nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 16 Packages Projects Releases Wiki Activity

mamba2: remove hardcoded 2x expansion factor and invalid d_inner % d_state check (#23082)

Browse Source 
* mamba2: remove hardcoded 2x expansion factor, support any expand value

* mamba2: remove invalid d_inner %% d_state check (unrelated parameters)

* Update convert_hf_to_gguf.py: make expand optional with default 2

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* mamba2: apply expand fix to refactored conversion/mamba.py

* also check for mamba_expand

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Sigbjørn Skjæret <1629204+CISC@users.noreply.github.com>
This commit is contained in:
Arsen Arutunan
2026-06-26 08:50:54 +03:00
committed by GitHub
co-authored by Sigbjørn Skjæret Sigbjørn Skjæret
parent 5c7c22c3e1
commit 960d628f46
3 changed files with 10 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/mamba-base.cpp
-1
View File
@@ -169,7 +169,6 @@ ggml_tensor * llm_build_mamba_base::build_mamba2_layer(llm_graph_input_rs * inp,
GGML_ASSERT(ubatch.equal_seqs());
GGML_ASSERT(ubatch.n_tokens == n_seq_tokens * n_seqs);
GGML_ASSERT(d_inner % n_head == 0);
GGML_ASSERT(d_inner % d_state == 0);
GGML_ASSERT(d_inner % n_group == 0);
ggml_tensor * conv_states_all = mctx_cur->get_r_l(il);
src/models/mamba2.cpp
+7 -6
View File
@@ -39,10 +39,11 @@ void llama_model_mamba2::load_arch_tensors(llama_model_loader &) {
const int64_t d_inner = hparams.ssm_d_inner;
const int64_t d_state = hparams.ssm_d_state;
const int64_t n_group = hparams.ssm_n_group;
const int64_t d_in_proj = 2*d_inner + 2*n_group*d_state + n_head;
const int64_t dt_rank = hparams.ssm_dt_rank;
const int64_t conv_dim = d_inner + 2 * n_group * d_state;
const int64_t d_in_proj = d_inner + conv_dim + dt_rank;
// only an expansion factor of 2 is supported for now
GGML_ASSERT(2 * n_embd == d_inner);
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
@@ -68,11 +69,11 @@ void llama_model_mamba2::load_arch_tensors(llama_model_loader &) {
layer.ssm_conv1d = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "weight", i), {d_conv, d_inner + 2*n_group*d_state}, 0);
layer.ssm_conv1d_b = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "bias", i), {d_inner + 2*n_group*d_state}, 0);
layer.ssm_dt_b = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), {n_head}, 0);
layer.ssm_dt_b = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), {dt_rank}, 0);
// no "weight" suffix for these
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), {1, n_head}, 0);
layer.ssm_d = create_tensor(tn(LLM_TENSOR_SSM_D, i), {1, n_head}, 0);
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), {1, dt_rank}, 0);
layer.ssm_d = create_tensor(tn(LLM_TENSOR_SSM_D, i), {1, dt_rank}, 0);
layer.ssm_norm = create_tensor(tn(LLM_TENSOR_SSM_NORM, "weight", i), {d_inner / n_group, n_group}, 0);