ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-02 18:10:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

iAdding support for dense Qwen-3.5 models (#1326)

Browse Source
This commit is contained in:
Kawrakow
2026-02-26 08:51:01 +01:00
committed by GitHub
parent 2616efa296
commit 0aa6f7e7cd
9 changed files with 263 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/llama-hparams.cpp
View File

@@ -507,6 +507,33 @@ void llm_load_hparams(
default: model.type = e_model::MODEL_UNKNOWN;
}
} break;
case LLM_ARCH_QWEN35:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key_or_arr(LLM_KV_ROPE_DIMENSION_SECTIONS, hparams.rope_sections, 4, true);
// Load linear attention (gated delta net) parameters
ml.get_key(LLM_KV_SSM_CONV_KERNEL, hparams.ssm_d_conv);
ml.get_key(LLM_KV_SSM_INNER_SIZE, hparams.ssm_d_inner);
ml.get_key(LLM_KV_SSM_STATE_SIZE, hparams.ssm_d_state);
ml.get_key(LLM_KV_SSM_TIME_STEP_RANK, hparams.ssm_dt_rank);
ml.get_key(LLM_KV_SSM_GROUP_COUNT, hparams.ssm_n_group);
// Mark recurrent layers (linear attention layers)
{
uint32_t full_attn_interval = 4;
ml.get_key(LLM_KV_FULL_ATTENTION_INTERVAL, full_attn_interval, false);
for (uint32_t i = 0; i < hparams.n_layer; ++i) {
hparams.recurrent_layer_arr[i] = ((i + 1) % full_attn_interval != 0);
}
}
switch (hparams.n_layer) {
case 24: model.type = e_model::MODEL_2B; break;
case 64: model.type = e_model::MODEL_27B; break;
default: model.type = e_model::MODEL_UNKNOWN;
}
} break;
case LLM_ARCH_QWEN3VLMOE:
{
ml.get_key(LLM_KV_NUM_DEEPSTACK_LAYERS, hparams.n_deepstack_layers, false);