llama : remove all_pos_0, all_pos_1, all_seq_id from llama_batch (#9745)

* refactor llama_batch_get_one

* adapt all examples

* fix simple.cpp

* fix llama_bench

* fix

* fix context shifting

* free batch before return

* use common_batch_add, reuse llama_batch in loop

* null terminated seq_id list

* fix save-load-state example

* fix perplexity

* correct token pos in llama_batch_allocr
This commit is contained in:
Xuan Son Nguyen 2024-10-18 23:18:01 +02:00 committed by GitHub
parent afd9909a64
commit cda0e4b648
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
22 changed files with 205 additions and 118 deletions

View file

@ -496,6 +496,8 @@ static bool compute_imatrix(llama_context * ctx, const common_params & params) {
// clear the KV cache
llama_kv_cache_clear(ctx);
llama_batch batch = llama_batch_init(n_batch, 0, 1);
for (int j = 0; j < num_batches; ++j) {
const int batch_start = start + j * n_batch;
const int batch_size = std::min(end - batch_start, n_batch);
@ -508,9 +510,14 @@ static bool compute_imatrix(llama_context * ctx, const common_params & params) {
tokens[batch_start] = llama_token_bos(llama_get_model(ctx));
}
// TODO: use batch.logits to save computations instead of relying on logits_all == true
if (llama_decode(ctx, llama_batch_get_one(tokens.data() + batch_start, batch_size, j * n_batch, 0))) {
common_batch_clear(batch);
for (int i = 0; i < batch_size; i++) {
common_batch_add(batch, tokens[batch_start + i], j*n_batch + i, {0}, true);
}
if (llama_decode(ctx, batch)) {
LOG_ERR("%s : failed to eval\n", __func__);
llama_batch_free(batch);
return false;
}
@ -523,6 +530,8 @@ static bool compute_imatrix(llama_context * ctx, const common_params & params) {
}
}
llama_batch_free(batch);
const auto t_end = std::chrono::high_resolution_clock::now();
if (i == 0) {