mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 34 Packages Projects Releases Wiki Activity

model: support MiMo-V2-Flash (#18328)

Browse Source 
* mimov2: convert ok

* rename mimov2 --> mimo2

* fix conversion

* runnable not incorrect

* use sink

* add_sliding_window_pattern

* add swa and per-layer n_head_kv

* correct params

* somewhat working

* correct gating func

* nits

* mimo2: wire RMS eps + MoE bias + converter guards

* add co-author

Co-authored-by: Aaryan-Kapoor <Aaryan-Kapoor@users.noreply.github.com>

* use add_rope_freq_base_swa

---------

Co-authored-by: Aaryan Kapoor <aaryankapoor2006@gmail.com>
Co-authored-by: Aaryan-Kapoor <Aaryan-Kapoor@users.noreply.github.com>
This commit is contained in:
Xuan-Son Nguyen
2025-12-24 23:07:08 +01:00
committed by GitHub
parent c184284230
commit 4cbafad4f0
11 changed files with 324 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-hparams.h
+4 -3
View File
@@ -123,10 +123,11 @@ struct llama_hparams {
llama_swa_type swa_type = LLAMA_SWA_TYPE_NONE;
// the size of the sliding window (0 - no SWA)
uint32_t n_swa = 0;
// if swa_layers[il] == true, then layer il is SWA
// if swa_layers[il] == false, then layer il is dense (i.e. non-SWA)
// if swa_layers[il] == 1, then layer il is SWA
// if swa_layers[il] == 0, then layer il is dense (i.e. non-SWA)
// by default, all layers are dense
std::array<bool, LLAMA_MAX_LAYERS> swa_layers;
// note: using uint32_t type for compatibility reason
std::array<uint32_t, LLAMA_MAX_LAYERS> swa_layers;
// for State Space Models
uint32_t ssm_d_conv = 0;