vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Cleanup

Browse source
This commit is contained in:
joshcarp 2024-05-07 14:44:59 -04:00
parent 308c817af4
commit 92ff0de243
3 changed files with 0 additions and 177 deletions
Download patch file Download diff file Expand all files Collapse all files

8
convert-hf-to-gguf.py
View file

@ -2916,12 +2916,8 @@ class OpenELM(Model):
head_dim = self.find_hparam(["head_dim"])
n_head = n_embd // head_dim
rot_pct = 1.0
self.gguf_writer.add_context_length(self.find_hparam(["max_context_length"]))
# self.gguf_writer.add_embedding_length(n_embd)
self.gguf_writer.add_block_count(self.block_count)
# self.gguf_writer.add_head_count(n_head)
# self.gguf_writer.add_head_count_kv(n_head)
self.gguf_writer.add_head_count_kv(n_head*10)
self.gguf_writer.add_head_count(n_head*10)
self.gguf_writer.add_rope_dimension_count(int(rot_pct * n_embd) // n_head)
@ -2979,8 +2975,6 @@ class OpenELM(Model):
def write_tensors(self):
block_count = self.hparams.get("num_transformer_layers", self.hparams.get("num_hidden_layers", self.hparams.get("num_transformer_layers")))
tensor_map = gguf.get_tensor_name_map(self.model_arch, block_count)
n_head = self.hparams.get("model_dim") // self.hparams.get("head_dim") # TODO: propagate this
foobar = {}
for name, data_torch in self.get_tensors():
old_dtype = data_torch.dtype
# convert any unsupported data types to float32
@ -3006,8 +3000,6 @@ class OpenELM(Model):
data = data.astype(np.float16)
print(f"{new_name}, n_dims = {n_dims}, {old_dtype} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data)
foobar[name] = (data_torch, new_name, data)
foobar
###### CONVERSION LOGIC ######

158
examples/eval-callback/eval-callback.cpp
View file

@ -38,115 +38,6 @@ static std::string ggml_nb_string(const ggml_tensor * t) {
return str;
}
static void ggml_print_tensor(uint8_t * data, ggml_type type, const int64_t * ne, const size_t * nb, int64_t n) {
GGML_ASSERT(n > 0);
float sum = 0;
for (int64_t i0 = 0; i0 < 3; i0++) {
if (i0 == n && ne[0] > 2*n) {
printf("..., ");
i0 = ne[0] - n;
}
size_t i = i0;//i3 * nb[3] + i2 * nb[2] + i1 * nb[1] + i0 * nb[0];
float v;
if (type == GGML_TYPE_F16) {
v = ggml_fp16_to_fp32(*(ggml_fp16_t *) data + i);
} else if (type == GGML_TYPE_F32) {
v = *(float *) data + i;
} else if (type == GGML_TYPE_I32) {
v = (float) *((int32_t *) data + i);
} else if (type == GGML_TYPE_I16) {
v = (float) *(int16_t *) data + i;
} else if (type == GGML_TYPE_I8) {
v = (float) *(int8_t *) data + i;
} else {
GGML_ASSERT(false);
}
printf("%12.4f", v);
sum += v;
}
printf("\n");
// for (int64_t i3 = 0; i3 < ne[3]; i3++) {
// printf(" [\n");
// for (int64_t i2 = 0; i2 < ne[2]; i2++) {
// if (i2 == n && ne[2] > 2*n) {
// printf(" ..., \n");
// i2 = ne[2] - n;
// }
// printf(" [\n");
// for (int64_t i1 = 0; i1 < ne[1]; i1++) {
// if (i1 == n && ne[1] > 2*n) {
// printf(" ..., \n");
// i1 = ne[1] - n;
// }
// printf(" [");
// for (int64_t i0 = 0; i0 < ne[0]; i0++) {
// if (i0 == n && ne[0] > 2*n) {
// printf("..., ");
// i0 = ne[0] - n;
// }
// size_t i = i0;//i3 * nb[3] + i2 * nb[2] + i1 * nb[1] + i0 * nb[0];
// float v;
// if (type == GGML_TYPE_F16) {
// v = ggml_fp16_to_fp32(*(ggml_fp16_t *) data + i);
// } else if (type == GGML_TYPE_F32) {
// v = *(float *) data + i;
// } else if (type == GGML_TYPE_I32) {
// v = (float) *((int32_t *) data + i);
// } else if (type == GGML_TYPE_I16) {
// v = (float) *(int16_t *) data + i;
// } else if (type == GGML_TYPE_I8) {
// v = (float) *(int8_t *) data + i;
// } else {
// GGML_ASSERT(false);
// }
// printf("%12.4f", v);
// sum += v;
// if (i0 < ne[0] - 1) printf(", ");
// }
// printf("],\n");
// }
// printf(" ],\n");
// }
// printf(" ]\n");
// printf(" sum = %f\n", sum);
// }
}
float Sum(float *arr, int64_t N){
float s = 0.0;
for (int i = 0; i < N; i++){
s += arr[i];
}
return s;
}
float PrintArr(const char * name, float * arr, int64_t N){
float sum = 0.0;
if (arr != NULL){
sum = Sum(arr, N);
printf("%s %d %10f \n",name, N, sum);
} else {
printf("%s %d %10f \n",name, 0, 0.0);
}
return sum;
}
size_t get_nth_element(const int64_t *ne, const size_t *nb, int64_t nth) {
size_t offset = 0;
size_t divisor = 1;
for (int i = 3; i >= 0; --i) {
size_t index = size_t(floor(nth / divisor)) % ne[i];
offset += index * nb[i]/4;
divisor *= ne[i];
}
return offset;
}
void print_tensor(const ggml_tensor * src0) {
float sum = 0;
@ -164,8 +55,6 @@ void print_tensor(const ggml_tensor * src0) {
if (i == n) {
buf2 += sprintf(buf2, "..., ");
}
// int64_t offset = get_nth_element(src0->ne, src0->nb, i);
// offset *= ggml_element_size(src0);
int64_t offset = i;
float v;
if (type == GGML_TYPE_F16) {
@ -186,17 +75,6 @@ void print_tensor(const ggml_tensor * src0) {
}
sum += v;
}
int i = 0;
while (i < ggml_nbytes(src0)/4){
float val = (((float *) src0->data)[i]);
float diff = abs(val - 0.0022226818837225437164306640625);
if (diff < 0.000001 ){
printf("found %s: %d = %f\n", src0->name, i, val);
}
i += 1;
}
int max_name_length = 15;
int max_dim_length = 15;
int max_str_length = 15;
@ -206,10 +84,6 @@ void print_tensor(const ggml_tensor * src0) {
max_dim_length, ggml_ne_string(src0).c_str(),
max_str_length, ggml_nb_string(src0).c_str(),
src0->data);
// printf("%s\n", buf);
}
/**
@ -231,50 +105,19 @@ static bool ggml_debug(struct ggml_tensor * t, bool ask, void * user_data) {
if (ask) {
return true; // Always retrieve data
}
char src1_str[128] = {0};
// if (src1) {
// sprintf(src1_str, "%s{%s}\n", src1->name, ggml_ne_string(src1).c_str());
// }
if (src0) {
print_tensor(src0);
// printf("%s{%s} n=%d %f\n", src0->name, ggml_ne_string(src0).c_str(),src0->ne[0], Sum(static_cast<float *>(src0->data), src0->ne[0]));
// printf("%s{%s}", src0->name, ggml_ne_string(src0).c_str());
// enum ggml_type type = src0->name == "inp_tokens" ? GGML_TYPE_I32:src0->type;
// ggml_print_tensor(static_cast<uint8_t *>(src0->data), src0->type, src0->ne, src0->nb, 3);
// PrintArr(src0->name, static_cast<float *>(src0->data), src0->ne[0]);
}
if (src1) {
print_tensor(src1);
// printf("%s{%s} n=%d %f\n", src1->name, ggml_ne_string(src1).c_str(),src0->ne[0], Sum(static_cast<float *>(src1->data), src1->ne[0]));
// enum ggml_type type = src1->name == "inp_tokens" ? GGML_TYPE_I32:src1->type;
// ggml_print_tensor(static_cast<uint8_t *>(src1->data), type, src1->ne, src1->nb, 3);
// ggml_print_tensor(static_cast<uint8_t *>(src1->data), src1->type, src1->ne, src1->nb, 3);
// PrintArr(src1->name, static_cast<float *>(src1->data), src1->ne[0]);
}
printf("%s ==\n", ggml_op_desc(t));
if (t) {
print_tensor(t);
// printf("%s{%s} n=%d %f\n", t->name, ggml_ne_string(t).c_str(),src0->ne[0], Sum(static_cast<float *>(t->data), t->ne[0]));
// printf("%s{%s}", t->name, ggml_ne_string(t).c_str());
// PrintArr(t->name, static_cast<float *>(t->data), t->ne[0]);
// ggml_print_tensor(static_cast<uint8_t *>(t->data), t->type, t->ne, t->nb, 3);
// printf("\n == \n");
}
printf("\n\n");
// printf("%24s = (%s) %10s(%s{%s}, %s}) = {%s}\n",
//
//
// t->name, ggml_op_desc(t), src0->name, ggml_ne_string(src0).c_str(),
// src1 ? src1_str : "",
// ggml_ne_string(t).c_str());
// copy the data from the GPU memory if needed
const bool is_host = ggml_backend_buffer_is_host(t->buffer);
@ -286,7 +129,6 @@ static bool ggml_debug(struct ggml_tensor * t, bool ask, void * user_data) {
if (!ggml_is_quantized(t->type)) {
uint8_t * data = is_host ? (uint8_t *) t->data : cb_data->data.data();
// ggml_print_tensor(data, t->type, t->ne, t->nb, 3);
}
return true;

11
llama.cpp
View file

@ -6955,17 +6955,6 @@ struct llm_build_context {
return lctx.inp_KQ_mask;
}
struct ggml_tensor * build_inp_KQ_mask2(int64_t n_kv, bool causal = true) {
if (causal) {
lctx.inp_KQ_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv, n_tokens);
} else {
lctx.inp_KQ_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, n_tokens);
}
cb(lctx.inp_KQ_mask, "KQ_mask", -1);
ggml_set_input(lctx.inp_KQ_mask);
return lctx.inp_KQ_mask;
}
struct ggml_tensor * build_inp_KQ_pos() {
lctx.inp_KQ_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, n_kv);
cb(lctx.inp_KQ_pos, "KQ_pos", -1);