ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

context : round n_tokens to next multiple of n_seqs when reserving (#14140)

Browse source 
This fixes RWKV inference which otherwise failed
when the worst case ubatch.n_seq_tokens rounded to 0.
This commit is contained in:
compilade 2025-06-12 02:56:04 -04:00 • committed by GitHub
parent 2e89f76b7a
commit a20b2b05bc
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
src/llama-context.cpp
View file

@ -1332,7 +1332,7 @@ ggml_cgraph * llama_context::graph_reserve(uint32_t n_tokens, uint32_t n_seqs, u
LLAMA_LOG_DEBUG("%s: reserving a graph for ubatch with n_tokens = %4u, n_seqs = %2u, n_outputs = %4u\n", __func__, n_tokens, n_seqs, n_outputs);
if (n_tokens % n_seqs != 0) {
n_tokens = (n_tokens / n_seqs) * n_seqs;
n_tokens = ((n_tokens + (n_seqs - 1)) / n_seqs) * n_seqs; // round to next multiple of n_seqs
n_outputs = std::min(n_outputs, n_tokens);
LLAMA_LOG_DEBUG("%s: making n_tokens a multiple of n_seqs - n_tokens = %u, n_seqs = %u, n_outputs = %u\n", __func__, n_tokens, n_seqs, n_outputs);