EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

model : add grok-2 support (#15539)

Browse Source 
* add grok-2 support

* type fix

* type fix

* type fix

* "fix" vocab for invalid sequences

* fix expert tensor mapping and spaces in vocab

* add chat template

* fix norm tensor mapping

* rename layer_out_norm to ffn_post_norm

* ensure ffn_post_norm is mapped

* fix experts merging

* remove erroneous FFN_GATE entry

* concatenate split tensors and add more metadata

* process all expert layers and try cat instead of hstack

* add support for community BPE vocab

* fix expert feed forward length and ffn_down concat

* commit this too

* add ffn_up/gate/down, unsure if sequence is right

* add ffn_gate/down/up to tensor names

* correct residual moe (still not working)

* mess--

* fix embedding scale being applied twice

* add built in chat template

* change beta fast for grok if default value

* remove spm vocab in favor of community bpe vocab

* change attention temp length metadata type to integer

* update attention temp length metadata

* remove comment

* replace M_SQRT2 with std::sqrt(2)

* add yarn metadata, move defaults to hparams
This commit is contained in:
Sigbjørn Skjæret
2025-09-14 23:00:59 +02:00
committed by GitHub
parent 6c019cb04e
commit b8e09f08b9
16 changed files with 281 additions and 96 deletions
Download Patch File Download Diff File Expand all files Collapse all files

100
src/llama-model.cpp
View File

@@ -685,7 +685,30 @@ void llama_model::load_hparams(llama_model_loader & ml) {
} break;
case LLM_ARCH_GROK:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
// defaults for old GGUFs
hparams.yarn_beta_fast = 8.0f;
hparams.f_logit_scale = 0.5773502691896257f;
hparams.f_embedding_scale = 78.38367176906169f;
hparams.f_attn_out_scale = 0.08838834764831845f;
hparams.f_attn_logit_softcapping = 30.0f;
hparams.f_router_logit_softcapping = 30.0f;
// no final_logit_softcapping in grok-1
hparams.f_final_logit_softcapping = 0.0f;
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp, false);
ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale, false);
ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale, false);
ml.get_key(LLM_KV_ATTENTION_OUTPUT_SCALE, hparams.f_attn_out_scale, false);
ml.get_key(LLM_KV_ATTN_LOGIT_SOFTCAPPING, hparams.f_attn_logit_softcapping, false);
ml.get_key(LLM_KV_ROUTER_LOGIT_SOFTCAPPING, hparams.f_router_logit_softcapping, false);
ml.get_key(LLM_KV_FINAL_LOGIT_SOFTCAPPING, hparams.f_final_logit_softcapping, false);
ml.get_key(LLM_KV_ATTENTION_TEMPERATURE_LENGTH, hparams.attn_temp_length, false);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR, hparams.yarn_ext_factor, false);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR, hparams.yarn_attn_factor, false);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_BETA_FAST, hparams.yarn_beta_fast, false);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_BETA_SLOW, hparams.yarn_beta_slow, false);
switch (hparams.n_layer) {
case 64: type = LLM_TYPE_314B; break;
@@ -2540,6 +2563,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
}
const int64_t n_ff_exp = hparams.n_ff_exp ? hparams.n_ff_exp : n_ff/* / n_expert_used*/; // grok-1 n_ff_exp == n_ff
for (int i = 0; i < n_layer; ++i) {
auto & layer = layers[i];
@@ -2554,12 +2578,19 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, TENSOR_NOT_REQUIRED);
layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0);
layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0);
layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, TENSOR_NOT_REQUIRED);
layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, TENSOR_NOT_REQUIRED);
layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, TENSOR_NOT_REQUIRED);
layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);
layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff_exp, n_expert}, TENSOR_NOT_REQUIRED);
layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp, n_embd, n_expert}, 0);
layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff_exp, n_expert}, 0);
layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, TENSOR_NOT_REQUIRED);
if (!layer.ffn_post_norm) {
layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, "weight", i), {n_embd}, 0);
}
}
} break;
case LLM_ARCH_DBRX:
@@ -7028,9 +7059,6 @@ struct llm_build_grok : public llm_graph_context {
inpL = build_inp_embd(model.tok_embd);
// multiply by embedding_multiplier_scale of 78.38367176906169
inpL = ggml_scale(ctx0, inpL, 78.38367176906169f);
// inp_pos - contains the positions
ggml_tensor * inp_pos = build_inp_pos();
@@ -7102,26 +7130,22 @@ struct llm_build_grok : public llm_graph_context {
inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
}
// Grok
// if attn_out_norm is present then apply it before adding the input
if (model.layers[il].attn_out_norm) {
cur = build_norm(cur,
model.layers[il].attn_out_norm, NULL,
LLM_NORM_RMS, il);
cb(cur, "attn_out_norm", il);
}
cur = build_norm(cur,
model.layers[il].attn_out_norm, NULL,
LLM_NORM_RMS, il);
cb(cur, "attn_out_norm", il);
ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
cb(ffn_inp, "ffn_inp", il);
// feed-forward network
// MoE branch
cur = build_norm(ffn_inp,
model.layers[il].ffn_norm, NULL,
LLM_NORM_RMS, il);
cb(cur, "ffn_norm", il);
cur = build_moe_ffn(cur,
// MoE branch
ggml_tensor * moe_out = build_moe_ffn(cur,
model.layers[il].ffn_gate_inp,
model.layers[il].ffn_up_exps,
model.layers[il].ffn_gate_exps,
@@ -7132,18 +7156,28 @@ struct llm_build_grok : public llm_graph_context {
false, 0.0,
LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
il);
cb(cur, "ffn_moe_out", il);
cb(moe_out, "ffn_moe_out", il);
// Grok
// if layer_out_norm is present then apply it before adding the input
// Idea: maybe ffn_out_norm is a better name
if (model.layers[il].layer_out_norm) {
cur = build_norm(cur,
model.layers[il].layer_out_norm, NULL,
LLM_NORM_RMS, il);
cb(cur, "layer_out_norm", il);
if (model.layers[il].ffn_up) {
ggml_tensor * ffn_out = build_ffn(cur,
model.layers[il].ffn_up, NULL, NULL,
model.layers[il].ffn_gate, NULL, NULL,
model.layers[il].ffn_down, NULL, NULL,
NULL,
LLM_FFN_GELU, LLM_FFN_PAR, il);
cb(ffn_out, "ffn_out", il);
cur = ggml_scale(ctx0, ggml_add(ctx0, ffn_out, moe_out), std::sqrt(2) / 2);
cb(cur, "ffn_out", il);
} else {
cur = moe_out;
}
cur = build_norm(cur,
model.layers[il].ffn_post_norm, NULL,
LLM_NORM_RMS, il);
cb(cur, "ffn_post_norm", il);
cur = ggml_add(ctx0, cur, ffn_inp);
cb(cur, "ffn_out", il);
@@ -7166,10 +7200,14 @@ struct llm_build_grok : public llm_graph_context {
// lm_head
cur = build_lora_mm(model.output, cur);
// Grok
// multiply logits by output_multiplier_scale of 0.5773502691896257
cur = ggml_scale(ctx0, cur, hparams.f_logit_scale);
cur = ggml_scale(ctx0, cur, 0.5773502691896257f);
// final logit soft-capping
if (hparams.f_final_logit_softcapping) {
cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
cur = ggml_tanh(ctx0, cur);
cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
}
cb(cur, "result_output", -1);
res->t_logits = cur;