Instructions to use VladHong/llama-cpp-K2-FAST2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use VladHong/llama-cpp-K2-FAST2 with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: llama cli -hf VladHong/llama-cpp-K2-FAST2
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: llama cli -hf VladHong/llama-cpp-K2-FAST2
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: ./build/bin/llama cli -hf VladHong/llama-cpp-K2-FAST2
Use Docker
docker model run hf.co/VladHong/llama-cpp-K2-FAST2
- Llama new
- LM Studio
- Jan
- Ollama
How to use VladHong/llama-cpp-K2-FAST2 with Ollama:
ollama run hf.co/VladHong/llama-cpp-K2-FAST2
- Unsloth Desktop
- Docker Model Runner
How to use VladHong/llama-cpp-K2-FAST2 with Docker Model Runner:
docker model run hf.co/VladHong/llama-cpp-K2-FAST2
- Lemonade
How to use VladHong/llama-cpp-K2-FAST2 with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull VladHong/llama-cpp-K2-FAST2
Run and chat with the model
lemonade run user.llama-cpp-K2-FAST2-{{QUANT_TAG}}List all available models
lemonade list
- Atomic Chat
File size: 10,702 Bytes
c59fd93 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 | #include "llama-kv-cache-dsa-iswa.h"
#include "llama-impl.h"
#include "llama-batch.h"
#include "llama-model.h"
#include <algorithm>
#include <cassert>
//
// llama_kv_cache_dsa_iswa
//
llama_kv_cache_dsa_iswa::llama_kv_cache_dsa_iswa(
const llama_model & model,
ggml_type type_k,
ggml_type type_v,
bool v_trans,
bool offload,
bool swa_full,
bool unified,
uint32_t kv_size,
uint32_t n_seq_max,
uint32_t n_ubatch,
uint32_t n_pad,
const layer_filter_cb & filter_mla,
const layer_filter_cb & filter_lid,
const layer_reuse_cb & reuse) : unified(unified) {
const auto & hparams = model.hparams;
// chain filters
const layer_filter_cb filter_dsa = [&](int32_t il) {
if (filter_mla && !filter_mla(il)) {
return false;
}
return !hparams.is_swa(il);
};
const layer_filter_cb filter_swa = [&](int32_t il) {
if (filter_mla && !filter_mla(il)) {
return false;
}
return hparams.is_swa(il);
};
const uint32_t size_dsa = kv_size;
// note: the SWA cache is always padded to 256 for performance
// https://github.com/ggml-org/llama.cpp/issues/17037
uint32_t size_swa = GGML_PAD(std::min(size_dsa, hparams.n_swa*(unified ? n_seq_max : 1) + n_ubatch), 256);
// when using full-size SWA cache, we set the SWA cache size to be equal to the base cache size
if (swa_full) {
LLAMA_LOG_WARN("%s: using full-size SWA cache (ref: %s)\n",
__func__, "https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055");
size_swa = size_dsa;
}
LLAMA_LOG_INFO("%s: creating DSA KV cache, size = %u cells\n", __func__, size_dsa);
kv_dsa = std::make_unique<llama_kv_cache_dsa>(
model, type_k, type_v,
v_trans, offload, unified, size_dsa, n_seq_max, n_pad,
0, LLAMA_SWA_TYPE_NONE, filter_dsa, filter_lid, reuse);
LLAMA_LOG_INFO("%s: creating SWA KV cache, size = %u cells\n", __func__, size_swa);
kv_swa = std::make_unique<llama_kv_cache>(
model, hparams, type_k, type_v,
v_trans, offload, unified, size_swa, n_seq_max, n_pad,
hparams.n_swa, hparams.swa_type, nullptr, filter_swa, reuse, nullptr);
}
void llama_kv_cache_dsa_iswa::clear(bool data) {
kv_dsa->clear(data);
kv_swa->clear(data);
}
bool llama_kv_cache_dsa_iswa::seq_rm(llama_seq_id seq_id, llama_pos p0, llama_pos p1) {
bool res = true;
res = res & kv_dsa->seq_rm(seq_id, p0, p1);
res = res & kv_swa->seq_rm(seq_id, p0, p1);
return res;
}
void llama_kv_cache_dsa_iswa::seq_cp(llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) {
kv_dsa->seq_cp(seq_id_src, seq_id_dst, p0, p1);
kv_swa->seq_cp(seq_id_src, seq_id_dst, p0, p1);
}
void llama_kv_cache_dsa_iswa::seq_keep(llama_seq_id seq_id) {
kv_dsa->seq_keep(seq_id);
kv_swa->seq_keep(seq_id);
}
void llama_kv_cache_dsa_iswa::seq_add(llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos shift) {
kv_dsa->seq_add(seq_id, p0, p1, shift);
kv_swa->seq_add(seq_id, p0, p1, shift);
}
void llama_kv_cache_dsa_iswa::seq_div(llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) {
kv_dsa->seq_div(seq_id, p0, p1, d);
kv_swa->seq_div(seq_id, p0, p1, d);
}
llama_pos llama_kv_cache_dsa_iswa::seq_pos_min(llama_seq_id seq_id) const {
// the DSA cache is a superset of the SWA cache, so we can just check the SWA cache
return kv_swa->seq_pos_min(seq_id);
}
llama_pos llama_kv_cache_dsa_iswa::seq_pos_max(llama_seq_id seq_id) const {
return kv_swa->seq_pos_max(seq_id);
}
std::map<ggml_backend_buffer_type_t, size_t> llama_kv_cache_dsa_iswa::memory_breakdown() const {
std::map<ggml_backend_buffer_type_t, size_t> mb = kv_dsa->memory_breakdown();
for (const auto & buft_size : kv_swa->memory_breakdown()) {
mb[buft_size.first] += buft_size.second;
}
return mb;
}
llama_memory_context_ptr llama_kv_cache_dsa_iswa::init_batch(llama_batch_allocr & balloc, uint32_t n_ubatch, bool embd_all) {
GGML_UNUSED(embd_all);
// first try simple split
do {
if (!unified) {
// requires equal splits, so we skip the simple split
break;
}
balloc.split_reset();
std::vector<llama_ubatch> ubatches;
while (true) {
auto ubatch = balloc.split_simple(n_ubatch);
if (ubatch.n_tokens == 0) {
break;
}
ubatches.push_back(std::move(ubatch)); // NOLINT
}
if (balloc.get_n_used() < balloc.get_n_tokens()) {
// failed to find a suitable split
break;
}
auto sinfos_mla = kv_dsa->get_mla()->prepare(ubatches);
if (sinfos_mla.empty()) {
break;
}
auto sinfos_lid = kv_dsa->get_lid()->prepare(ubatches);
if (sinfos_lid.empty()) {
break;
}
auto sinfos_swa = kv_swa->prepare(ubatches);
if (sinfos_swa.empty()) {
break;
}
assert(sinfos_mla.size() == sinfos_swa.size());
return std::make_unique<llama_kv_cache_dsa_iswa_context>(
this, std::move(sinfos_mla), std::move(sinfos_lid), std::move(sinfos_swa), std::move(ubatches));
} while (false);
// if it fails, try equal split
do {
balloc.split_reset();
std::vector<llama_ubatch> ubatches;
while (true) {
auto ubatch = balloc.split_equal(n_ubatch, !unified, 0);
if (ubatch.n_tokens == 0) {
break;
}
ubatches.push_back(std::move(ubatch)); // NOLINT
}
if (balloc.get_n_used() < balloc.get_n_tokens()) {
// failed to find a suitable split
break;
}
auto sinfos_mla = kv_dsa->get_mla()->prepare(ubatches);
if (sinfos_mla.empty()) {
break;
}
auto sinfos_lid = kv_dsa->get_lid()->prepare(ubatches);
if (sinfos_lid.empty()) {
break;
}
auto sinfos_swa = kv_swa->prepare(ubatches);
if (sinfos_swa.empty()) {
break;
}
assert(sinfos_mla.size() == sinfos_swa.size());
return std::make_unique<llama_kv_cache_dsa_iswa_context>(
this, std::move(sinfos_mla), std::move(sinfos_lid), std::move(sinfos_swa), std::move(ubatches));
} while (false);
return std::make_unique<llama_kv_cache_dsa_iswa_context>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
}
llama_memory_context_ptr llama_kv_cache_dsa_iswa::init_full() {
return std::make_unique<llama_kv_cache_dsa_iswa_context>(this);
}
llama_memory_context_ptr llama_kv_cache_dsa_iswa::init_update(llama_context * lctx, bool optimize) {
return std::make_unique<llama_kv_cache_dsa_iswa_context>(this, lctx, optimize);
}
bool llama_kv_cache_dsa_iswa::get_can_shift() const {
return kv_dsa->get_can_shift() &&
kv_swa->get_can_shift() &&
kv_dsa->get_mla()->get_size() == kv_swa->get_size();
}
void llama_kv_cache_dsa_iswa::state_write(llama_io_write_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) const {
if ((flags & LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY) == 0) {
kv_dsa->state_write(io, seq_id, flags);
}
kv_swa->state_write(io, seq_id, flags);
}
void llama_kv_cache_dsa_iswa::state_read(llama_io_read_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) {
if ((flags & LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY) == 0) {
kv_dsa->state_read(io, seq_id, flags);
}
kv_swa->state_read(io, seq_id, flags);
}
llama_kv_cache_dsa * llama_kv_cache_dsa_iswa::get_dsa() const {
return kv_dsa.get();
}
llama_kv_cache * llama_kv_cache_dsa_iswa::get_swa() const {
return kv_swa.get();
}
//
// llama_kv_cache_dsa_iswa_context
//
llama_kv_cache_dsa_iswa_context::llama_kv_cache_dsa_iswa_context(llama_memory_status status) : status(status) {}
llama_kv_cache_dsa_iswa_context::llama_kv_cache_dsa_iswa_context(
llama_kv_cache_dsa_iswa * kv) :
ctx_dsa(kv->get_dsa()->init_full()),
ctx_swa(kv->get_swa()->init_full()),
status(llama_memory_status_combine(ctx_dsa->get_status(), ctx_swa->get_status())) {
}
llama_kv_cache_dsa_iswa_context::llama_kv_cache_dsa_iswa_context(
llama_kv_cache_dsa_iswa * kv,
llama_context * lctx,
bool optimize) :
ctx_dsa(kv->get_dsa()->init_update(lctx, optimize)),
ctx_swa(kv->get_swa()->init_update(lctx, optimize)),
status(llama_memory_status_combine(ctx_dsa->get_status(), ctx_swa->get_status())) {
}
llama_kv_cache_dsa_iswa_context::llama_kv_cache_dsa_iswa_context(
llama_kv_cache_dsa_iswa * kv,
slot_info_vec_t sinfos_mla,
slot_info_vec_t sinfos_lid,
slot_info_vec_t sinfos_swa,
std::vector<llama_ubatch> ubatches) :
ubatches(std::move(ubatches)),
// note: here we copy the ubatches. not sure if this is ideal
ctx_dsa(new llama_kv_cache_dsa_context(kv->get_dsa(), std::move(sinfos_mla), std::move(sinfos_lid), this->ubatches)),
ctx_swa(new llama_kv_cache_context(kv->get_swa(), std::move(sinfos_swa), this->ubatches)),
status(llama_memory_status_combine(ctx_dsa->get_status(), ctx_swa->get_status())) {
}
llama_kv_cache_dsa_iswa_context:: ~llama_kv_cache_dsa_iswa_context() = default;
bool llama_kv_cache_dsa_iswa_context::next() {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
ctx_dsa->next();
ctx_swa->next();
if (++i_next >= ubatches.size()) {
return false;
}
return true;
}
bool llama_kv_cache_dsa_iswa_context::apply() {
assert(!llama_memory_status_is_fail(status));
bool res = true;
res = res & ctx_dsa->apply();
res = res & ctx_swa->apply();
return res;
}
llama_memory_status llama_kv_cache_dsa_iswa_context::get_status() const {
return status;
}
const llama_ubatch & llama_kv_cache_dsa_iswa_context::get_ubatch() const {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
return ubatches[i_next];
}
const llama_kv_cache_dsa_context * llama_kv_cache_dsa_iswa_context::get_dsa() const {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
return static_cast<const llama_kv_cache_dsa_context *>(ctx_dsa.get());
}
const llama_kv_cache_context * llama_kv_cache_dsa_iswa_context::get_swa() const {
assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
return static_cast<const llama_kv_cache_context *>(ctx_swa.get());
}
|