Port universal assisted decoding to llama-server (#699)

* port universal assisted decoding to server

* fix calls

* fix LOG_INFO

* fix llama_detokenize call

* use emplace_back
This commit is contained in:
g2mt
2025-08-17 23:22:23 -07:00
committed by GitHub
parent 6b2c84b099
commit 06bed7e01b
5 changed files with 160 additions and 55 deletions

View File

@@ -282,6 +282,11 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
} }
} }
for (auto & rep : params.replacements_draft) {
string_process_escapes(rep.first);
string_process_escapes(rep.second);
}
if (!params.kv_overrides.empty()) { if (!params.kv_overrides.empty()) {
params.kv_overrides.emplace_back(); params.kv_overrides.emplace_back();
params.kv_overrides.back().key[0] = 0; params.kv_overrides.back().key[0] = 0;
@@ -731,6 +736,14 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
} }
return true; return true;
} }
if (arg == "--spec-replace") {
CHECK_ARG
std::string target = argv[i];
CHECK_ARG
std::string draft = argv[i];
params.replacements_draft.emplace_back(std::move(target), std::move(draft));
return true;
}
if (arg == "--cfg-negative-prompt") { if (arg == "--cfg-negative-prompt") {
CHECK_ARG CHECK_ARG
sparams.cfg_negative_prompt = argv[i]; sparams.cfg_negative_prompt = argv[i];

View File

@@ -148,6 +148,8 @@ struct gpt_params {
std::vector<llama_model_tensor_buft_override> tensor_buft_overrides; std::vector<llama_model_tensor_buft_override> tensor_buft_overrides;
std::vector<std::pair<int,int>> offload_policy; std::vector<std::pair<int,int>> offload_policy;
std::vector<std::pair<std::string, std::string>> replacements_draft; // main to speculative model replacements
bool lora_init_without_apply = false; // only load lora to memory, but do not apply it to ctx (user can manually apply lora later using llama_lora_adapter_apply) bool lora_init_without_apply = false; // only load lora to memory, but do not apply it to ctx (user can manually apply lora later using llama_lora_adapter_apply)
std::vector<llama_lora_adapter_info> lora_adapters; // lora adapter path with user defined scale std::vector<llama_lora_adapter_info> lora_adapters; // lora adapter path with user defined scale

View File

@@ -6,25 +6,32 @@
#include <cstring> #include <cstring>
#include <algorithm> #include <algorithm>
#include <map>
#define SPEC_VOCAB_MAX_SIZE_DIFFERENCE 128 #define SPEC_VOCAB_MAX_SIZE_DIFFERENCE 128
#define SPEC_VOCAB_CHECK_START_TOKEN_ID 5 #define SPEC_VOCAB_CHECK_START_TOKEN_ID 5
struct llama_speculative { struct llama_speculative {
struct llama_context * ctx; struct llama_context * ctx_tgt; // only used for retokenizing from ctx_dft
struct llama_context * ctx_dft;
struct llama_sampling_context * smpl; struct llama_sampling_context * smpl;
llama_batch batch; llama_batch batch;
std::vector<llama_token> prompt; std::vector<llama_token> prompt_dft;
bool vocab_dft_compatible = true; // whether retokenization is needed
std::map<std::string, std::string> tgt_dft_replacements = {};
}; };
struct llama_speculative * llama_speculative_init( struct llama_speculative * llama_speculative_init(
struct llama_context * ctx_tgt,
struct llama_context * ctx_dft) { struct llama_context * ctx_dft) {
auto * result = new llama_speculative { auto * result = new llama_speculative {
/* .ctx = */ ctx_dft, /* .ctx_tgt = */ ctx_tgt,
/* .smpl = */ nullptr, /* .ctx_dft = */ ctx_dft,
/* .batch = */ llama_batch_init(llama_n_batch(ctx_dft), 0, 1), /* .smpl = */ nullptr,
/* .prompt = */ {}, /* .batch = */ llama_batch_init(llama_n_batch(ctx_dft), 0, 1),
/* .prompt_dft = */ {},
/* .vocab_dft_compatible = */ false,
}; };
// TODO: optimize or pass from outside? // TODO: optimize or pass from outside?
@@ -56,6 +63,9 @@ struct llama_speculative * llama_speculative_init(
} }
#endif #endif
result->vocab_dft_compatible = llama_speculative_are_compatible(ctx_tgt, ctx_dft);
LLAMA_LOG_INFO("vocab_dft_compatible = %d\n", result->vocab_dft_compatible);
return result; return result;
} }
@@ -87,18 +97,18 @@ bool llama_speculative_are_compatible(
LLAMA_LOG_INFO("%s: vocab_type dft: %d\n", __func__, vocab_type_dft); LLAMA_LOG_INFO("%s: vocab_type dft: %d\n", __func__, vocab_type_dft);
if (vocab_type_tgt != vocab_type_dft) { if (vocab_type_tgt != vocab_type_dft) {
LLAMA_LOG_ERROR("%s: draft model vocab type must match target model to use speculation but " LLAMA_LOG_INFO("%s: draft model vocab type must match target model to use speculation but ", __func__);
"vocab_type_dft = %d while vocab_type_tgt = %d\n", __func__, vocab_type_dft, vocab_type_tgt); LLAMA_LOG_INFO("vocab_type_dft = %d while vocab_type_tgt = %d\n", vocab_type_dft, vocab_type_tgt);
return false; return false;
} }
if (llama_add_bos_token(model_tgt) != llama_add_bos_token(model_dft) || if (
llama_add_bos_token(model_tgt) != llama_add_bos_token(model_dft) ||
llama_add_eos_token(model_tgt) != llama_add_eos_token(model_dft) || llama_add_eos_token(model_tgt) != llama_add_eos_token(model_dft) ||
llama_token_bos(model_tgt) != llama_token_bos(model_dft) || llama_token_bos(model_tgt) != llama_token_bos(model_dft) ||
llama_token_eos(model_tgt) != llama_token_eos(model_dft)) { llama_token_eos(model_tgt) != llama_token_eos(model_dft)
LLAMA_LOG_ERROR("%s: draft vocab special tokens must match target vocab to use speculation\n", __func__); ) {
LLAMA_LOG_ERROR("%s: tgt: bos = %d (%d), eos = %d (%d)\n", __func__, llama_token_bos(model_tgt), llama_add_bos_token(model_tgt), llama_token_eos(model_tgt), llama_add_eos_token(model_tgt)); LLAMA_LOG_INFO("%s: draft model special tokens must match target model to use speculation\n", __func__);
LLAMA_LOG_ERROR("%s: dft: bos = %d (%d), eos = %d (%d)\n", __func__, llama_token_bos(model_dft), llama_add_bos_token(model_dft), llama_token_eos(model_dft), llama_add_eos_token(model_dft));
return false; return false;
} }
@@ -106,12 +116,14 @@ bool llama_speculative_are_compatible(
const int n_vocab_tgt = llama_n_vocab(model_tgt); const int n_vocab_tgt = llama_n_vocab(model_tgt);
const int n_vocab_dft = llama_n_vocab(model_dft); const int n_vocab_dft = llama_n_vocab(model_dft);
const int model_diff = std::abs(n_vocab_tgt - n_vocab_dft); const int model_diff = n_vocab_tgt > n_vocab_dft
? n_vocab_tgt - n_vocab_dft
: n_vocab_dft - n_vocab_tgt;
if (model_diff > SPEC_VOCAB_MAX_SIZE_DIFFERENCE) { if (model_diff > SPEC_VOCAB_MAX_SIZE_DIFFERENCE) {
LLAMA_LOG_ERROR("%s: draft model vocab must closely match target model to use speculation but " LLAMA_LOG_INFO("%s: draft model vocab must closely match target model to use speculation but ", __func__);
"target vocab size %d does not match draft vocab size %d - difference %d, max allowed %d\n", LLAMA_LOG_INFO("target vocab size %d does not match draft vocab size %d - difference %d, max allowed %d\n",
__func__, n_vocab_tgt, n_vocab_dft, model_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE); n_vocab_tgt, n_vocab_dft, model_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE);
return false; return false;
} }
@@ -119,8 +131,8 @@ bool llama_speculative_are_compatible(
const char * token_text_tgt = llama_token_get_text(model_tgt, i); const char * token_text_tgt = llama_token_get_text(model_tgt, i);
const char * token_text_dft = llama_token_get_text(model_dft, i); const char * token_text_dft = llama_token_get_text(model_dft, i);
if (std::strcmp(token_text_tgt, token_text_dft) != 0) { if (std::strcmp(token_text_tgt, token_text_dft) != 0) {
LLAMA_LOG_ERROR("%s: draft vocab vocab must match target vocab to use speculation but " LLAMA_LOG_INFO("%s: draft model vocab must match target model to use speculation but ", __func__);
"token %d content differs - target '%s', draft '%s'\n", __func__, i, LLAMA_LOG_INFO("token %d content differs - target '%s', draft '%s'\n", i,
llama_token_to_piece(ctx_tgt, i).c_str(), llama_token_to_piece(ctx_tgt, i).c_str(),
llama_token_to_piece(ctx_dft, i).c_str()); llama_token_to_piece(ctx_dft, i).c_str());
return false; return false;
@@ -131,30 +143,88 @@ bool llama_speculative_are_compatible(
return true; return true;
} }
void llama_speculative_add_replacement_tgt_dft(
struct llama_speculative * spec,
const char *source, const char *dest) {
spec->tgt_dft_replacements[source] = dest;
}
static std::string replace_to_dft(
struct llama_speculative * spec,
const std::string& input) {
std::string result = input;
for (const auto & pair : spec->tgt_dft_replacements) {
size_t pos = result.find(pair.first);
while (pos != std::string::npos) {
result.replace(pos, pair.first.length(), pair.second);
pos = result.find(pair.first, pos + pair.second.length());
}
}
return result;
}
static std::string replace_to_tgt(
struct llama_speculative * spec,
const std::string& input) {
std::vector<std::pair<std::string, std::string>> sorted_pairs(spec->tgt_dft_replacements.begin(), spec->tgt_dft_replacements.end());
std::sort(sorted_pairs.begin(), sorted_pairs.end(), [](const auto &a, const auto &b) {
return a.second.length() > b.second.length(); // Sort by length in descending order
});
std::string result = input;
for (const auto & pair : sorted_pairs) {
size_t pos = 0;
while ((pos = result.find(pair.second, pos)) != std::string::npos) {
result.replace(pos, pair.second.length(), pair.first);
pos += pair.first.length();
}
}
return result;
}
std::vector<llama_token> llama_speculative_gen_draft( std::vector<llama_token> llama_speculative_gen_draft(
struct llama_speculative * spec, struct llama_speculative * spec,
struct llama_speculative_params params, struct llama_speculative_params params,
const std::vector<llama_token> & prompt_tgt, const std::vector<llama_token> & prompt_tgt_main_model, // specified in target model vocab
llama_token id_last) { llama_token id_last) {
auto & batch = spec->batch; auto & batch = spec->batch;
auto & ctx = spec->ctx; auto & ctx_tgt = spec->ctx_tgt;
auto & ctx_dft = spec->ctx_dft;
auto & smpl = spec->smpl; auto & smpl = spec->smpl;
auto & prompt = spec->prompt; auto & prompt_dft = spec->prompt_dft;
int reuse_i = 0; int reuse_i = 0;
int reuse_n = 0; int reuse_n = 0;
const int n_ctx = llama_n_ctx(ctx) - params.n_draft; const int n_ctx = llama_n_ctx(ctx_dft) - params.n_draft;
std::vector<llama_token> prompt_tgt_draft_model;
if (!spec->vocab_dft_compatible) {
std::string text;
text = llama_detokenize(ctx_tgt, prompt_tgt_main_model, true);
text = replace_to_dft(spec, text);
LLAMA_LOG_INFO("%s: main->draft detokenized string: '%s'\n", __func__, text.c_str());
prompt_tgt_draft_model = llama_tokenize(ctx_dft, text, false, true);
// convert id_last to draft vocab
std::vector<llama_token> id_last_vec(1, id_last);
text = llama_detokenize(ctx_tgt, id_last_vec);
LLAMA_LOG_INFO("main->draft detokenized id_last(%d): '%s'\n", id_last, text.c_str());
id_last = llama_tokenize(ctx_dft, text, false, true)[0];
}
// prompt_tgt's tokens will always be compatible with ctx_dft
const std::vector<llama_token> &prompt_tgt =
spec->vocab_dft_compatible ? prompt_tgt_main_model : prompt_tgt_draft_model;
const int i_start = std::max<int>(0, (int) prompt_tgt.size() - n_ctx); const int i_start = std::max<int>(0, (int) prompt_tgt.size() - n_ctx);
// reuse as much as possible from the old draft context // reuse as much as possible from the old draft context
// ideally, the draft context should be as big as the target context and we will always reuse the entire prompt // ideally, the draft context should be as big as the target context and we will always reuse the entire prompt
for (int i = 0; i < (int) prompt.size(); ++i) { for (int i = 0; i < (int) prompt_dft.size(); ++i) {
int cur = 0; int cur = 0;
while (i_start + cur < (int) prompt_tgt.size() && while (i_start + cur < (int) prompt_tgt.size() &&
i + cur < (int) prompt.size() && i + cur < (int) prompt_dft.size() &&
prompt_tgt[i_start + cur] == prompt[i + cur]) { prompt_tgt[i_start + cur] == prompt_dft[i + cur]) {
cur++; cur++;
} }
@@ -164,21 +234,21 @@ std::vector<llama_token> llama_speculative_gen_draft(
} }
} }
// LLAMA_LOG_INFO("%s: reuse_i = %d, reuse_n = %d, prompt = %d\n", __func__, reuse_i, reuse_n, (int) prompt.size()); LLAMA_LOG_INFO("%s: reuse_i = %d, reuse_n = %d, prompt = %d\n", __func__, reuse_i, reuse_n, (int) prompt_dft.size());
std::vector<llama_token> result; std::vector<llama_token> result;
result.reserve(params.n_draft); result.reserve(params.n_draft);
if (reuse_n == 0) { if (reuse_n == 0) {
llama_kv_cache_clear(ctx); llama_kv_cache_clear(ctx_dft);
prompt.clear(); prompt_dft.clear();
} else { } else {
// this happens when a previous draft has been discarded (for example, due to being too small), but the // this happens when a previous draft has been discarded (for example, due to being too small), but the
// target model agreed with it. in this case, we simply pass back the previous results to save compute // target model agreed with it. in this case, we simply pass back the previous results to save compute
if (reuse_i + reuse_n < (int) prompt.size() && prompt[reuse_i + reuse_n] == id_last) { if (reuse_i + reuse_n < (int) prompt_dft.size() && prompt_dft[reuse_i + reuse_n] == id_last) {
for (int i = reuse_i + reuse_n + 1; i < (int) prompt.size(); ++i) { for (int i = reuse_i + reuse_n + 1; i < (int) prompt_dft.size(); ++i) {
result.push_back(prompt[i]); result.push_back(prompt_dft[i]);
if (params.n_draft <= (int) result.size()) { if (params.n_draft <= (int) result.size()) {
break; break;
@@ -189,16 +259,16 @@ std::vector<llama_token> llama_speculative_gen_draft(
} }
if (reuse_i > 0) { if (reuse_i > 0) {
llama_kv_cache_seq_rm (ctx, 0, 0, reuse_i); llama_kv_cache_seq_rm (ctx_dft, 0, 0, reuse_i);
llama_kv_cache_seq_add(ctx, 0, reuse_i, -1, -reuse_i); llama_kv_cache_seq_add(ctx_dft, 0, reuse_i, -1, -reuse_i);
prompt.erase(prompt.begin(), prompt.begin() + reuse_i); prompt_dft.erase(prompt_dft.begin(), prompt_dft.begin() + reuse_i);
} }
if (reuse_n < (int) prompt.size()) { if (reuse_n < (int) prompt_dft.size()) {
llama_kv_cache_seq_rm (ctx, 0, reuse_n, -1); llama_kv_cache_seq_rm (ctx_dft, 0, reuse_n, -1);
prompt.erase(prompt.begin() + reuse_n, prompt.end()); prompt_dft.erase(prompt_dft.begin() + reuse_n, prompt_dft.end());
} }
} }
@@ -209,28 +279,28 @@ std::vector<llama_token> llama_speculative_gen_draft(
//LLAMA_LOG_INFO("i = %d, i_start = %d, reuse_n = %d, i - i_start = %d, id = %6d\n", i, i_start, reuse_n, i - i_start, prompt_tgt[i]); //LLAMA_LOG_INFO("i = %d, i_start = %d, reuse_n = %d, i - i_start = %d, id = %6d\n", i, i_start, reuse_n, i - i_start, prompt_tgt[i]);
llama_batch_add(batch, prompt_tgt[i], i - i_start, { 0 }, false); llama_batch_add(batch, prompt_tgt[i], i - i_start, { 0 }, false);
prompt.push_back(prompt_tgt[i]); prompt_dft.push_back(prompt_tgt[i]);
} }
// we should rarely end-up here during normal decoding // we should rarely end-up here during normal decoding
if (batch.n_tokens > 0) { if (batch.n_tokens > 0) {
//LLAMA_LOG_INFO("%s: draft prompt batch: %s\n", __func__, string_from(ctx, batch).c_str()); //LLAMA_LOG_INFO("%s: draft prompt batch: %s\n", __func__, string_from(ctx_dft, batch).c_str());
llama_decode(ctx, batch); llama_decode(ctx_dft, batch);
} }
const llama_pos n_past = prompt.size(); const llama_pos n_past = prompt_dft.size();
// LLAMA_LOG_INFO("%s: n_past = %d\n", __func__, n_past); // LLAMA_LOG_INFO("%s: n_past = %d\n", __func__, n_past);
llama_batch_clear(batch); llama_batch_clear(batch);
llama_batch_add (batch, id_last, n_past, { 0 }, true); llama_batch_add (batch, id_last, n_past, { 0 }, true);
prompt.push_back(id_last); prompt_dft.push_back(id_last);
//LLAMA_LOG_INFO("%s: draft prompt: %s\n", __func__, string_from(ctx, prompt).c_str()); //LLAMA_LOG_INFO("%s: draft prompt: %s\n", __func__, string_from(ctx_dft, prompt_dft).c_str());
llama_decode(ctx, batch); llama_decode(ctx_dft, batch);
llama_sampling_reset(smpl); llama_sampling_reset(smpl);
@@ -238,19 +308,19 @@ std::vector<llama_token> llama_speculative_gen_draft(
for (int i = 0; i < params.n_draft; ++i) { for (int i = 0; i < params.n_draft; ++i) {
llama_batch_clear(batch); llama_batch_clear(batch);
llama_sampling_sample(smpl, ctx, nullptr, 0); llama_sampling_sample(smpl, ctx_dft, nullptr, 0);
const auto * cur_p = llama_sampling_get_candidates(smpl); const auto * cur_p = llama_sampling_get_candidates(smpl);
// for (int k = 0; k < std::min(3, (int) cur_p->size); ++k) { // for (int k = 0; k < std::min(3, (int) cur_p->size); ++k) {
// LLAMA_LOG_INFO(" - draft candidate %3d, pos %3d: %6d (%8.3f) '%s'\n", // LLAMA_LOG_INFO(" - draft candidate %3d, pos %3d: %6d (%8.3f) '%s'\n",
// k, i, cur_p->data[k].id, cur_p->data[k].p, llama_token_to_piece(ctx, cur_p->data[k].id).c_str()); // k, i, cur_p->data[k].id, cur_p->data[k].p, llama_token_to_piece(ctx_dft, cur_p->data[k].id).c_str());
// } // }
// add drafted token for each sequence // add drafted token for each sequence
const llama_token id = cur_p->data[0].id; const llama_token id = cur_p->data[0].id;
llama_sampling_accept(smpl, ctx, id, true); llama_sampling_accept(smpl, ctx_dft, id, true);
result.push_back(id); result.push_back(id);
@@ -266,10 +336,20 @@ std::vector<llama_token> llama_speculative_gen_draft(
llama_batch_add(batch, id, n_past + i + 1, { 0 }, true); llama_batch_add(batch, id, n_past + i + 1, { 0 }, true);
// evaluate the drafted tokens on the draft model // evaluate the drafted tokens on the draft model
llama_decode(ctx, batch); llama_decode(ctx_dft, batch);
prompt.push_back(id); prompt_dft.push_back(id);
} }
if (!spec->vocab_dft_compatible) {
std::string detokenized = llama_detokenize(ctx_dft, result, true);
detokenized = replace_to_tgt(spec, detokenized);
LLAMA_LOG_INFO("draft->main detokenized string: '%s'\n", detokenized.c_str());
result = llama_tokenize(ctx_tgt, detokenized, false, true);
if (result.size() > (size_t)params.n_draft) {
result.resize(params.n_draft);
}
}
return result; return result;
} }

View File

@@ -13,10 +13,17 @@ struct llama_speculative_params {
float p_min = 0.75f; // min probability required to accept a token in the draft float p_min = 0.75f; // min probability required to accept a token in the draft
}; };
struct llama_speculative * llama_speculative_init(struct llama_context * ctx_dft); struct llama_speculative * llama_speculative_init(
struct llama_context * ctx_tgt,
struct llama_context * ctx_dft
);
void llama_speculative_free(struct llama_speculative * spec); void llama_speculative_free(struct llama_speculative * spec);
void llama_speculative_add_replacement_tgt_dft(
struct llama_speculative * spec,
const char *source, const char *dest);
bool llama_speculative_are_compatible( bool llama_speculative_are_compatible(
const struct llama_context * ctx_tgt, const struct llama_context * ctx_tgt,
const struct llama_context * ctx_dft); const struct llama_context * ctx_dft);

View File

@@ -910,7 +910,7 @@ struct server_context {
chat_templates = llama_chat_templates_from_model(model, params.chat_template); chat_templates = llama_chat_templates_from_model(model, params.chat_template);
} }
GGML_ASSERT(chat_templates.template_default.get() != nullptr); GGML_ASSERT(chat_templates.template_default.get() != nullptr);
// Load draft model for speculative decoding if specified // Load draft model for speculative decoding if specified
if (!params.model_draft.empty()) { if (!params.model_draft.empty()) {
LOG_INFO("loading draft model", {{"model", params.model_draft}}); LOG_INFO("loading draft model", {{"model", params.model_draft}});
@@ -933,8 +933,7 @@ struct server_context {
} }
if (!llama_speculative_are_compatible(ctx, llama_init_dft.context)) { if (!llama_speculative_are_compatible(ctx, llama_init_dft.context)) {
LOG_ERROR("the draft model is not compatible with the target model", {}); LOG_INFO("the draft model is not compatible with the target model. tokens will be translated between the draft and target models.", {{}});
return false;
} }
const int n_ctx_dft = llama_n_ctx(llama_init_dft.context); const int n_ctx_dft = llama_n_ctx(llama_init_dft.context);
@@ -1029,11 +1028,15 @@ struct server_context {
return; return;
} }
slot.spec = llama_speculative_init(slot.ctx_dft); slot.spec = llama_speculative_init(ctx, slot.ctx_dft);
if (slot.spec == nullptr) { if (slot.spec == nullptr) {
LOG_ERROR("failed to create speculator", {}); LOG_ERROR("failed to create speculator", {});
return; return;
} }
for (auto & pair : params.replacements_draft) {
llama_speculative_add_replacement_tgt_dft(slot.spec, pair.first.c_str(), pair.second.c_str());
}
} }
slot.reset(); slot.reset();