CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-28 08:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

context : fix n_outputs during reserve (#15858)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-09-08 10:26:36 +03:00
committed by GitHub
parent cf0e3ba150
commit 663027fd54
2 changed files with 4 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llama-context.cpp
View File

@@ -285,8 +285,8 @@ llama_context::llama_context(
const uint32_t n_seqs = cparams.kv_unified ? 1 : cparams.n_seq_max; const uint32_t n_seqs = cparams.kv_unified ? 1 : cparams.n_seq_max;
const uint32_t n_tokens = std::min(cparams.n_ctx, cparams.n_ubatch); const uint32_t n_tokens = std::min(cparams.n_ctx, cparams.n_ubatch);
// avoid reserving graphs with zero outputs // avoid reserving graphs with zero outputs - assume one output per sequence
n_outputs = 1; n_outputs = n_seqs;
LLAMA_LOG_DEBUG("%s: worst-case: n_tokens = %d, n_seqs = %d, n_outputs = %d\n", __func__, n_tokens, n_seqs, n_outputs); LLAMA_LOG_DEBUG("%s: worst-case: n_tokens = %d, n_seqs = %d, n_outputs = %d\n", __func__, n_tokens, n_seqs, n_outputs);

3
src/llama-graph.cpp
View File

@@ -1431,7 +1431,8 @@ ggml_tensor * llm_graph_context::build_attn(
// [TAG_NO_CACHE_PAD] // [TAG_NO_CACHE_PAD]
// TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams // TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams
assert(!ubatch.equal_seqs() || (k_cur->ne[3] == 1 && k_cur->ne[3] == ubatch.n_seqs_unq)); // but it might not be worth it: https://github.com/ggml-org/llama.cpp/pull/15636
//assert(!ubatch.equal_seqs() || (k_cur->ne[3] == 1 && k_cur->ne[3] == ubatch.n_seqs_unq));
ggml_tensor * q = q_cur; ggml_tensor * q = q_cur;
ggml_tensor * k = k_cur; ggml_tensor * k = k_cur;