8000 llama : refactor llama_context, llama_kv_cache, llm_build_context (v2) by ggerganov · Pull Request #12181 · ggml-org/llama.cpp · GitHub
[go: up one dir, main page]
Skip to content

llama : refactor llama_context, llama_kv_cache, llm_build_context (v2) #12181

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 16 commits into from
Mar 13, 2025
Merged

llama : refactor llama_context, llama_kv_cache, llm_build_context (v2) #12181

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
5590925
llama : refactor llama_context, llama_kv_cache, llm_build_context
ggerganov Mar 4, 2025
75624a2
graph : don't mutate the KV cache during defrag
ggerganov Mar 4, 2025
5aa3518
context : reduce virtuals + remove test function
ggerganov Mar 5, 2025
0a6648c
context : move interface implementation to source file + factory
ggerganov Mar 5, 2025
cc9fa25
graph : move KV cache build functions to llama_context impl
ggerganov Mar 5, 2025
29c9ef5
graph : remove model reference from build_pooling
ggerganov Mar 5, 2025
bc82560
graph : remove llama_model reference
ggerganov Mar 5, 2025
ff95ffd
kv_cache : provide rope factors
ggerganov Mar 6, 2025
562a478
graph : rework inputs to use only unique_ptr, remove attn input abstr…
ggerganov Mar 6, 2025
d0cb319
context : remove llama_context_i abstraction
ggerganov Mar 6, 2025
a4fc4e8
context : clean-up
ggerganov Mar 7, 2025
af9f6b8
graph : clean-up
ggerganov Mar 7, 2025
226ff01
llama : remove redundant keywords (struct, enum)
ggerganov Mar 7, 2025
5fc6dbd
model : adapt gemma3
ggerganov Mar 12, 2025
70ef653
graph : restore same attention ops as on master
ggerganov Mar 12, 2025
31b8eab
llama : remove TODO + fix indent
ggerganov Mar 13, 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
Prev Previous commit
Next Next commit
graph : rework inputs to use only unique_ptr, remove attn input abstr… 
…action

ggml-ci
  • Loading branch information
@ggerganov
ggerganov committed Mar 12, 2025
commit 562a47870b60cd2a403991a60498818f7dcbb10c
6 changes: 3 additions & 3 deletions src/llama-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1874,13 +1874,11 @@ llm_graph_result_ptr llama_context_kv_self::build_kv_self_shift(

//GGML_ASSERT(kv_self->size == n_ctx);

auto inp = std::make_shared<llm_graph_input_k_shift>(kv_self.get());
auto inp = std::make_unique<llm_graph_input_k_shift>(kv_self.get());

inp->k_shift = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, cparams.n_ctx);
ggml_set_input(inp->k_shift);

res->add_input(inp);

for (uint32_t il = 0; il < n_layer; ++il) {
const int64_t n_head_kv = hparams.n_head_kv(il);
const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
Expand All @@ -1899,6 +1897,8 @@ llm_graph_result_ptr llama_context_kv_self::build_kv_self_shift(
ggml_build_forward_expand(gf, cur);
}

res->add_input(std::move(inp));

return res;
}

Expand Down
Loading
0