echodict/llama.cpp
version https://git-lfs.github.com/spec/v1 oid sha256:cfc44b7ba25614df70e6b65e3341cae0310163bd32fd31a6b928a542df433faf size 30786
0610
1#include "llama-batch.h"2 3#include "llama-impl.h"4#include "llama-vocab.h"5#include "llama-memory.h"6 7#include <cassert>8#include <cstring>9#include <algorithm>10#include <sstream>11 12llama_batch_allocr::llama_batch_allocr(uint32_t n_pos_per_embd) : n_pos_per_embd(n_pos_per_embd) {13 const char * LLAMA_BATCH_DEBUG = getenv("LLAMA_BATCH_DEBUG");14 debug = LLAMA_BATCH_DEBUG ? atoi(LLAMA_BATCH_DEBUG) : 0;15 16 seq_pos.resize(LLAMA_MAX_SEQ);17 seq_cpl.resize(LLAMA_MAX_SEQ);18 for (auto & cur : seq_cpl) {19 cur.resize(LLAMA_MAX_SEQ);20 }21 22 seq_idx.resize(LLAMA_MAX_SEQ, -1);23}24 25bool llama_batch_allocr::init(26 const llama_batch & batch_inp,27 const llama_vocab & vocab,28 const llama_memory_i * memory,29 uint32_t n_embd,30 uint32_t n_seq_max,31 bool output_all) {32 clear();33 34 batch = batch_inp;35 36 this->vocab = &vocab;37 38 GGML_ASSERT(batch.n_tokens > 0);39 40 //41 // validate input batch42 //43 44 if (n_seq_max > LLAMA_MAX_SEQ) {45 LLAMA_LOG_ERROR("%s: n_seq_max = %d > %d\n", __func__, n_seq_max, LLAMA_MAX_SEQ);46 return false;47 }48 49 if (batch.token) {50 for (int32_t i = 0; i < batch.n_tokens; ++i) {51 if (batch.token[i] < 0 || (uint32_t) batch.token[i] >= vocab.n_tokens()) {52 LLAMA_LOG_ERROR("%s: invalid token[%d] = %d\n", __func__, i, batch.token[i]);53 return false;54 }55 }56 }57 58 if (batch.seq_id) {59 for (int32_t i = 0; i < batch.n_tokens; ++i) {60 for (int32_t s = 0; s < batch.n_seq_id[i]; ++s) {61 if (batch.seq_id && (batch.seq_id[i][s] < 0 || batch.seq_id[i][s] >= (llama_seq_id) n_seq_max)) {62 LLAMA_LOG_ERROR("%s: invalid seq_id[%d][%d] = %d >= %d\n", __func__, i, s, batch.seq_id[i][s], (llama_seq_id) n_seq_max);63 return false;64 }65 }66 }67 }68 69 //70 // auto-generate missing fields71 //72 73 if (!batch.n_seq_id) {74 n_seq_id.resize(batch.n_tokens);75 for (int32_t i = 0; i < batch.n_tokens; i++) {76 n_seq_id[i] = seq_id_0.size();77 }78 batch.n_seq_id = n_seq_id.data();79 }80 81 if (!batch.seq_id) {82 seq_id.resize(batch.n_tokens + 1);83 seq_id[batch.n_tokens] = NULL;84 for (int32_t i = 0; i < batch.n_tokens; i++) {85 seq_id[i] = seq_id_0.data();86 }87 batch.seq_id = seq_id.data();88 }89 90 if (!batch.pos) {91 pos.resize(batch.n_tokens);92 93 // initialize the starting position for each sequence based on the positions in the memory94 llama_pos p0[LLAMA_MAX_SEQ];95 for (uint32_t s = 0; s < n_seq_max; ++s) {96 if (!memory) {97 // if no memory -> start from 098 p0[s] = 0;99 } else {100 p0[s] = memory->seq_pos_max(s) + 1;101 }102 }103 104 for (int32_t i = 0; i < batch.n_tokens; i++) {105 const llama_seq_id seq_id = batch.seq_id[i][0];106 107 pos[i] = p0[seq_id];108 109 // update the starting position for all sequences that are assigned to the this token110 for (int32_t s = 0; s < batch.n_seq_id[i]; ++s) {111 const llama_seq_id seq_id = batch.seq_id[i][s];112 113 p0[seq_id] = pos[i] + 1;114 }115 }116 117 batch.pos = pos.data();118 }119 120 if (!batch.logits) {121 if (output_all) {122 // return the output for all tokens123 output.resize(batch.n_tokens, true);124 } else {125 // return the output only for the last token126 output.resize(batch.n_tokens, false);127 output[output.size() - 1] = true;128 }129 130 batch.logits = output.data();131 } else if (output_all) {132 bool warn = false;133 134 for (int32_t i = 0; i < batch.n_tokens; ++i) {135 if (batch.logits[i] == 0) {136 warn = true;137 }138 }139 140 if (warn) {141 LLAMA_LOG_WARN("%s: embeddings required but some input tokens were not marked as outputs -> overriding\n", __func__);142 143 output.resize(batch.n_tokens, true);144 batch.logits = output.data();145 }146 }147 148 //149 // compute stats150 //151 152 this->n_embd = n_embd;153 this->n_seq_max = n_seq_max;154 155 // count the outputs in this batch156 for (int32_t i = 0; i < batch.n_tokens; ++i) {157 n_outputs += batch.logits[i] != 0;158 }159 160 has_cpl = false;161 162 // determine coupled sequences163 // these are pairs of sequences that have at least one token in the input batch that is assigned to both of them164 for (int32_t i = 0; i < batch.n_tokens; ++i) {165 const llama_seq_id s0 = batch.seq_id[i][0];166 167 for (int32_t s = 0; s < batch.n_seq_id[i]; ++s) {168 const llama_seq_id s1 = batch.seq_id[i][s];169 170 seq_pos[s1].insert(batch.pos[i]);171 172 if (s > 0) {173 // mark that sequence s1 is coupled to s0174 seq_cpl[s1][s0] = true;175 176 // note: tracking the other way around is not necessary for now177 //seq_cpl[s0][s1] = true;178 179 has_cpl = true;180 }181 }182 }183 184 // precompute the sequence sets for each token and determine the unique sequence ids that participate in the batch185 {186 seq_set_t seq_set_unq;187 188 for (int32_t i = 0; i < batch.n_tokens; ++i) {189 seq_set_t cur;190 for (int32_t s = 0; s < batch.n_seq_id[i]; ++s) {191 const llama_seq_id seq_id = batch.seq_id[i][s];192 193 cur .set(seq_id);194 seq_set_unq.set(seq_id);195 }196 197 seq_set.push_back(cur);198 seq_set_map[cur].push_back(i);199 }200 201 for (uint32_t s = 0; s < n_seq_max; ++s) {202 if (seq_set_unq.test(s)) {203 seq_idx[s] = seq_id_unq.size();204 seq_id_unq.push_back(s);205 }206 }207 }208 209 if (debug > 0) {210 LLAMA_LOG_DEBUG("%s: input batch info:\n", __func__);211 212 llama_ubatch ubatch {213 /*.b_equal_seqs =*/ false,214 /*.n_tokens =*/ (uint32_t) batch.n_tokens,215 /*.n_seq_tokens =*/ (uint32_t) 1,216 /*.n_seqs =*/ (uint32_t) batch.n_tokens,217 /*.n_seqs_unq =*/ (uint32_t) this->seq_id_unq.size(),218 /*.n_pos =*/ n_pos_per_embd,219 /*.token =*/ batch.token,220 /*.embd =*/ batch.embd,221 /*.pos =*/ batch.pos,222 /*.n_seq_id =*/ batch.n_seq_id,223 /*.seq_id =*/ batch.seq_id,224 /*.seq_id_unq =*/ this->seq_id_unq.data(),225 /*.seq_idx =*/ this->seq_idx.data(),226 /*.output =*/ batch.logits,227 /*.data =*/ {},228 };229 230 ubatch_print(ubatch, debug);231 232 LLAMA_LOG_DEBUG("%s: seq = [\n", __func__);233 for (int s0 = 0; s0 < (int) seq_pos.size(); ++s0) {234 if (seq_pos[s0].empty()) {235 continue;236 }237 238 std::stringstream ss;239 for (int s1 = 0; s1 < (int) seq_cpl[s0].size(); ++s1) {240 if (seq_cpl[s0][s1]) {241 ss << s1 << " ";242 }243 }244 245 LLAMA_LOG_DEBUG("%s: %4d: pos = [%4d, %4d], cpl = %s\n",246 __func__, s0, seq_pos_min(s0), seq_pos_max(s0), ss.str().empty() ? "-" : ss.str().c_str());247 }248 LLAMA_LOG_DEBUG("%s: ]\n", __func__);249 }250 251 //252 // consistency checks253 //254 255 if (n_pos_per_embd > 1) {256 // M-RoPE case: allow position to "jump" forward only (non-continuous positions are allowed)257 for (uint32_t s = 0; s < n_seq_max; ++s) {258 if (seq_pos[s].empty()) {259 continue;260 }261 262 const llama_pos p0 = memory ? memory->seq_pos_max(s) : -1;263 264 if (batch.token) {265 if (p0 >= 0 && p0 >= seq_pos_min(s)) {266 LLAMA_LOG_ERROR(267 "%s: the tokens of sequence %d in the input batch have inconsistent sequence positions:\n"268 " - the last position stored in the memory module of the context (i.e. the KV cache) for sequence %d is X = %d\n"269 " - the tokens for sequence %d in the input batch have a starting position of Y = %d\n"270 " for M-RoPE, it is required that the position satisfies: X < Y\n",271 __func__, s, s, p0, s, seq_pos_min(s));272 273 return false;274 }275 } else {276 // embedding inputs can have overlapping positions277 if (p0 >= 0 && p0 > seq_pos_min(s)) {278 LLAMA_LOG_ERROR(279 "%s: the tokens of sequence %d in the input batch have inconsistent sequence positions:\n"280 " - the last position stored in the memory module of the context (i.e. the KV cache) for sequence %d is X = %d\n"281 " - the tokens for sequence %d in the input batch have a starting position of Y = %d\n"282 " for M-RoPE, it is required that the position satisfies: X <= Y\n",283 __func__, s, s, p0, s, seq_pos_min(s));284 285 return false;286 }287 }288 }289 } else {290 for (uint32_t s = 0; s < n_seq_max; ++s) {291 if (seq_pos[s].empty()) {292 continue;293 }294 295 const llama_pos p0 = memory ? memory->seq_pos_max(s) : -1;296 297 if (p0 >= 0) {298 bool ok = true;299 300 if (seq_pos_min(s) != p0 + 1) {301 ok = false;302 }303 304 if (!ok) {305 LLAMA_LOG_ERROR(306 "%s: the tokens of sequence %d in the input batch have inconsistent sequence positions:\n"307 " - the last position stored in the memory module of the context (i.e. the KV cache) for sequence %d is X = %d\n"308 " - the tokens for sequence %d in the input batch have a starting position of Y = %d\n"309 " it is required that the sequence positions remain consecutive: Y = X + 1\n",310 __func__, s, s, p0, s, seq_pos_min(s));311 312 return false;313 }314 }315 316 if (seq_pos_max(s) - seq_pos_min(s) + 1 > (int) seq_pos[s].size()) {317 LLAMA_LOG_ERROR("%s: sequence %d positions are not continuous\n", __func__, s);318 return false;319 }320 }321 }322 323 if (memory) {324 for (uint32_t s0 = 0; s0 < n_seq_max; ++s0) {325 for (uint32_t s1 = 0; s1 < n_seq_max; ++s1) {326 if (seq_cpl[s0][s1]) {327 if (memory->seq_pos_min(s0) != memory->seq_pos_min(s1) ||328 memory->seq_pos_max(s0) != memory->seq_pos_max(s1)) {329 LLAMA_LOG_ERROR("%s: sequence %d is coupled to %d in the input batch, but have divereged\n", __func__, s0, s1);330 return false;331 }332 }333 }334 }335 }336 337 // disallow partial sequence sub-sets:338 //339 // invalid: x340 // i: 0 1 2 ...341 // ---------------------------------------342 // seq_id[i][0]: 0 0 1343 // seq_id[i][1]: 1 1 2344 // seq_id[i][2]: 2345 //346 // disallow decreasing sequence positions:347 //348 // invalid: x349 // i: 0 1 2 3 4 5 6 ...350 // ---------------------------------------351 // pos[i]: 4 5 0 1 6 2 3352 // seq_id[i][0]: 0 0 1 1 0 1 0353 //354 {355 seq_set_t cur_seq_set[LLAMA_MAX_SEQ];356 for (uint32_t s = 0; s < n_seq_max; ++s) {357 cur_seq_set[s].set();358 }359 360 llama_pos cur_seq_pos[LLAMA_MAX_SEQ];361 for (uint32_t s = 0; s < n_seq_max; ++s) {362 cur_seq_pos[s] = -1;363 }364 365 for (int32_t i = 0; i < batch.n_tokens; ++i) {366 const llama_pos pos = batch.pos[i];367 368 for (int32_t s = 0; s < batch.n_seq_id[i]; ++s) {369 const llama_seq_id seq_id = batch.seq_id[i][s];370 371 cur_seq_set[seq_id] &= seq_set[i];372 373 if (cur_seq_set[seq_id].none()) {374 LLAMA_LOG_ERROR("%s: sequence %d belongs to incompatible sequence sets (not allowed)\n", __func__, seq_id);375 return false;376 }377 378 if (pos < cur_seq_pos[seq_id]) {379 LLAMA_LOG_ERROR("%s: sequence %d positions are decreasing (not allowed)\n", __func__, seq_id);380 return false;381 }382 }383 }384 }385 386 split_reset();387 388 return true;389}390 391llama_ubatch llama_batch_allocr::ubatch_reserve(uint32_t n_seq_tokens, uint32_t n_seqs) {392 const uint32_t n_tokens = n_seq_tokens*n_seqs;393 394 clear();395 split_reset();396 397 const int64_t n_pos_all = (int64_t) n_tokens*n_pos_per_embd;398 399 auto udata = std::make_shared<llama_ubatch::data_t>();400 401 udata->token .resize(n_tokens);402 udata->embd .clear();403 udata->pos .resize(n_pos_all);404 udata->n_seq_id .resize(n_tokens);405 udata->seq_id .resize(n_tokens);406 udata->seq_id_unq.resize(0);407 udata->seq_idx .resize(LLAMA_MAX_SEQ, -1);408 udata->output .resize(n_tokens);409 410 for (uint32_t s = 0; s < n_seqs; ++s) {411 udata->seq_idx[s] = s;412 udata->seq_id_unq.push_back(s);413 }414 415 llama_ubatch res {416 /*.b_equal_seqs =*/ true,417 /*.n_tokens =*/ n_tokens,418 /*.n_seq_tokens =*/ n_seq_tokens,419 /*.n_seqs =*/ n_seqs,420 /*.n_seqs_unq =*/ n_seqs,421 /*.n_pos =*/ n_pos_per_embd,422 423 /*.token =*/ udata->token.data(),424 /*.embd =*/ nullptr,425 /*.pos =*/ udata->pos.data(),426 /*.n_seq_id =*/ udata->n_seq_id.data(),427 /*.seq_id =*/ udata->seq_id.data(),428 /*.seq_id_unq =*/ udata->seq_id_unq.data(),429 /*.seq_idx =*/ udata->seq_idx.data(),430 /*.output =*/ udata->output.data(),431 /*.data =*/ std::move(udata),432 };433 434 return res;435}436 437const llama_batch & llama_batch_allocr::get_batch() const {438 return batch;439}440 441uint32_t llama_batch_allocr::get_n_tokens() const {442 return batch.n_tokens;443}444 445uint32_t llama_batch_allocr::get_n_outputs() const {446 return n_outputs;447}448 449uint32_t llama_batch_allocr::get_n_used() const {450 return n_used;451}452 453std::vector<int32_t> & llama_batch_allocr::get_out_ids() {454 return out_ids;455}456 457llama_pos llama_batch_allocr::seq_pos_min(llama_seq_id seq_id) const {458 return seq_pos[seq_id].empty() ? -1 : *seq_pos[seq_id].begin();459}460 461llama_pos llama_batch_allocr::seq_pos_max(llama_seq_id seq_id) const {462 return seq_pos[seq_id].empty() ? -1 : *seq_pos[seq_id].rbegin();463}464 465void llama_batch_allocr::split_reset() {466 out_ids.clear();467 468 n_used = 0;469 470 used.clear();471 used.resize(get_n_tokens(), false);472}473 474llama_ubatch llama_batch_allocr::split_simple(uint32_t n_ubatch) {475 // find the first unused token476 uint32_t cur_idx = 0;477 while (cur_idx < used.size() && used[cur_idx]) {478 ++cur_idx;479 }480 481 // we are done482 if (cur_idx >= used.size()) {483 return {};484 }485 486 std::vector<int32_t> idxs;487 488 while (true) {489 idxs.push_back(cur_idx);490 491 used[cur_idx] = true;492 ++n_used;493 494 ++cur_idx;495 496 if (cur_idx >= used.size()) {497 break;498 }499 500 if (idxs.size() >= n_ubatch) {501 break;502 }503 }504 505 return ubatch_add(idxs, idxs.size(), false);506}507 508llama_ubatch llama_batch_allocr::split_equal(uint32_t n_ubatch, bool sequential) {509 if (sequential && has_cpl) {510 LLAMA_LOG_ERROR("%s: sequential split is not supported when there are coupled sequences in the input batch (you may need to use the -kvu flag)\n", __func__);511 512 return {};513 }514 515 std::vector<seq_set_t> cur_seq_set;516 517 llama_seq_id last_seq_id = -1;518 519 // determine the non-overlapping sequence sets participating in this ubatch520 for (int32_t i = 0; i < batch.n_tokens; ++i) {521 if (used[i]) {522 continue;523 }524 525 bool add = true;526 527 for (uint32_t s = 0; s < cur_seq_set.size(); ++s) {528 // no overlap with existing sequence sets:529 if (!(cur_seq_set[s] & seq_set[i]).none()) {530 add = false;531 break;532 }533 }534 535 // accept only increasing sequence ids536 if (sequential) {537 add = add && (cur_seq_set.empty() || batch.seq_id[i][0] == last_seq_id + 1);538 }539 540 if (add) {541 cur_seq_set.push_back(seq_set[i]);542 543 last_seq_id = batch.seq_id[i][0];544 545 if (cur_seq_set.size() > n_ubatch) {546 break;547 }548 }549 }550 551 const uint32_t n_seqs = cur_seq_set.size();552 553 // we are done554 if (n_seqs == 0) {555 return {};556 }557 558 // the current batch index of each sequence set559 std::vector<int32_t> cur_idx(n_seqs, 0);560 561 for (uint32_t s = 0; s < n_seqs; ++s) {562 while (used[seq_set_map[cur_seq_set[s]][cur_idx[s]]]) {563 ++cur_idx[s];564 }565 }566 567 // the list of batch indices for each sequence set568 // at the end we will concat these to get the final ubatch569 std::vector<idx_vec_t> idxs_per_seq(n_seqs);570 571 while (true) {572 // we can only add new n_seq_tokens tokens if all the sequence sets have at least one more unused token and573 // if we haven't reached n_ubatch574 bool can_expand = true;575 576 for (uint32_t s = 0; s < n_seqs; ++s) {577 if (cur_idx[s] >= (int32_t) seq_set_map[cur_seq_set[s]].size()) {578 can_expand = false;579 break;580 }581 }582 583 if (!can_expand) {584 break;585 }586 587 for (uint32_t s = 0; s < n_seqs; ++s) {588 const int32_t idx = seq_set_map[cur_seq_set[s]][cur_idx[s]];589 590 idxs_per_seq[s].push_back(idx);591 592 used[idx] = true;593 ++n_used;594 595 ++cur_idx[s];596 }597 598 if ((idxs_per_seq[0].size() + 1)*n_seqs > n_ubatch) {599 break;600 }601 }602 603 // concat the per-sequence-set lists604 std::vector<int32_t> idxs;605 606 for (uint32_t s = 0; s < n_seqs; ++s) {607 idxs.insert(idxs.end(), idxs_per_seq[s].begin(), idxs_per_seq[s].end());608 }609 610 return ubatch_add(idxs, n_seqs, true);611}612 613llama_ubatch llama_batch_allocr::split_seq(uint32_t n_ubatch) {614 // find the first unused token615 uint32_t cur_idx = 0;616 while (cur_idx < used.size() && used[cur_idx]) {617 ++cur_idx;618 }619 620 // we are done621 if (cur_idx >= used.size()) {622 return {};623 }624 625 // this is the starting sequence set626 // we allow adding tokens only if their sequence set is a subset of the current sequence set627 auto cur_seq_set = seq_set[cur_idx];628 629 std::vector<int32_t> idxs;630 631 while (true) {632 idxs.push_back(cur_idx);633 634 used[cur_idx] = true;635 ++n_used;636 637 if (idxs.size() >= n_ubatch) {638 break;639 }640 641 do {642 ++cur_idx;643 } while (cur_idx < get_n_tokens() && (used[cur_idx] || ((cur_seq_set & seq_set[cur_idx]) != seq_set[cur_idx])));644 645 if (cur_idx == get_n_tokens()) {646 break;647 }648 649 cur_seq_set = seq_set[cur_idx];650 }651 652 return ubatch_add(idxs, 1, true);653}654 655void llama_batch_allocr::clear() {656 n_outputs = 0;657 658 batch = {};659 660 pos .clear();661 n_seq_id .clear();662 seq_id .clear();663 seq_id_unq.clear();664 output .clear();665 666 for (auto & cur : seq_pos) {667 cur.clear();668 }669 670 for (auto & cur : seq_cpl) {671 std::fill(cur.begin(), cur.end(), false);672 }673 674 seq_set.clear();675 676 seq_set_map.clear();677 678 std::fill(seq_idx.begin(), seq_idx.end(), -1);679}680 681llama_ubatch llama_batch_allocr::ubatch_add(const std::vector<int32_t> & idxs, uint32_t n_seqs, bool equal_seqs) {682 const uint32_t n_tokens = idxs.size();683 684 assert(n_tokens%n_seqs == 0);685 686 auto udata = std::make_shared<llama_ubatch::data_t>();687 688 const int64_t n_embd_all = batch.embd ? (int64_t) n_tokens*n_embd : 0;689 const int64_t n_pos_all = (int64_t) n_tokens*n_pos_per_embd;690 691 udata->token .resize(n_tokens);692 udata->embd .resize(n_embd_all);693 udata->pos .resize(n_pos_all);694 udata->n_seq_id .resize(n_tokens);695 udata->seq_id .resize(n_tokens);696 udata->seq_id_unq.resize(0);697 udata->seq_idx .resize(LLAMA_MAX_SEQ, -1);698 udata->output .resize(n_tokens);699 700 udata->seq_id_data.reserve(n_tokens);701 702 seq_set_t seq_set_unq;703 704 for (size_t i = 0; i < idxs.size(); ++i) {705 if (batch.token) {706 udata->token[i] = batch.token[idxs[i]];707 }708 709 if (batch.embd) {710 memcpy(udata->embd.data() + i*n_embd, batch.embd + (int64_t) idxs[i]*n_embd, n_embd*sizeof(float));711 }712 713 for (size_t j = 0; j < (size_t)n_pos_per_embd; ++j) {714 // if we are using M-RoPE715 // if the current batch is text, we need to broadcast the same position across all RoPE sections716 // otherwise, the input batch is image embeddings, we copy the positions as-is717 // if we are not using M-RoPE, there is only one position per token (this loop runs only once)718 size_t src_off = batch.token ? 0 : j*batch.n_tokens;719 udata->pos[j*n_tokens + i] = batch.pos[src_off + idxs[i]];720 }721 722 udata->n_seq_id[i] = batch.n_seq_id[idxs[i]];723 udata->output[i] = batch.logits[idxs[i]];724 725 for (int s = 0; s < udata->n_seq_id[i]; ++s) {726 const llama_seq_id seq_id = batch.seq_id[idxs[i]][s];727 728 udata->seq_id_data.push_back(seq_id);729 seq_set_unq.set(seq_id);730 }731 732 if (udata->output[i]) {733 out_ids.push_back(idxs[i]);734 }735 }736 737 llama_seq_id * seq_id_ptr = udata->seq_id_data.data();738 for (size_t i = 0; i < idxs.size(); ++i) {739 udata->seq_id[i] = seq_id_ptr;740 seq_id_ptr += udata->n_seq_id[i];741 }742 743 for (uint32_t s = 0; s < n_seq_max; ++s) {744 if (seq_set_unq.test(s)) {745 udata->seq_idx[s] = udata->seq_id_unq.size();746 udata->seq_id_unq.push_back(s);747 }748 }749 750 llama_ubatch res {751 /*.b_equal_seqs =*/ equal_seqs,752 /*.n_tokens =*/ n_tokens,753 /*.n_seq_tokens =*/ n_tokens/n_seqs,754 /*.n_seqs =*/ n_seqs,755 /*.n_seqs_unq =*/ (uint32_t) udata->seq_id_unq.size(),756 /*.n_pos =*/ n_pos_per_embd,757 758 /*.token =*/ batch.token ? udata->token.data() : nullptr,759 /*.embd =*/ batch.embd ? udata->embd.data() : nullptr,760 /*.pos =*/ udata->pos.data(),761 /*.n_seq_id =*/ udata->n_seq_id.data(),762 /*.seq_id =*/ udata->seq_id.data(),763 /*.seq_id_unq =*/ udata->seq_id_unq.data(),764 /*.seq_idx =*/ udata->seq_idx.data(),765 /*.output =*/ udata->output.data(),766 /*.data =*/ std::move(udata),767 };768 769 if (debug > 0) {770 LLAMA_LOG_DEBUG("%s: added ubatch to split:\n", __func__);771 772 ubatch_print(res, debug);773 }774 775 return res;776}777 778void llama_batch_allocr::ubatch_print(const llama_ubatch & ubatch, int debug) {779 if (debug > 0) {780 LLAMA_LOG_DEBUG("%s: equal_seqs = %d\n", __func__, ubatch.equal_seqs());781 LLAMA_LOG_DEBUG("%s: n_tokens = %d\n", __func__, ubatch.n_tokens);782 LLAMA_LOG_DEBUG("%s: n_seq_tokens = %d\n", __func__, ubatch.n_seq_tokens);783 LLAMA_LOG_DEBUG("%s: n_seqs = %d\n", __func__, ubatch.n_seqs);784 LLAMA_LOG_DEBUG("%s: n_seqs_unq = %d\n", __func__, ubatch.n_seqs_unq);785 786 std::stringstream ss_seq_id_unq;787 std::stringstream ss_seq_idx;788 789 ss_seq_id_unq << "[ ";790 ss_seq_idx << "[";791 792 for (uint32_t s = 0; s < ubatch.n_seqs_unq; ++s) {793 ss_seq_id_unq << ubatch.seq_id_unq[s] << " ";794 }795 796 for (uint32_t s = 0; s < LLAMA_MAX_SEQ; ++s) {797 if (ubatch.seq_idx[s] >= 0) {798 ss_seq_idx << ubatch.seq_idx[s]%10;799 } else {800 ss_seq_idx << ".";801 }802 }803 804 ss_seq_id_unq << "]";805 ss_seq_idx << "]";806 807 LLAMA_LOG_DEBUG("%s: token = %p\n", __func__, (void *) ubatch.token);808 LLAMA_LOG_DEBUG("%s: embd = %p\n", __func__, (void *) ubatch.embd);809 LLAMA_LOG_DEBUG("%s: pos = %p\n", __func__, (void *) ubatch.pos);810 LLAMA_LOG_DEBUG("%s: n_seq_id = %p\n", __func__, (void *) ubatch.n_seq_id);811 LLAMA_LOG_DEBUG("%s: seq_id = %p\n", __func__, (void *) ubatch.seq_id);812 LLAMA_LOG_DEBUG("%s: seq_id_unq = %s\n", __func__, ss_seq_id_unq.str().c_str());813 LLAMA_LOG_DEBUG("%s: seq_idx = %s\n", __func__, ss_seq_idx.str().c_str());814 LLAMA_LOG_DEBUG("%s: output = %p\n", __func__, (void *) ubatch.output);815 LLAMA_LOG_DEBUG("%s: n_outputs = %d\n", __func__, n_outputs);816 817 if (debug > 1) {818 int seq_id_max = 0;819 for (uint32_t i = 0; i < ubatch.n_tokens; ++i) {820 for (int s = 0; s < ubatch.n_seq_id[i]; ++s) {821 for (int s = 0; s < ubatch.n_seq_id[i]; ++s) {822 seq_id_max = std::max(seq_id_max, ubatch.seq_id[i][s]);823 }824 }825 }826 ++seq_id_max;827 828 LLAMA_LOG_DEBUG("%s: token = [\n", __func__);829 for (uint32_t i = 0; i < ubatch.n_tokens; ++i) {830 std::vector<int8_t> seq_id(seq_id_max);831 832 for (int s = 0; s < ubatch.n_seq_id[i]; ++s) {833 seq_id[ubatch.seq_id[i][s]] = 1;834 }835 836 std::stringstream ss;837 for (int s = 0; s < seq_id_max; ++s) {838 if (seq_id[s]) {839 ss << s%10;840 } else {841 ss << ".";842 }843 }844 845 if (ubatch.token) {846 LLAMA_LOG_DEBUG("%s: %4d: id = %6d (%16s), pos = %4d, n_seq_id = %2d, seq_id = [%s], output = %d\n",847 __func__, i, ubatch.token[i], vocab->token_to_piece(ubatch.token[i]).c_str(),848 ubatch.pos[i], ubatch.n_seq_id[i], ss.str().c_str(), ubatch.output[i]);849 } else {850 LLAMA_LOG_DEBUG("%s: %4d: [embd], pos = %4d, n_seq_id = %2d, seq_id = [%s], output = %d\n",851 __func__, i, ubatch.pos[i], ubatch.n_seq_id[i], ss.str().c_str(), ubatch.output[i]);852 }853 }854 LLAMA_LOG_DEBUG("%s: ]\n", __func__);855 }856 }857}858 859//860// interface implementation861//862 863struct llama_batch llama_batch_get_one(864 llama_token * tokens,865 int32_t n_tokens) {866 return {867 /*n_tokens =*/ n_tokens,868 /*tokens =*/ tokens,869 /*embd =*/ nullptr,870 /*pos =*/ nullptr,871 /*n_seq_id =*/ nullptr,872 /*seq_id =*/ nullptr,873 /*logits =*/ nullptr,874 };875}876 877struct llama_batch llama_batch_init(int32_t n_tokens_alloc, int32_t embd, int32_t n_seq_max) {878 llama_batch batch = {879 /*n_tokens =*/ 0,880 /*tokens =*/ nullptr,881 /*embd =*/ nullptr,882 /*pos =*/ nullptr,883 /*n_seq_id =*/ nullptr,884 /*seq_id =*/ nullptr,885 /*logits =*/ nullptr,886 };887 888 if (embd) {889 batch.embd = (float *) malloc(sizeof(float) * n_tokens_alloc * embd);890 } else {891 batch.token = (llama_token *) malloc(sizeof(llama_token) * n_tokens_alloc);892 }893 894 batch.pos = (llama_pos *) malloc(sizeof(llama_pos) * n_tokens_alloc);895 batch.n_seq_id = (int32_t *) malloc(sizeof(int32_t) * n_tokens_alloc);896 batch.seq_id = (llama_seq_id **) malloc(sizeof(llama_seq_id *) * (n_tokens_alloc + 1));897 for (int i = 0; i < n_tokens_alloc; ++i) {898 batch.seq_id[i] = (llama_seq_id *) malloc(sizeof(llama_seq_id) * n_seq_max);899 }900 batch.seq_id[n_tokens_alloc] = nullptr;901 902 batch.logits = (int8_t *) malloc(sizeof(int8_t) * n_tokens_alloc);903 904 return batch;905}906 907void llama_batch_free(struct llama_batch batch) {908 if (batch.token) free(batch.token);909 if (batch.embd) free(batch.embd);910 if (batch.pos) free(batch.pos);911 if (batch.n_seq_id) free(batch.n_seq_id);912 if (batch.seq_id) {913 for (int i = 0; batch.seq_id[i] != nullptr; ++i) {914 free(batch.seq_id[i]);915 }916 free(batch.seq_id);917 }918 if (batch.logits) free(batch.logits);919}920