tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-15 04:33:06 -04:00
Code Issues Packages Projects Releases Wiki Activity

llama-graph : fix recurrent state copy

Browse Source 
The `state_copy` shuffle assumes everything is moved at once,
which is not true when `states_extra` is copied back to the cache
before copying the range of states between `head` and `head + n_seqs`.
This is only a problem if any of the cells in [`head`, `head + n_seqs`)
have an `src` in [`head + n_seqs`, `head + n_kv`),
which does happen when `n_ubatch > 1` in the `llama-parallel` example.

Changing the order of the operations avoids the potential overwrite
before use, although when copies are avoided (like with Mamba2),
this will require further changes.

* llama-graph : rename n_state to state_size in build_recurrent_state

This naming should reduce confusion between the state size
and the number of states.
This commit is contained in:
Francis Couture-Harpin
2025-06-10 00:19:13 -04:00
parent dd6495ddc9
commit 62a9f34bae
2 changed files with 20 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
src/llama-graph.cpp
View File

@@ -1426,7 +1426,7 @@ ggml_tensor * llm_graph_context::build_recurrent_state(
ggml_cgraph * gf,
ggml_tensor * s,
ggml_tensor * state_copy,
int32_t n_state,
int32_t state_size,
int32_t n_seqs,
bool avoid_copies) const {
const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
@@ -1435,28 +1435,35 @@ ggml_tensor * llm_graph_context::build_recurrent_state(
const auto kv_head = kv_state->get_head();
const auto rs_zero = kv_state->get_rs_z();
ggml_tensor * states = ggml_reshape_2d(ctx0, s, n_state, kv_state->get_size());
ggml_tensor * states = ggml_reshape_2d(ctx0, s, state_size, kv_state->get_size());
// Clear a single state which will then be copied to the other cleared states.
// Note that this is a no-op when the view is zero-sized.
ggml_tensor * state_zero = ggml_view_1d(ctx0, states, n_state*(rs_zero >= 0), rs_zero*states->nb[1]*(rs_zero >= 0));
ggml_tensor * state_zero = ggml_view_1d(ctx0, states, state_size*(rs_zero >= 0), rs_zero*states->nb[1]*(rs_zero >= 0));
ggml_build_forward_expand(gf, ggml_scale_inplace(ctx0, state_zero, 0));
ggml_tensor * output_states;
if (!avoid_copies) {
// copy states
// NOTE: assuming the copy destinations are ALL contained between kv_head and kv_head + n_kv
// {state_size, kv_size} -> {state_size, n_seqs}
output_states = ggml_get_rows(ctx0, states, ggml_view_1d(ctx0, state_copy, n_seqs, 0));
ggml_build_forward_expand(gf, output_states);
} else {
// FIXME: make the gathering operation happen before the copy below
// (maybe with an optional lambda function passed as a parameter instead of `avoid_copies`?)
output_states = states;
}
// copy extra states which won't be changed further (between n_seqs and n_kv)
ggml_tensor * states_extra = ggml_get_rows(ctx0, states, ggml_view_1d(ctx0, state_copy, n_kv - n_seqs, n_seqs*state_copy->nb[0]));
ggml_build_forward_expand(gf,
ggml_cpy(ctx0,
states_extra,
ggml_view_1d(ctx0, s, n_state*(n_kv - n_seqs), (kv_head + n_seqs)*n_state*ggml_element_size(s))));
ggml_view_1d(ctx0, s, state_size*(n_kv - n_seqs), (kv_head + n_seqs)*state_size*ggml_element_size(s))));
if (!avoid_copies) {
// copy states
// NOTE: assuming the copy destinations are ALL contained between kv_head and kv_head + n_kv
// this shrinks the tensors's ne[1] to n_seqs
states = ggml_get_rows(ctx0, states, ggml_view_1d(ctx0, state_copy, n_seqs, 0));
}
return states;
return output_states;
}
ggml_tensor * llm_graph_context::build_rwkv_token_shift_load(

2
src/llama-graph.h
View File

@@ -597,7 +597,7 @@ struct llm_graph_context {
ggml_cgraph * gf,
ggml_tensor * s,
ggml_tensor * state_copy,
int32_t n_state,
int32_t state_size,
int32_t n_seqs,
bool avoid_copies = false) const;