Skip to content

llama : fix attention layer count sanity check #6550

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 2 commits into from
Apr 8, 2024
Merged

llama : fix attention layer count sanity check #6550

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6804714
llama : fix attention layer count sanity check
ggerganov Apr 8, 2024
7bab4c0
llama : fix parentheses in attention layer count sanity check
compilade Apr 8, 2024
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
9 changes: 7 additions & 2 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13468,7 +13468,8 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
const std::string name = ggml_get_name(meta);

// TODO: avoid hardcoded tensor names - use the TN_* constants
if (name.find("attn_v.weight") != std::string::npos || name.find("attn_qkv.weight") != std::string::npos) {
if (name.find("attn_v.weight") != std::string::npos ||
name.find("attn_qkv.weight") != std::string::npos) {
++qs.n_attention_wv;
} else if (name == LLM_TN(model.arch)(LLM_TENSOR_OUTPUT, "weight")) {
qs.has_output = true;
Expand All @@ -13478,7 +13479,11 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
qs.n_ffn_down = qs.n_ffn_gate = qs.n_ffn_up = (int)model.hparams.n_layer;

// sanity checks
GGML_ASSERT(qs.n_attention_wv == (int)model.hparams.n_layer && "n_attention_wv != n_layer is unexpected");
//
// - qs.n_attention_wv == 0 for Mamba models
// - qs.n_attention_wv == model.hparams.n_layer for Transformer models
//
GGML_ASSERT((qs.n_attention_wv == 0 || qs.n_attention_wv == (int)model.hparams.n_layer) && "n_attention_wv is unexpected");

size_t total_size_org = 0;
size_t total_size_new = 0;
Expand Down