vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : custom attention mask + parallel decoding + no context swaps (#3228)

Browse source 
* tests : verify that RoPE is "additive"

* llama : replace ggml_diag_mask_inf with ggml_add (custom -inf mask)

* ggml : ggml_rope now takes a vector with positions instead of n_past

* metal : add rope_f16 kernel + optimize cpy kernels

* llama : unified KV cache + batch inference API

* llama : add new llama_decode() API that works with llama_batch

* llama : add cell_max heuristic for more efficient kv_cache

* llama : extend llama_kv_cache API

* llama : more robust cell_max heuristic + wip shift

* metal : disable concurrency optimization

* llama : add llama_kv_cache_shift_seq + no more context swaps

* llama : apply K-cache roping for Falcon and Baichuan

* speculative : fix KV cache management

* parallel : example for serving multiple users in parallel

* parallel : disable hot-plug to avoid cache fragmentation

* fixes : speculative KV cache + llama worst-case graph

* llama : extend batch API to select which logits to output

* llama : fix worst case graph build

* ggml-cuda : update rope implementation for parallel decoding (#3254)

* ggml-cuda : update rope implementation for parallel decoding

* better solution for p0 computation

* fix rope

* simpler rope implementation

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* make : add parallel to build + fix static functions in llama.cpp

* simple : fix token counting

* parallel : various improvements

* llama : fix cell_max logic + rename functions

* parallel : try smaller batches when the KV cache is fragmented

* parallel : fix sequence termination criteria

* llama : silence errors KV cache errors

* parallel : remove new line from prompt

* parallel : process system prompt once + configurable paramters + llama API

* parallel : remove question with short answers

* parallel : count cache misses

* parallel : print misses on each request

* parallel : minor

* llama : fix n_kv to never become 0

* parallel : rename hot-plug to continuous-batching

* llama : improve llama_batch API + simplify parallel example

* simple : add parallel decoding support

* simple : improve comments + free batch

* ggml-cuda : add rope f16, restore performance with parallel decoding (#3272)

* ggml-cuda : add rope f16, restore performance

* offload KQ_mask with all models

* fix rope shift

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* llama : disable MPI for now

ggml-ci

* train : make KQ_pos memory buffer permanent via dummy scale op

* ggml : revert change to ggml_cpy, add ggml_cont_Nd instead (#3275)

ggml-ci

* parallel : fix bug (extra BOS) + smaller token_prev array

* parallel : fix cases where the input prompts can overflow the batch

* parallel : add disabled experimental batch chunking in powers of two

* llama : llama.h formatting + comments

* simple : add README.md

* llama : fix kv cache heuristic when context is less than 32

* parallel : fix crash when `-n -1`

* llama : simplify returns if/else branches

* metal : use mm kernels for batch size > 2

* examples : utilize new llama_get_logits_ith()

* examples : add example for batched decoding

* examples : do not eval prompt 2 times (close #3348)

* server : clear the KV cache beyond n_past before llama_decode

* server : avoid context swaps by shifting the KV cache

---------

Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
Georgi Gerganov 2023-09-28 19:04:36 +03:00 • committed by GitHub
parent 45855b3f1c
commit ec893798b7
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
35 changed files with 2700 additions and 673 deletions
Download patch file Download diff file Expand all files Collapse all files

36
examples/main/main.cpp
View file

@ -124,7 +124,7 @@ int main(int argc, char ** argv) {
console::init(params.simple_io, params.use_color);
atexit([]() { console::cleanup(); });
if (params.perplexity) {
if (params.logits_all) {
printf("\n************\n");
printf("%s: please use the 'perplexity' tool for perplexity calculations\n", __func__);
printf("************\n\n");
@ -200,15 +200,6 @@ int main(int argc, char ** argv) {
params.n_threads, std::thread::hardware_concurrency(), llama_print_system_info());
}
// export the cgraph and exit
if (params.export_cgraph) {
llama_eval_export(ctx, "llama.ggml");
llama_free(ctx);
llama_free_model(model);
return 0;
}
std::string path_session = params.path_prompt_cache;
std::vector<llama_token> session_tokens;
@ -508,18 +499,23 @@ int main(int argc, char ** argv) {
break;
}
const int n_left = n_past - params.n_keep;
LOG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d\n", n_past, n_left, n_ctx, params.n_keep);
const int n_left = n_past - params.n_keep - 1;
const int n_discard = n_left/2;
// always keep the first token - BOS
n_past = std::max(1, params.n_keep);
n_past_guidance = std::max(1, params.n_keep + guidance_offset);
LOG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
n_past, n_left, n_ctx, params.n_keep, n_discard);
llama_kv_cache_seq_rm (ctx, 0, params.n_keep + 1 , params.n_keep + n_discard + 1);
llama_kv_cache_seq_shift(ctx, 0, params.n_keep + 1 + n_discard, n_past, -n_discard);
n_past -= n_discard;
if (ctx_guidance) {
n_past_guidance -= n_discard;
}
LOG("after swap: n_past = %d, n_past_guidance = %d\n", n_past, n_past_guidance);
// insert n_left/2 tokens at the start of embd from last_tokens
embd.insert(embd.begin(), last_tokens.begin() + n_ctx - n_left/2 - embd.size(), last_tokens.end() - embd.size());
LOG("embd: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd));
LOG("clear session path\n");
@ -580,7 +576,7 @@ int main(int argc, char ** argv) {
for (int i = 0; i < input_size; i += params.n_batch) {
int n_eval = std::min(input_size - i, params.n_batch);
if (llama_eval(ctx_guidance, input_buf + i, n_eval, n_past_guidance, params.n_threads)) {
if (llama_decode(ctx_guidance, llama_batch_get_one(input_buf + i, n_eval, n_past_guidance, 0), params.n_threads)) {
LOG_TEE("%s : failed to eval\n", __func__);
return 1;
}
@ -597,7 +593,7 @@ int main(int argc, char ** argv) {
LOG("eval: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd));
if (llama_eval(ctx, &embd[i], n_eval, n_past, params.n_threads)) {
if (llama_decode(ctx, llama_batch_get_one(&embd[i], n_eval, n_past, 0), params.n_threads)) {
LOG_TEE("%s : failed to eval\n", __func__);
return 1;
}