Skip to content

Migrate Qwen3.5 to IMROPE #19443

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

Closed
pwilkin wants to merge 1 commit into from
Closed

Migrate Qwen3.5 to IMROPE #19443

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 3 additions & 2 deletions src/llama-model.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2418,6 +2418,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp, false);
ml.get_key(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp, false);
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);
Expand Down Expand Up @@ -8755,8 +8756,6 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_PANGU_EMBED:
case LLM_ARCH_AFMOE:
case LLM_ARCH_QWEN3NEXT:
case LLM_ARCH_QWEN3_5:
case LLM_ARCH_QWEN3_5_MOE:
case LLM_ARCH_MIMO2:
case LLM_ARCH_STEP35:
return LLAMA_ROPE_TYPE_NEOX;
Expand All @@ -8765,6 +8764,8 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
return LLAMA_ROPE_TYPE_MROPE;
case LLM_ARCH_QWEN3VL:
case LLM_ARCH_QWEN3VLMOE:
case LLM_ARCH_QWEN3_5:
case LLM_ARCH_QWEN3_5_MOE:
return LLAMA_ROPE_TYPE_IMROPE;

case LLM_ARCH_GLM4:
Expand Down
13 changes: 8 additions & 5 deletions src/models/qwen3-5.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -132,15 +132,18 @@ ggml_tensor * llm_build_qwen3_5::build_layer_attn(

Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);

Qcur = ggml_rope_ext(
int sections[4];
std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);

Qcur = ggml_rope_multi(
ctx0, Qcur, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
ext_factor, attn_factor, beta_fast, beta_slow);

Kcur = ggml_rope_ext(
Kcur = ggml_rope_multi(
ctx0, Kcur, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base,
freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
ext_factor, attn_factor, beta_fast, beta_slow);

cb(Qcur, "Qcur", il);
cb(Kcur, "Kcur", il);
Expand Down
Loading