr, "%s: n_text_head = %d\n", __func__, hparams.n_text_head);
fprintf(stderr, "%s: n_text_layer = %d\n", __func__, hparams.n_text_layer);
fprintf(stderr, "%s: n_mels = %d\n", __func__, hparams.n_mels);
fprintf(stderr, "%s: f16 = %d\n", __func__, hparams.f16);
fprintf(stderr, "%s: ftype (src) = %d\n", __func__, hparams.ftype);
fprintf(stderr, "%s: qntvr (src) = %d\n", __func__, qntvr_src);
fprintf(stderr, "%s: ftype (dst) = %d\n", __func__, ftype_dst);
fprintf(stderr, "%s: qntvr (dst) = %d\n", __func__, GGML_QNT_VERSION);

fout.write((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
fout.write((char *) &hparams.n_audio_ctx, sizeof(hparams.n_audio_ctx));
Expand All @@ -103,7 +109,7 @@ bool whisper_model_quantize(const std::string & fname_inp, const std::string & f
fout.write((char *) &hparams.n_text_head, sizeof(hparams.n_text_head));
fout.write((char *) &hparams.n_text_layer, sizeof(hparams.n_text_layer));
fout.write((char *) &hparams.n_mels, sizeof(hparams.n_mels));
fout.write((char *) &ftype, sizeof(hparams.f16));
fout.write((char *) &ftype, sizeof(hparams.ftype));
}

// load mel filters
Expand Down
5 changes: 5 additions & 0 deletions examples/whisper/whisper.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -861,6 +861,8 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con
model.type = e_model::MODEL_LARGE;
}

hparams.ftype %= GGML_QNT_VERSION_FACTOR;

// for the big tensors, we have the option to store the data in 16-bit floats or quantized
// in order to save memory and also to speed up the computation
wctx.wtype = ggml_ftype_to_ggml_type((ggml_ftype) (model.hparams.ftype));
Expand All @@ -871,6 +873,8 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con

const size_t scale = model.hparams.ftype ? 1 : 2;

const int32_t qntvr = hparams.ftype / GGML_QNT_VERSION_FACTOR;

fprintf(stderr, "%s: n_vocab = %d\n", __func__, hparams.n_vocab);
fprintf(stderr, "%s: n_audio_ctx = %d\n", __func__, hparams.n_audio_ctx);
fprintf(stderr, "%s: n_audio_state = %d\n", __func__, hparams.n_audio_state);
Expand All @@ -882,6 +886,7 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con
fprintf(stderr, "%s: n_text_layer = %d\n", __func__, hparams.n_text_layer);
fprintf(stderr, "%s: n_mels = %d\n", __func__, hparams.n_mels);
fprintf(stderr, "%s: ftype = %d\n", __func__, model.hparams.ftype);
fprintf(stderr, "%s: qntvr = %d\n", __func__, qntvr);
fprintf(stderr, "%s: type = %d\n", __func__, model.type);

// print memory requirements
Expand Down
3 changes: 3 additions & 0 deletions include/ggml/ggml.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -190,6 +190,9 @@
#define GGML_FILE_MAGIC 0x67676d6c // "ggml"
#define GGML_FILE_VERSION 1

#define GGML_QNT_VERSION 0 // bump this on quantization format changes
#define GGML_QNT_VERSION_FACTOR 1000 // do not change this

#define GGML_MAX_DIMS 4
#define GGML_MAX_NODES 4096
#define GGML_MAX_PARAMS 16
Expand Down

0 comments on commit effcfa6

Please sign in to comment.