public_git/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/whisper.cpp.git synced 2026-10-04 05:21:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

talk-llama : sync llama.cpp

Browse Source
This commit is contained in:
Georgi Gerganov
2026-02-27 20:57:58 +02:00
parent 4734056067
commit 84f8db71d8
41 changed files with 1734 additions and 2227 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/talk-llama/llama-context.cpp
+13 -174
View File
@@ -22,6 +22,8 @@ llama_context::llama_context(
const llama_model & model,
llama_context_params params) :
model(model),
cvec(std::make_unique<llama_adapter_cvec>()),
loras(std::make_unique<llama_adapter_loras>()),
balloc(std::make_unique<llama_batch_allocr>(model.hparams.n_pos_per_embd())) {
// TODO warning when creating llama_context with awkward ctx size that is not a power of 2,
// may need to be backend-dependent
@@ -710,8 +712,6 @@ int64_t llama_context::output_resolve_row(int32_t i) const {
}
float * llama_context::get_logits_ith(int32_t i) {
int64_t j = -1;
output_reorder();
try {
@@ -719,26 +719,7 @@ float * llama_context::get_logits_ith(int32_t i) {
throw std::runtime_error("no logits");
}
// TODO: use output_resolve_row()
if (i < 0) {
j = n_outputs + i;
if (j < 0) {
throw std::runtime_error(format("negative index out of range [0, %d)", n_outputs));
}
} else if ((size_t) i >= output_ids.size()) {
throw std::runtime_error(format("out of range [0, %zu)", output_ids.size()));
} else {
j = output_ids[i];
}
if (j < 0) {
throw std::runtime_error(format("batch.logits[%d] != true", i));
}
if (j >= n_outputs) {
// This should not happen
throw std::runtime_error(format("corrupt output buffer (j=%" PRId64 ", n_outputs=%d)", j, n_outputs));
}
const int64_t j = output_resolve_row(i);
return logits.data + j*model.vocab.n_tokens();
} catch (const std::exception & err) {
LLAMA_LOG_ERROR("%s: invalid logits id %d, reason: %s\n", __func__, i, err.what());
@@ -761,8 +742,6 @@ llama_token * llama_context::get_sampled_tokens() const{
}
float * llama_context::get_embeddings_ith(int32_t i) {
int64_t j = -1;
output_reorder();
try {
@@ -770,26 +749,7 @@ float * llama_context::get_embeddings_ith(int32_t i) {
throw std::runtime_error("no embeddings");
}
// TODO: use output_resolve_row()
if (i < 0) {
j = n_outputs + i;
if (j < 0) {
throw std::runtime_error(format("negative index out of range [0, %d)", n_outputs));
}
} else if ((size_t) i >= output_ids.size()) {
throw std::runtime_error(format("out of range [0, %zu)", output_ids.size()));
} else {
j = output_ids[i];
}
if (j < 0) {
throw std::runtime_error(format("batch.logits[%d] != true", i));
}
if (j >= n_outputs) {
// This should not happen
throw std::runtime_error(format("corrupt output buffer (j=%" PRId64 ", n_outputs=%d)", j, n_outputs));
}
const int64_t j = output_resolve_row(i);
const uint32_t n_embd_out = model.hparams.n_embd_out();
return embd.data + j*n_embd_out;
} catch (const std::exception & err) {
@@ -1065,11 +1025,11 @@ void llama_context::set_adapters_lora(llama_adapter_lora ** adapters, size_t n_a
return;
}
loras.clear();
loras.reset(new llama_adapter_loras());
for (size_t i = 0; i < n_adapters; i ++) {
if (scales[i] != 0.0f) {
loras[adapters[i]] = scales[i];
loras->insert({adapters[i], scales[i]});
}
}
@@ -1079,14 +1039,14 @@ void llama_context::set_adapters_lora(llama_adapter_lora ** adapters, size_t n_a
bool llama_context::adapters_lora_are_same(llama_adapter_lora ** adapters, size_t n_adapters, float * scales) {
LLAMA_LOG_DEBUG("%s: adapters = %p\n", __func__, (void *) adapters);
if (n_adapters != loras.size()) {
if (n_adapters != loras->size()) {
return false;
}
for (size_t i = 0; i < n_adapters; i ++) {
auto it = loras.find(adapters[i]);
auto it = loras->find(adapters[i]);
if (it == loras.end() || it->second != scales[i]) {
if (it == loras->end() || it->second != scales[i]) {
return false;
}
}
@@ -1104,7 +1064,7 @@ bool llama_context::set_adapter_cvec(
// TODO: should we reserve?
return cvec.apply(model, data, len, n_embd, il_start, il_end);
return cvec->apply(model, data, len, n_embd, il_start, il_end);
}
llm_graph_result * llama_context::process_ubatch(const llama_ubatch & ubatch, llm_graph_type gtype, llama_memory_context_i * mctx, ggml_status & ret) {
@@ -2081,8 +2041,8 @@ llm_graph_params llama_context::graph_params(
/*.gtype =*/ gtype,
/*.sched =*/ sched.get(),
/*.backend_cpu =*/ backend_cpu,
/*.cvec =*/ &cvec,
/*.loras =*/ &loras,
/*.cvec =*/ cvec.get(),
/*.loras =*/ loras.get(),
/*.mctx =*/ mctx,
/*.cross =*/ &cross,
/*.samplers =*/ sampling.samplers,
@@ -2480,64 +2440,6 @@ size_t llama_context::state_write_data(llama_io_write_i & io) {
// TODO: add more model-specific info which should prevent loading the session file if not identical
}
// write output ids
{
LLAMA_LOG_DEBUG("%s: - writing output ids\n", __func__);
const auto n_outputs = this->n_outputs;
const auto & output_ids = this->output_ids;
std::vector<int32_t> w_output_pos;
w_output_pos.resize(n_outputs);
// build a more compact representation of the output ids
for (size_t i = 0; i < n_batch(); ++i) {
// map an output id to a position in the batch
int64_t pos = output_ids[i];
if (pos >= 0) {
GGML_ASSERT(pos < n_outputs);
w_output_pos[pos] = i;
}
}
io.write(&n_outputs, sizeof(n_outputs));
if (n_outputs) {
io.write(w_output_pos.data(), n_outputs * sizeof(int32_t));
}
}
// [TAG_CONTEXT_STATE_LOGITS]
// write logits
{
LLAMA_LOG_DEBUG("%s: - writing logits\n", __func__);
const uint64_t logits_size = std::min((uint64_t) this->logits.size, (uint64_t) n_outputs * model.vocab.n_tokens());
io.write(&logits_size, sizeof(logits_size));
if (logits_size) {
io.write(logits.data, logits_size * sizeof(float));
}
}
// write embeddings
{
LLAMA_LOG_DEBUG("%s: - writing embeddings\n", __func__);
const uint64_t embd_size = std::min((uint64_t) this->embd.size, (uint64_t) n_outputs * model.hparams.n_embd);
io.write(&embd_size, sizeof(embd_size));
if (embd_size) {
io.write(embd.data, embd_size * sizeof(float));
}
}
// TODO: handle sampling buffers and samplers state ?
// https://github.com/ggml-org/llama.cpp/pull/17004
if (memory != nullptr) {
LLAMA_LOG_DEBUG("%s: - writing memory module\n", __func__);
memory->state_write(io);
@@ -2563,70 +2465,6 @@ size_t llama_context::state_read_data(llama_io_read_i & io) {
// TODO: add more info which needs to be identical but which is not verified otherwise
}
// read output ids
{
LLAMA_LOG_DEBUG("%s: - reading output ids\n", __func__);
auto n_outputs = this->n_outputs;
io.read_to(&n_outputs, sizeof(n_outputs));
if (n_outputs > output_reserve(n_outputs)) {
throw std::runtime_error("could not reserve outputs");
}
std::vector<int32_t> output_pos;
if (n_outputs) {
output_pos.resize(n_outputs);
io.read_to(output_pos.data(), n_outputs * sizeof(int32_t));
for (int32_t i = 0; i < (int32_t) output_pos.size(); ++i) {
int32_t id = output_pos[i];
if ((uint32_t) id >= n_batch()) {
throw std::runtime_error(format("invalid output id, %d does not fit in batch size of %u", id, n_batch()));
}
this->output_ids[id] = i;
}
this->n_outputs = n_outputs;
}
}
// read logits
{
LLAMA_LOG_DEBUG("%s: - reading logits\n", __func__);
uint64_t logits_size;
io.read_to(&logits_size, sizeof(logits_size));
if (this->logits.size < logits_size) {
throw std::runtime_error("logits buffer too small");
}
if (logits_size) {
io.read_to(this->logits.data, logits_size * sizeof(float));
}
}
// read embeddings
{
LLAMA_LOG_DEBUG("%s: - reading embeddings\n", __func__);
uint64_t embd_size;
io.read_to(&embd_size, sizeof(embd_size));
if (this->embd.size < embd_size) {
throw std::runtime_error("embeddings buffer too small");
}
if (embd_size) {
io.read_to(this->embd.data, embd_size * sizeof(float));
}
}
// TODO: handle sampling buffers and samplers state ?
// https://github.com/ggml-org/llama.cpp/pull/17004
if (memory) {
LLAMA_LOG_DEBUG("%s: - reading memory module\n", __func__);
@@ -2759,6 +2597,7 @@ void llama_context::opt_init(struct llama_model * model, struct llama_opt_params
llama_set_param(model->cls_b, param_filter, param_filter_ud);
llama_set_param(model->cls_out, param_filter, param_filter_ud);
llama_set_param(model->cls_out_b, param_filter, param_filter_ud);
llama_set_param(model->cls_norm, param_filter, param_filter_ud);
for (struct llama_layer & layer : model->layers) {
for (size_t i = 0; i < sizeof(layer)/sizeof(struct ggml_tensor *); ++i) {