tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 03:55:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

batch : remove logits_all flag (#14141)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-12 11:49:26 +03:00
committed by GitHub
parent e2c0b6e46a
commit c3ee46fab4
10 changed files with 17 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/llama-batch.cpp
View File

@ -105,12 +105,7 @@ void llama_sbatch::add_seq_to_ubatch(llama_ubatch & ubatch, llama_sbatch_seq & s
ubatch.seq_id = batch->seq_id + seq.offset;
}
}
if (logits_all) {
for (size_t i = 0; i < length; ++i) {
ubatch.output[ubatch.n_tokens + i] = 1;
out_ids.push_back(ids[seq.offset + i]);
}
} else if (batch->logits) {
if (batch->logits) {
if (ubatch.equal_seqs) {
for (size_t i = 0; i < length; ++i) {
size_t id = ids[seq.offset + i];
@ -197,11 +192,10 @@ llama_ubatch llama_sbatch::split_seq(size_t n_ubatch) {
return ubatch;
}
llama_sbatch::llama_sbatch(const llama_batch & batch, size_t n_embd, bool simple_split, bool logits_all) {
llama_sbatch::llama_sbatch(const llama_batch & batch, size_t n_embd, bool simple_split) {
GGML_ASSERT(batch.n_tokens >= 0);
this->batch = &batch;
this->n_embd = n_embd;
this->logits_all = logits_all;
n_tokens = batch.n_tokens;
ids.resize(n_tokens);