Skip to content

llama : reuse compute graphs #14482

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 5 commits into
base: master
Choose a base branch
from
+403 −191
Open

llama : reuse compute graphs #14482

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
76681e3
llama : reuse compute graphs
ggerganov Jul 1, 2025
0d2038f
llama-bench : add graph reuse parameter
ggerganov Jul 7, 2025
5143876
cont : remove the parameter and the sched resets
ggerganov Jul 8, 2025
ee7c644
graph : rename update() to can_reuse()
ggerganov Jul 8, 2025
600e69f
params : remove is_same()
ggerganov Jul 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions include/llama.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1429,6 +1429,7 @@ extern "C" {

int32_t n_p_eval;
int32_t n_eval;
int32_t n_reused;
};

struct llama_perf_sampler_data {
Expand Down
175 changes: 87 additions & 88 deletions src/llama-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -227,8 +227,8 @@ llama_context::llama_context(

LLAMA_LOG_DEBUG("%s: max_nodes = %zu\n", __func__, max_nodes);

// buffer used to store the computation graph and the tensor meta data
buf_compute_meta.resize(ggml_tensor_overhead()*max_nodes + ggml_graph_overhead_custom(max_nodes, false));
gf_res_prev.reset(new llm_graph_result(max_nodes));
gf_res_reserve.reset(new llm_graph_result(max_nodes));

// TODO: move these checks to ggml_backend_sched
// enabling pipeline parallelism in the scheduler increases memory usage, so it is only done when necessary
Expand Down Expand Up @@ -388,10 +388,6 @@ ggml_backend_sched_t llama_context::get_sched() const {
return sched.get();
}

ggml_context * llama_context::get_ctx_compute() const {
return ctx_compute.get();
}

uint32_t llama_context::n_ctx() const {
return cparams.n_ctx;
}
Expand Down Expand Up @@ -678,38 +674,59 @@ bool llama_context::apply_adapter_cvec(
return cvec.apply(model, data, len, n_embd, il_start, il_end);
}

llm_graph_result_ptr llama_context::process_ubatch(const llama_ubatch & ubatch, llm_graph_type gtype, llama_memory_context_i * mctx, ggml_status & ret) {
llm_graph_result_i * llama_context::process_ubatch(const llama_ubatch & ubatch, llm_graph_type gtype, llama_memory_context_i * mctx, ggml_status & ret) {
if (mctx && !mctx->apply()) {
LLAMA_LOG_ERROR("%s: failed to apply memory context\n", __func__);
ret = GGML_STATUS_FAILED;
return nullptr;
}

auto * gf = graph_init();
if (!gf) {
LLAMA_LOG_ERROR("%s: failed to initialize graph\n", __func__);
ret = GGML_STATUS_FAILED;
return nullptr;
}
auto * res = gf_res_prev.get();
auto * gf = res->get_gf();

auto res = graph_build(ctx_compute.get(), gf, ubatch, gtype, mctx);
if (!res) {
LLAMA_LOG_ERROR("%s: failed to build graph\n", __func__);
ret = GGML_STATUS_FAILED;
return nullptr;
}
// the new graph parameters
// in order to correctly reuse a graph, it's full topology has to be uniquely determined by these parameters
const auto gparams = graph_params(res, ubatch, mctx, gtype);

// LLAMA_LOG_INFO("graph build time: %.3f ms (%d nodes, %d leafs)\n", (ggml_time_us() - t_start_us)/1000.0, gf->n_nodes, gf->n_leafs);
if (res->can_reuse(gparams)) {
//LLAMA_LOG_DEBUG("%s: reusing previous graph\n", __func__);

if (!ggml_backend_sched_alloc_graph(sched.get(), gf)) {
LLAMA_LOG_ERROR("%s: failed to allocate graph\n", __func__);
ret = GGML_STATUS_ALLOC_FAILED;
return nullptr;
n_reused++;
} else {
res->reset();

ggml_backend_sched_reset(sched.get());
ggml_backend_sched_set_eval_callback(sched.get(), cparams.cb_eval, cparams.cb_eval_user_data);

//const auto t_start_us = ggml_time_us();

gf = model.build_graph(gparams);

//LLAMA_LOG_INFO("graph build time: %.3f ms\n", (ggml_time_us() - t_start_us)/1000.0);

if (!gf) {
LLAMA_LOG_ERROR("%s: failed to initialize graph\n", __func__);
ret = GGML_STATUS_FAILED;
return nullptr;
}

if (!ggml_backend_sched_alloc_graph(sched.get(), gf)) {
LLAMA_LOG_ERROR("%s: failed to allocate graph\n", __func__);
ret = GGML_STATUS_ALLOC_FAILED;
return nullptr;
}
}

res->set_inputs(&ubatch);
// set the input data for the input tensors
{
//const auto t_start_us = ggml_time_us();

res->set_inputs(&ubatch);

const auto status = graph_compute(gf, ubatch.n_tokens > 1);
//LLAMA_LOG_INFO("graph set inputs time: %.3f ms\n", (ggml_time_us() - t_start_us)/1000.0);
}

const auto status = graph_compute(res->get_gf(), ubatch.n_tokens > 1);
if (status != GGML_STATUS_SUCCESS) {
LLAMA_LOG_ERROR("%s: failed to compute graph, compute status: %d\n", __func__, status);
ret = status;
Expand Down Expand Up @@ -767,9 +784,6 @@ int llama_context::encode(const llama_batch & batch_inp) {

n_outputs = n_tokens;

ggml_backend_sched_reset(sched.get());
ggml_backend_sched_set_eval_callback(sched.get(), cparams.cb_eval, cparams.cb_eval_user_data);

const auto causal_attn_org = cparams.causal_attn;

// always use non-causal attention for encoder graphs
Expand All @@ -778,7 +792,7 @@ int llama_context::encode(const llama_batch & batch_inp) {
cparams.causal_attn = false;

ggml_status status;
const auto res = process_ubatch(ubatch, LLM_GRAPH_TYPE_ENCODER, nullptr, status);
const auto * res = process_ubatch(ubatch, LLM_GRAPH_TYPE_ENCODER, nullptr, status);

cparams.causal_attn = causal_attn_org;

Expand Down Expand Up @@ -844,10 +858,6 @@ int llama_context::encode(const llama_batch & batch_inp) {
}
}

// Reset state for the next token before backend sync, to allow the CPU activities in the reset to
// overlap with device computation.
ggml_backend_sched_reset(sched.get());

// TODO: hacky solution
if (model.arch == LLM_ARCH_T5 && t_embd) {
//cross.t_embd = t_embd;
Expand Down Expand Up @@ -1005,11 +1015,8 @@ int llama_context::decode(const llama_batch & batch_inp) {
n_outputs = n_outputs_new;
}

ggml_backend_sched_reset(sched.get());
ggml_backend_sched_set_eval_callback(sched.get(), cparams.cb_eval, cparams.cb_eval_user_data);

ggml_status status;
const auto res = process_ubatch(ubatch, LLM_GRAPH_TYPE_DECODER, mctx.get(), status);
const auto * res = process_ubatch(ubatch, LLM_GRAPH_TYPE_DECODER, mctx.get(), status);

if (!res) {
// the last ubatch failed or was aborted -> remove all positions of that ubatch from the KV cache
Expand Down Expand Up @@ -1190,10 +1197,6 @@ int llama_context::decode(const llama_batch & batch_inp) {
// wait for the computation to finish (automatically done when obtaining the model output)
//synchronize();

// Reset state for the next token before backend sync, to allow the CPU activities in the reset to
// overlap with device computation.
ggml_backend_sched_reset(sched.get());

return 0;
}

Expand Down Expand Up @@ -1275,20 +1278,8 @@ uint32_t llama_context::output_reserve(int32_t n_outputs) {
// graph
//

int32_t llama_context::graph_max_nodes() const {
return std::max<int32_t>(65536, 5*model.n_tensors());
}

ggml_cgraph * llama_context::graph_init() {
ggml_init_params params = {
/*.mem_size =*/ buf_compute_meta.size(),
/*.mem_buffer =*/ buf_compute_meta.data(),
/*.no_alloc =*/ true,
};

ctx_compute.reset(ggml_init(params));

return ggml_new_graph_custom(ctx_compute.get(), graph_max_nodes(), false);
uint32_t llama_context::graph_max_nodes() const {
return std::max<uint32_t>(65536u, 5u*model.n_tensors());
}

ggml_cgraph * llama_context::graph_reserve(uint32_t n_tokens, uint32_t n_seqs, uint32_t n_outputs, const llama_memory_context_i * mctx) {
Expand All @@ -1301,6 +1292,9 @@ ggml_cgraph * llama_context::graph_reserve(uint32_t n_tokens, uint32_t n_seqs, u
LLAMA_LOG_DEBUG("%s: making n_tokens a multiple of n_seqs - n_tokens = %u, n_seqs = %u, n_outputs = %u\n", __func__, n_tokens, n_seqs, n_outputs);
}

gf_res_prev->reset();
ggml_backend_sched_reset(sched.get());

// store the n_outputs as it is, and restore it afterwards
// TODO: not sure if needed, might simplify in the future by removing this
const auto save_n_outputs = this->n_outputs;
Expand All @@ -1310,17 +1304,15 @@ ggml_cgraph * llama_context::graph_reserve(uint32_t n_tokens, uint32_t n_seqs, u
llama_batch_allocr balloc(model.hparams.n_pos_per_embd());
llama_ubatch ubatch = balloc.ubatch_reserve(n_tokens/n_seqs, n_seqs);

auto * gf = graph_init();
auto res = graph_build(ctx_compute.get(), gf, ubatch, LLM_GRAPH_TYPE_DEFAULT, mctx);
auto * res = gf_res_reserve.get();

this->n_outputs = save_n_outputs;
const auto gparams = graph_params(res, ubatch, mctx, LLM_GRAPH_TYPE_DEFAULT);

if (!res) {
LLAMA_LOG_ERROR("%s: failed to build worst-case graph\n", __func__);
return nullptr;
}
res->reset();

ggml_backend_sched_reset(sched.get());
auto * gf = model.build_graph(gparams);

this->n_outputs = save_n_outputs;

// initialize scheduler with the specified graph
if (!ggml_backend_sched_reserve(sched.get(), gf)) {
Expand All @@ -1331,28 +1323,27 @@ ggml_cgraph * llama_context::graph_reserve(uint32_t n_tokens, uint32_t n_seqs, u
return gf;
}

llm_graph_result_ptr llama_context::graph_build(
ggml_context * ctx,
ggml_cgraph * gf,
const llama_ubatch & ubatch,
llm_graph_type gtype,
const llama_memory_context_i * mctx) {
return model.build_graph(
{
/*.ctx =*/ ctx,
/*.arch =*/ model.arch,
/*.hparams =*/ model.hparams,
/*.cparams =*/ cparams,
/*.ubatch =*/ ubatch,
/*.sched =*/ sched.get(),
/*.backend_cpu =*/ backend_cpu,
/*.cvec =*/ &cvec,
/*.loras =*/ &loras,
/*.mctx =*/ mctx,
/*.cross =*/ &cross,
/*.n_outputs =*/ n_outputs,
/*.cb =*/ graph_get_cb(),
}, gf, gtype);
llm_graph_params llama_context::graph_params(
llm_graph_result_i * res,
const llama_ubatch & ubatch,
const llama_memory_context_i * mctx,
llm_graph_type gtype) const {
return {
/*.arch =*/ model.arch,
/*.hparams =*/ model.hparams,
/*.cparams =*/ cparams,
/*.ubatch =*/ ubatch,
/*.gtype =*/ gtype,
/*.sched =*/ sched.get(),
/*.backend_cpu =*/ backend_cpu,
/*.cvec =*/ &cvec,
/*.loras =*/ &loras,
/*.mctx =*/ mctx,
/*.cross =*/ &cross,
/*.n_outputs =*/ n_outputs,
/*.cb =*/ graph_get_cb(),
/*.res =*/ res,
};
}

ggml_status llama_context::graph_compute(
Expand Down Expand Up @@ -1930,6 +1921,7 @@ llama_perf_context_data llama_context::perf_get_data() const {
data.t_eval_ms = 1e-3 * t_eval_us;
data.n_p_eval = std::max(1, n_p_eval);
data.n_eval = std::max(1, n_eval);
data.n_reused = std::max(0, n_reused);

return data;
}
Expand All @@ -1938,6 +1930,7 @@ void llama_context::perf_reset() {
t_start_us = ggml_time_us();
t_eval_us = n_eval = 0;
t_p_eval_us = n_p_eval = 0;
n_reused = 0;
}

//
Expand Down Expand Up @@ -2064,8 +2057,13 @@ void llama_context::opt_epoch_iter(
break;
}

auto * gf = graph_init();
auto res = graph_build(ctx_compute.get(), gf, ubatch, LLM_GRAPH_TYPE_DEFAULT, mctx.get());
auto * res = gf_res_prev.get();

const auto gparams = graph_params(res, ubatch, mctx.get(), LLM_GRAPH_TYPE_DEFAULT);

res->reset();

auto * gf = model.build_graph(gparams);

struct ggml_context * ctx_compute_opt;
{
Expand Down Expand Up @@ -2807,6 +2805,7 @@ void llama_perf_context_print(const llama_context * ctx) {
LLAMA_LOG_INFO("%s: eval time = %10.2f ms / %5d runs (%8.2f ms per token, %8.2f tokens per second)\n",
__func__, data.t_eval_ms, data.n_eval, data.t_eval_ms / data.n_eval, 1e3 / data.t_eval_ms * data.n_eval);
LLAMA_LOG_INFO("%s: total time = %10.2f ms / %5d tokens\n", __func__, (t_end_ms - data.t_start_ms), (data.n_p_eval + data.n_eval));
LLAMA_LOG_INFO("%s: graphs reused = %10d\n", __func__, data.n_reused);
}

void llama_perf_context_reset(llama_context * ctx) {
Expand Down
28 changes: 11 additions & 17 deletions src/llama-context.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,8 +35,6 @@ struct llama_context {

ggml_backend_sched_t get_sched() const;

ggml_context * get_ctx_compute() const;

uint32_t n_ctx() const;
uint32_t n_ctx_per_seq() const;
uint32_t n_batch() const;
Expand Down Expand Up @@ -96,7 +94,7 @@ struct llama_context {
// if memory_context is provided, it will be applied first to the context's memory
// ret contains the status of the graph computation
// returns nullptr only if ret != GGML_STATUS_SUCCESS
llm_graph_result_ptr process_ubatch(
llm_graph_result_i * process_ubatch(
const llama_ubatch & ubatch,
llm_graph_type gtype,
llama_memory_context_i * mctx,
Expand Down Expand Up @@ -188,10 +186,7 @@ struct llama_context {
//

public:
int32_t graph_max_nodes() const;

// zero-out inputs and create the ctx_compute for the compute graph
ggml_cgraph * graph_init();
uint32_t graph_max_nodes() const;

// returns the result of ggml_backend_sched_graph_compute_async execution
ggml_status graph_compute(ggml_cgraph * gf, bool batched);
Expand All @@ -200,12 +195,11 @@ struct llama_context {
ggml_cgraph * graph_reserve(uint32_t n_tokens, uint32_t n_seqs, uint32_t n_outputs, const llama_memory_context_i * mctx);

private:
llm_graph_result_ptr graph_build(
ggml_context * ctx,
ggml_cgraph * gf,
const llama_ubatch & ubatch,
llm_graph_type gtype,
const llama_memory_context_i * mctx);
llm_graph_params graph_params(
llm_graph_result_i * res,
const llama_ubatch & ubatch,
const llama_memory_context_i * mctx,
llm_graph_type gtype) const;

llm_graph_cb graph_get_cb() const;

Expand Down Expand Up @@ -258,8 +252,6 @@ struct llama_context {
ggml_backend_t backend_cpu = nullptr;
std::vector<ggml_backend_ptr> backends;

ggml_context_ptr ctx_compute;

// training
ggml_opt_context_t opt_ctx = nullptr;

Expand All @@ -275,8 +267,8 @@ struct llama_context {
std::vector<ggml_backend_t> backend_ptrs;
std::vector<ggml_backend_buffer_type_t> backend_buft;

// memory buffers used to evaluate the model
std::vector<uint8_t> buf_compute_meta;
llm_graph_result_ptr gf_res_prev;
llm_graph_result_ptr gf_res_reserve;

// host buffer for the model output (logits and embeddings)
ggml_backend_buffer_ptr buf_output;
Expand All @@ -294,4 +286,6 @@ struct llama_context {

mutable int32_t n_p_eval = 0; // number of tokens in eval calls for the prompt (with batch size > 1)
mutable int32_t n_eval = 0; // number of eval calls

mutable int32_t n_reused = 0; // number of times the previous graph was reused
};
Loading
Loading