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

uppdate: remove use_awq arg

Browse source
This commit is contained in:
namtranase 2023-12-25 17:13:06 +07:00
parent 2187a8debe
commit 13f60c417d
4 changed files with 24 additions and 75 deletions
Download patch file Download diff file Expand all files Collapse all files

4
common/common.cpp
View file

@ -149,8 +149,6 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
break;
}
params.seed = std::stoul(argv[i]);
} else if (arg == "--use-awq") {
params.use_awq = true;
} else if (arg == "-t" || arg == "--threads") {
if (++i >= argc) {
invalid_param = true;
@ -811,7 +809,6 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
printf(" (can be specified more than once for multiple prompts).\n");
printf(" --color colorise output to distinguish prompt and user input from generations\n");
printf(" -s SEED, --seed SEED RNG seed (default: -1, use random seed for < 0)\n");
printf(" --use-awq Using AWQ quantization model in inferences\n");
printf(" -t N, --threads N number of threads to use during generation (default: %d)\n", params.n_threads);
printf(" -tb N, --threads-batch N\n");
printf(" number of threads to use during batch and prompt processing (default: same as --threads)\n");
@ -1021,7 +1018,6 @@ struct llama_model_params llama_model_params_from_gpt_params(const gpt_params &
mparams.tensor_split = params.tensor_split;
mparams.use_mmap = params.use_mmap;
mparams.use_mlock = params.use_mlock;
mparams.use_awq = params.use_awq;
if (params.kv_overrides.empty()) {
mparams.kv_overrides = NULL;
} else {

1
common/common.h
View file

@ -125,7 +125,6 @@ struct gpt_params {
bool infill = false; // use infill mode
bool dump_kv_cache = false; // dump the KV cache contents for debugging purposes
bool no_kv_offload = false; // disable KV offloading
bool use_awq = false; // use AWQ quantization infer
std::string cache_type_k = "f16"; // KV cache data type for the K
std::string cache_type_v = "f16"; // KV cache data type for the V

93
llama.cpp
View file

@ -1261,7 +1261,6 @@ struct llama_hparams {
float f_clamp_kqv;
float f_max_alibi_bias;
bool use_awq;
bool operator!=(const llama_hparams & other) const {
if (this->vocab_only != other.vocab_only) return true;
@ -3478,9 +3477,9 @@ static bool llm_load_tensors(
layer.ffn_down = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, backend_split);
layer.ffn_up = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, backend_split);
if (model.hparams.use_awq) {
layer.ffn_act = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_ACT, "scales", i), {n_ff}, backend);
}
// AWQ ScaleActivation layer
layer.ffn_act = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_ACT, "scales", i), {n_ff}, backend, false);
}
} break;
case LLM_ARCH_STABLELM:
@ -3754,7 +3753,6 @@ static int llama_model_load(const std::string & fname, llama_model & model, cons
llama_model_loader ml(fname, params.use_mmap, params.kv_overrides);
model.hparams.vocab_only = params.vocab_only;
model.hparams.use_awq = params.use_awq;
llm_load_arch (ml, model);
llm_load_hparams(ml, model);
@ -3800,7 +3798,6 @@ enum llm_rope_type {
enum llm_ffn_op_type {
LLM_FFN_SILU,
LLM_FFN_GELU,
LLM_FFN_GELU_ACT,
LLM_FFN_RELU,
LLM_FFN_RELU_SQR,
};
@ -3968,6 +3965,7 @@ static struct ggml_tensor * llm_build_ffn(
struct ggml_tensor * gate_b,
struct ggml_tensor * down,
struct ggml_tensor * down_b,
struct ggml_tensor * act_scales,
llm_ffn_op_type type_op,
llm_ffn_gate_type type_gate,
const llm_build_cb & cb,
@ -4012,6 +4010,10 @@ static struct ggml_tensor * llm_build_ffn(
{
cur = ggml_gelu(ctx, cur);
cb(cur, "ffn_gelu", il);
if (act_scales != NULL) {
cur = ggml_div(ctx, cur, act_scales);
cb(cur, "ffn_act", il);
}
} break;
case LLM_FFN_RELU:
{
@ -4045,55 +4047,6 @@ static struct ggml_tensor * llm_build_ffn(
return cur;
}
static struct ggml_tensor * llm_build_ffn_mpt_awq(
struct ggml_context * ctx,
struct ggml_tensor * cur,
struct ggml_tensor * up,
struct ggml_tensor * up_b,
struct ggml_tensor * down,
struct ggml_tensor * down_b,
struct ggml_tensor * act_scales,
llm_ffn_op_type type_op,
llm_ffn_gate_type type_gate,
const llm_build_cb & cb,
int il) {
struct ggml_tensor * tmp = ggml_mul_mat(ctx, up, cur);
cb(tmp, "ffn_up", il);
if (up_b) {
tmp = ggml_add(ctx, tmp, up_b);
cb(tmp, "ffn_up_b", il);
}
cur = tmp;
switch (type_op) {
case LLM_FFN_GELU_ACT:
{
cur = ggml_gelu(ctx, cur);
cb(cur, "ffn_relu", il);
cur = ggml_div(ctx, cur, act_scales);
cb(cur, "ffn_div(gelu)", il);
} break;
}
if (type_gate == LLM_FFN_PAR) {
cur = ggml_mul(ctx, cur, tmp);
cb(cur, "ffn_gate_par", il);
}
cur = ggml_mul_mat(ctx, down, cur);
if (down_b) {
cb(cur, "ffn_down", il);
}
if (down_b) {
cur = ggml_add(ctx, cur, down_b);
}
return cur;
}
// if max_alibi_bias > 0 then apply ALiBi
static struct ggml_tensor * llm_build_kqv(
struct ggml_context * ctx,
@ -4379,6 +4332,7 @@ struct llm_build_context {
model.layers[il].ffn_up, NULL,
model.layers[il].ffn_gate, NULL,
model.layers[il].ffn_down, NULL,
NULL,
LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
cb(cur, "ffn_out", il);
} else {
@ -4558,6 +4512,7 @@ struct llm_build_context {
model.layers[il].ffn_up, NULL,
model.layers[il].ffn_gate, NULL,
model.layers[il].ffn_down, NULL,
NULL,
LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
cb(cur, "ffn_out", il);
}
@ -4672,6 +4627,7 @@ struct llm_build_context {
model.layers[il].ffn_up, NULL,
NULL, NULL,
model.layers[il].ffn_down, NULL,
NULL,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il);
}
@ -4776,6 +4732,7 @@ struct llm_build_context {
model.layers[il].ffn_up, model.layers[il].ffn_up_b,
NULL, NULL,
model.layers[il].ffn_down, model.layers[il].ffn_down_b,
NULL,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il);
}
@ -4980,6 +4937,7 @@ struct llm_build_context {
model.layers[il].ffn_up, model.layers[il].ffn_up_b,
NULL, NULL,
model.layers[il].ffn_down, model.layers[il].ffn_down_b,
NULL,
LLM_FFN_RELU_SQR, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il);
}
@ -5066,6 +5024,7 @@ struct llm_build_context {
model.layers[il].ffn_up, NULL,
model.layers[il].ffn_gate, NULL,
model.layers[il].ffn_down, NULL,
NULL,
LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
cb(cur, "ffn_out", il);
}
@ -5161,6 +5120,7 @@ struct llm_build_context {
model.layers[il].ffn_up, model.layers[il].ffn_up_b,
NULL, NULL,
model.layers[il].ffn_down, model.layers[il].ffn_down_b,
NULL,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il);
}
@ -5246,19 +5206,12 @@ struct llm_build_context {
NULL,
LLM_NORM, cb, il);
cb(cur, "ffn_norm", il);
if (hparams.use_awq) {
cur = llm_build_ffn_mpt_awq(ctx0, cur,
model.layers[il].ffn_up, NULL,
model.layers[il].ffn_down, NULL,
model.layers[il].ffn_act,
LLM_FFN_GELU_ACT, LLM_FFN_SEQ, cb, il);
} else {
cur = llm_build_ffn(ctx0, cur,
model.layers[il].ffn_up, NULL,
NULL, NULL,
model.layers[il].ffn_down, NULL,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
}
cur = llm_build_ffn(ctx0, cur,
model.layers[il].ffn_up, NULL,
NULL, NULL,
model.layers[il].ffn_down, NULL,
model.layers[il].ffn_act,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il);
}
@ -5366,6 +5319,7 @@ struct llm_build_context {
model.layers[il].ffn_up, NULL,
model.layers[il].ffn_gate, NULL,
model.layers[il].ffn_down, NULL,
NULL,
LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
cb(cur, "ffn_out", il);
}
@ -5478,6 +5432,7 @@ struct llm_build_context {
model.layers[il].ffn_up, NULL,
model.layers[il].ffn_gate, NULL,
model.layers[il].ffn_down, NULL,
NULL,
LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
cb(cur, "ffn_out", il);
}
@ -5585,6 +5540,7 @@ struct llm_build_context {
model.layers[il].ffn_up, model.layers[il].ffn_up_b,
NULL, NULL,
model.layers[il].ffn_down, model.layers[il].ffn_down_b,
NULL,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(ffn_output, "ffn_out", il);
}
@ -9111,7 +9067,6 @@ struct llama_model_params llama_model_default_params() {
/*.vocab_only =*/ false,
/*.use_mmap =*/ true,
/*.use_mlock =*/ false,
/*.use_awq =*/ false,
};
#ifdef GGML_USE_METAL

1
llama.h
View file

@ -195,7 +195,6 @@ extern "C" {
bool vocab_only; // only load the vocabulary, no weights
bool use_mmap; // use mmap if possible
bool use_mlock; // force system to keep model in RAM
bool use_awq; // whether to use awq quantization
};
struct llama_context_params {