Skip to content

graph : normalize Q, K, V shapes and add comments #12449

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 3 commits into from
Mar 18, 2025
+403 −247
Merged

graph : normalize Q, K, V shapes and add comments #12449

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
0b79556
graph : normalize Q, K, V shapes and add comments
ggerganov Mar 18, 2025
8d89dec
context : synchronize before getting cross attention data
ggerganov Mar 18, 2025
a745e6a
model : fix command-r attention norm check
ggerganov Mar 18, 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
2 changes: 1 addition & 1 deletion src/llama-graph.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1378,7 +1378,7 @@ ggml_tensor * llm_graph_context::build_attn(
// note: storing RoPE-ed version of K in the KV cache
ggml_build_forward_expand(gf, ggml_cpy(ctx0, k_cur, k_cache_view));

assert(v_cur->ne[0] == n_embd_v_gqa && v_cur->ne[1] == n_tokens);
v_cur = ggml_reshape_2d(ctx0, v_cur, n_embd_v_gqa, n_tokens);
Copy link
Member Author

@ggerganov ggerganov Mar 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This should fix the T5 bug: #12435


ggml_tensor * v_cache_view = nullptr;

Expand Down
24 changes: 12 additions & 12 deletions src/llama-graph.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -487,9 +487,9 @@ struct llm_graph_context {

ggml_tensor * build_attn_mha(
ggml_cgraph * gf,
ggml_tensor * q,
ggml_tensor * k,
ggml_tensor * v,
ggml_tensor * q, // [n_embd_head_q, n_tokens, n_head_q]
ggml_tensor * k, // [n_embd_head_k, n_tokens, n_head_k]
ggml_tensor * v, // [n_embd_head_v, n_tokens, n_head_v] (v_trans == false)
ggml_tensor * kq_b,
ggml_tensor * kq_mask,
bool v_trans,
Expand All @@ -502,9 +502,9 @@ struct llm_graph_context {
ggml_cgraph * gf,
ggml_tensor * wo,
ggml_tensor * wo_b,
ggml_tensor * q_cur,
ggml_tensor * k_cur,
ggml_tensor * v_cur,
ggml_tensor * q_cur, // [n_embd_head_q, n_head_q, n_tokens]
ggml_tensor * k_cur, // [n_embd_head_k, n_head_k, n_tokens]
ggml_tensor * v_cur, // [n_embd_head_v, n_head_v, n_tokens]
ggml_tensor * kq_b,
float kq_scale,
int il) const;
Expand All @@ -516,9 +516,9 @@ struct llm_graph_context {
ggml_cgraph * gf,
ggml_tensor * wo,
ggml_tensor * wo_b,
ggml_tensor * q_cur,
ggml_tensor * k_cur,
ggml_tensor * v_cur,
ggml_tensor * q_cur, // [n_embd_head_q, n_head_q, n_tokens]
ggml_tensor * k_cur, // [n_embd_head_k, n_head_k, n_tokens]
ggml_tensor * v_cur, // [n_embd_head_v, n_head_v, n_tokens]
ggml_tensor * kq_b,
float kq_scale,
int il) const;
Expand All @@ -530,9 +530,9 @@ struct llm_graph_context {
ggml_cgraph * gf,
ggml_tensor * wo,
ggml_tensor * wo_b,
ggml_tensor * q_cur,
ggml_tensor * k_cur,
ggml_tensor * v_cur,
ggml_tensor * q_cur, // [n_embd_head_q, n_head_q, n_tokens]
ggml_tensor * k_cur, // [n_embd_head_k, n_head_k, n_tokens]
ggml_tensor * v_cur, // [n_embd_head_v, n_head_v, n_tokens]
ggml_tensor * kq_b,
float kq_scale,
int il) const;
Expand Down
Loading
Loading