8000 batch : auto-gen positions + verify multi-sequence input by ggerganov · Pull Request #14177 · ggml-org/llama.cpp · GitHub
[go: up one dir, main page]
Skip to content

batch : auto-gen positions + verify multi-sequence input #14177

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 6 commits into from
Jun 15, 2025
Merged

batch : auto-gen positions + verify multi-sequence input #14177

Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
42b2ae3
batch : verify multi-sequence input batches
ggerganov Jun 13, 2025
b67fe0d
cont : auto-gen positions + verify multi-seq input
ggerganov Jun 14, 2025
2437143
cont : first print debug info, then perform validation
ggerganov Jun 14, 2025
91b7792
cont : fix position auto-gen + add comments
ggerganov Jun 14, 2025
a6b0e85
cont : engrish [no ci]
ggerganov Jun 14, 2025
32b5eee
cont : comments [no ci]
ggerganov Jun 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
cont : first print debug info, then perform validation 
ggml-ci
  • Loading branch information
@ggerganov
ggerganov committed Jun 14, 2025
commit 2437143e0a839154d65f5b42e7988f7d580dc291
60 changes: 30 additions & 30 deletions src/llama-batch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -402,36 +402,6 @@ bool llama_batch_allocr::init(
}
}

for (int32_t s = 0; s < LLAMA_MAX_PARALLEL_SEQUENCES; ++s) {
if (seq_pos[s].empty()) {
continue;
}

if (memory && seq_pos_min(s) != memory->seq_pos_max(s) + 1) {
LLAMA_LOG_ERROR("%s: sequence %d does not start from the last position stored in the memory\n", __func__, s);
return false;
}

if (seq_pos_max(s) - seq_pos_min(s) + 1 > (int) seq_pos[s].size()) {
LLAMA_LOG_ERROR("%s: sequence %d is not contiguous\n", __func__, s);
return false;
}
}

if (memory) {
for (int32_t s0 = 0; s0 < LLAMA_MAX_PARALLEL_SEQUENCES; ++s0) {
for (int32_t s1 = 0; s1 < LLAMA_MAX_PARALLEL_SEQUENCES; ++s1) {
if (seq_cpl[s0][s1]) {
if (memory->seq_pos_min(s0) != memory->seq_pos_min(s1) ||
memory->seq_pos_max(s0) != memory->seq_pos_max(s1)) {
LLAMA_LOG_ERROR("%s: sequence %d is coupled to %d in the input batch, but have divereged\n", __func__, s0, s1);
return false;
}
}
}
}
}

if (debug > 0) {
LLAMA_LOG_DEBUG("%s: input batch info:\n", __func__);
LLAMA_LOG_DEBUG("%s: n_tokens = %d\n", __func__, batch.n_tokens);
Expand Down Expand Up @@ -497,6 +467,36 @@ bool llama_batch_allocr::init(
}
}

for (int32_t s = 0; s < LLAMA_MAX_PARALLEL_SEQUENCES; ++s) {
if (seq_pos[s].empty()) {
continue;
}

if (memory && seq_pos_min(s) != memory->seq_pos_max(s) + 1) {
LLAMA_LOG_ERROR("%s: sequence %d does not start from the last position stored in the memory\n", __func__, s);
return false;
}

if (seq_pos_max(s) - seq_pos_min(s) + 1 > (int) seq_pos[s].size()) {
LLAMA_LOG_ERROR("%s: sequence %d is not contiguous\n", __func__, s);
return false;
}
}

if (memory) {
for (int32_t s0 = 0; s0 < LLAMA_MAX_PARALLEL_SEQUENCES; ++s0) {
for (int32_t s1 = 0; s1 < LLAMA_MAX_PARALLEL_SEQUENCES; ++s1) {
if (seq_cpl[s0][s1]) {
if (memory->seq_pos_min(s0) != memory->seq_pos_min(s1) ||
memory->seq_pos_max(s0) != memory->seq_pos_max(s1)) {
LLAMA_LOG_ERROR("%s: sequence %d is coupled to %d in the input batch, but have divereged\n", __func__, s0, s1);
return false;
}
}
}
}
}

return true;
}

Expand Down
Loading
0