Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 09:15:18 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	.github/labeler.yml
#	.github/workflows/server.yml
#	.gitignore
#	CMakeLists.txt
#	Makefile
#	README-sycl.md
#	README.md
#	llama.cpp
#	requirements/requirements-convert-hf-to-gguf-update.txt
#	requirements/requirements-convert-hf-to-gguf.txt
#	requirements/requirements-convert-legacy-llama.txt
#	scripts/sync-ggml.last
#	tests/test-tokenizer-random.py
This commit is contained in:
Concedo
2024-06-22 01:33:44 +08:00
parent 13398477a1 557b653dc9
commit 92afdfcae4
44 changed files with 10304 additions and 8631 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/embedding/embedding.cpp
+11 -10
View File
@@ -18,9 +18,10 @@ static std::vector<std::string> split_lines(const std::string & s) {
return lines;
}
static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & tokens, int seq_id) {
for (size_t i = 0; i < tokens.size(); i++) {
llama_batch_add(batch, tokens[i], i, { seq_id }, i == tokens.size() - 1);
static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & tokens, llama_seq_id seq_id) {
size_t n_tokens = tokens.size();
for (size_t i = 0; i < n_tokens; i++) {
llama_batch_add(batch, tokens[i], i, { seq_id }, true);
}
}
@@ -41,13 +42,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
// try to get sequence embeddings - supported only when pooling_type is not NONE
const float * embd = llama_get_embeddings_seq(ctx, batch.seq_id[i][0]);
if (embd == NULL) {
embd = llama_get_embeddings_ith(ctx, i);
if (embd == NULL) {
fprintf(stderr, "%s: failed to get embeddings for token %d\n", __func__, i);
continue;
}
}
GGML_ASSERT(embd != NULL && "failed to get sequence embeddings");
float * out = output + batch.seq_id[i][0] * n_embd;
//TODO: I would also add a parameter here to enable normalization or not.
@@ -98,6 +93,12 @@ int main(int argc, char ** argv) {
const int n_ctx_train = llama_n_ctx_train(model);
const int n_ctx = llama_n_ctx(ctx);
const enum llama_pooling_type pooling_type = llama_pooling_type(ctx);
if (pooling_type == LLAMA_POOLING_TYPE_NONE) {
fprintf(stderr, "%s: error: pooling type NONE not supported\n", __func__);
return 1;
}
if (n_ctx > n_ctx_train) {
fprintf(stderr, "%s: warning: model was trained on only %d context tokens (%d specified)\n",
__func__, n_ctx_train, n_ctx);
examples/gritlm/gritlm.cpp
+4 -2
View File
@@ -44,6 +44,7 @@ static std::vector<std::vector<float>> encode(llama_context * ctx, const std::ve
// clear previous kv_cache values (irrelevant for embeddings)
llama_kv_cache_clear(ctx);
llama_set_embeddings(ctx, true);
llama_set_causal_attn(ctx, false);
// run model
@@ -98,7 +99,9 @@ static std::string generate(llama_context * ctx, const std::string & prompt, boo
llama_token eos_token = llama_token_eos(mdl);
llama_kv_cache_clear(ctx);
llama_set_embeddings(ctx, false);
llama_set_causal_attn(ctx, true);
llama_batch bat = llama_batch_init(llama_n_batch(ctx), 0, 1);
std::vector<llama_token> inputs = llama_tokenize(mdl, prompt, false, true);
@@ -166,8 +169,7 @@ int main(int argc, char * argv[]) {
llama_model * mdl = llama_load_model_from_file(params.model.c_str(), mparams);
// create new context - set to embedding mode
cparams.embeddings = true;
// create generation context
llama_context * ctx = llama_new_context_with_model(mdl, cparams);
// ### Embedding/Representation ###
examples/infill/infill.cpp
+11 -2
View File
@@ -224,7 +224,11 @@ int main(int argc, char ** argv) {
inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
embd_inp = inp_pfx;
embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());
embd_inp.push_back(llama_token_middle(model));
const llama_token middle_token = llama_token_middle(model);
if (middle_token >= 0) {
embd_inp.push_back(middle_token);
}
LOG("prefix: \"%s\"\n", log_tostr(params.input_prefix));
LOG("suffix: \"%s\"\n", log_tostr(params.input_suffix));
@@ -529,7 +533,12 @@ int main(int argc, char ** argv) {
inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
embd_inp = inp_pfx;
embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());
embd_inp.push_back(llama_token_middle(model));
const llama_token middle_token = llama_token_middle(model);
if (middle_token >= 0) {
embd_inp.push_back(middle_token);
}
embd.clear();
n_remain = params.n_predict;
n_past = 0;
examples/llama.swiftui/llama.swiftui/Models/LlamaState.swift
+22 -15
View File
@@ -131,22 +131,29 @@ class LlamaState: ObservableObject {
messageLog += "\(text)"
while await llamaContext.n_cur < llamaContext.n_len {
let result = await llamaContext.completion_loop()
messageLog += "\(result)"
Task.detached {
while await llamaContext.n_cur < llamaContext.n_len {
let result = await llamaContext.completion_loop()
await MainActor.run {
self.messageLog += "\(result)"
}
}
let t_end = DispatchTime.now().uptimeNanoseconds
let t_generation = Double(t_end - t_heat_end) / self.NS_PER_S
let tokens_per_second = Double(await llamaContext.n_len) / t_generation
await llamaContext.clear()
await MainActor.run {
self.messageLog += """
\n
Done
Heat up took \(t_heat)s
Generated \(tokens_per_second) t/s\n
"""
}
}
let t_end = DispatchTime.now().uptimeNanoseconds
let t_generation = Double(t_end - t_heat_end) / NS_PER_S
let tokens_per_second = Double(await llamaContext.n_len) / t_generation
await llamaContext.clear()
messageLog += """
\n
Done
Heat up took \(t_heat)s
Generated \(tokens_per_second) t/s\n
"""
}
func bench() async {
examples/retrieval/retrieval.cpp
+10 -3
View File
@@ -73,9 +73,10 @@ static std::vector<chunk> chunk_file(const std::string & filename, int chunk_siz
return chunks;
}
static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & tokens, int seq_id) {
for (size_t i = 0; i < tokens.size(); i++) {
llama_batch_add(batch, tokens[i], i, { seq_id }, i == tokens.size() - 1);
static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & tokens, llama_seq_id seq_id) {
size_t n_tokens = tokens.size();
for (size_t i = 0; i < n_tokens; i++) {
llama_batch_add(batch, tokens[i], i, { seq_id }, true);
}
}
@@ -160,6 +161,12 @@ int main(int argc, char ** argv) {
const int n_ctx_train = llama_n_ctx_train(model);
const int n_ctx = llama_n_ctx(ctx);
const enum llama_pooling_type pooling_type = llama_pooling_type(ctx);
if (pooling_type == LLAMA_POOLING_TYPE_NONE) {
fprintf(stderr, "%s: error: pooling type NONE not supported\n", __func__);
return 1;
}
if (n_ctx > n_ctx_train) {
fprintf(stderr, "%s: warning: model was trained on only %d context tokens (%d specified)\n",
__func__, n_ctx_train, n_ctx);
examples/server/server.cpp
+7 -2
View File
@@ -1595,7 +1595,7 @@ struct server_context {
} else {
std::string prompt;
if (task.data.contains("prompt") && task.data.at("prompt").is_string()) {
json_value(task.data, "prompt", std::string());
prompt = json_value(task.data, "prompt", std::string());
}
slot = get_available_slot(prompt);
@@ -2039,7 +2039,12 @@ struct server_context {
prefix_tokens.insert(prefix_tokens.begin(), llama_token_bos(model)); // always add BOS
prefix_tokens.insert(prefix_tokens.end(), llama_token_suffix(model));
prefix_tokens.insert(prefix_tokens.end(), suffix_tokens.begin(), suffix_tokens.end());
prefix_tokens.push_back(llama_token_middle(model));
const llama_token middle_token = llama_token_middle(model);
if (middle_token >= 0) {
prefix_tokens.push_back(middle_token);
}
prompt_tokens = prefix_tokens;
} else {
prompt_tokens = tokenize(slot.prompt, system_prompt.empty()); // add BOS if there isn't system prompt
examples/sycl/win-build-sycl.bat
+3 -3
View File
@@ -13,16 +13,16 @@ if %errorlevel% neq 0 goto ERROR
:: for FP16
:: faster for long-prompt inference
:: cmake -G "MinGW Makefiles" .. -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release -DLLAMA_SYCL_F16=ON
:: cmake -G "MinGW Makefiles" .. -DLLAMA_SYCL=ON -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release -DLLAMA_SYCL_F16=ON
:: for FP32
cmake -G "MinGW Makefiles" .. -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release
cmake -G "Ninja" .. -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release
if %errorlevel% neq 0 goto ERROR
:: build example/main only
:: make main
:: build all binary
make -j
cmake --build . -j
if %errorlevel% neq 0 goto ERROR
cd ..