|
@@ -1506,6 +1506,11 @@ void llama_model::load_hparams(llama_model_loader & ml) {
|
|
|
ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale);
|
|
ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale);
|
|
|
ml.get_key(LLM_KV_ATTENTION_SCALE, hparams.f_attention_scale);
|
|
ml.get_key(LLM_KV_ATTENTION_SCALE, hparams.f_attention_scale);
|
|
|
|
|
|
|
|
|
|
+ // Granite uses rope_finetuned as a switch for rope, so default to true
|
|
|
|
|
+ bool rope_finetuned = true;
|
|
|
|
|
+ ml.get_key(LLM_KV_ROPE_SCALING_FINETUNED, rope_finetuned, false);
|
|
|
|
|
+ hparams.rope_finetuned = rope_finetuned;
|
|
|
|
|
+
|
|
|
switch (hparams.n_layer) {
|
|
switch (hparams.n_layer) {
|
|
|
case 32: type = LLM_TYPE_3B; break;
|
|
case 32: type = LLM_TYPE_3B; break;
|
|
|
case 40: type = LLM_TYPE_3B; break;
|
|
case 40: type = LLM_TYPE_3B; break;
|
|
@@ -1513,6 +1518,40 @@ void llama_model::load_hparams(llama_model_loader & ml) {
|
|
|
default: type = LLM_TYPE_UNKNOWN;
|
|
default: type = LLM_TYPE_UNKNOWN;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
|
|
+ // For Granite MoE Shared
|
|
|
|
|
+ ml.get_key(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp, /* required */ false);
|
|
|
|
|
+ } break;
|
|
|
|
|
+ case LLM_ARCH_GRANITE_HYBRID:
|
|
|
|
|
+ {
|
|
|
|
|
+ ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
|
|
|
|
|
+ ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale, /* required */ false);
|
|
|
|
|
+ ml.get_key(LLM_KV_RESIDUAL_SCALE, hparams.f_residual_scale, /* required */ false);
|
|
|
|
|
+ ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale, /* required */ false);
|
|
|
|
|
+ ml.get_key(LLM_KV_ATTENTION_SCALE, hparams.f_attention_scale, /* required */ false);
|
|
|
|
|
+
|
|
|
|
|
+ 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);
|
|
|
|
|
+
|
|
|
|
|
+ // Granite uses rope_finetuned as a switch for rope, so default to true
|
|
|
|
|
+ bool rope_finetuned = true;
|
|
|
|
|
+ ml.get_key(LLM_KV_ROPE_SCALING_FINETUNED, rope_finetuned, false);
|
|
|
|
|
+ hparams.rope_finetuned = rope_finetuned;
|
|
|
|
|
+
|
|
|
|
|
+ // A layer is recurrent IFF the n_head_kv value is set to 0
|
|
|
|
|
+ for (uint32_t i = 0; i < hparams.n_layer; ++i) {
|
|
|
|
|
+ hparams.recurrent_layer_arr[i] = hparams.n_head_kv(i) == 0;
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
|
|
|
|
|
+
|
|
|
|
|
+ switch (hparams.n_layer) {
|
|
|
|
|
+ // TODO: Add llm type label (not sure this is useful)
|
|
|
|
|
+ default: type = LLM_TYPE_UNKNOWN;
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
// For Granite MoE Shared
|
|
// For Granite MoE Shared
|
|
|
ml.get_key(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp, /* required */ false);
|
|
ml.get_key(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp, /* required */ false);
|
|
|
} break;
|
|
} break;
|
|
@@ -3364,6 +3403,99 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
|
|
|
}
|
|
}
|
|
|
}
|
|
}
|
|
|
} break;
|
|
} break;
|
|
|
|
|
+ case LLM_ARCH_GRANITE_HYBRID:
|
|
|
|
|
+ {
|
|
|
|
|
+ // mamba2 Mixer SSM params
|
|
|
|
|
+ // NOTE: int64_t for tensor dimensions
|
|
|
|
|
+ const int64_t d_conv = hparams.ssm_d_conv;
|
|
|
|
|
+ const int64_t d_inner = hparams.ssm_d_inner;
|
|
|
|
|
+ const int64_t d_state = hparams.ssm_d_state;
|
|
|
|
|
+ const int64_t n_ssm_head = hparams.ssm_dt_rank;
|
|
|
|
|
+ const int64_t n_group = hparams.ssm_n_group;
|
|
|
|
|
+ const int64_t d_in_proj = 2*d_inner + 2*n_group*d_state + n_ssm_head;
|
|
|
|
|
+
|
|
|
|
|
+ // only an expansion factor of 2 is supported for now
|
|
|
|
|
+ GGML_ASSERT(2 * n_embd == d_inner);
|
|
|
|
|
+
|
|
|
|
|
+ // embeddings
|
|
|
|
|
+ tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
|
|
|
|
|
+
|
|
|
|
|
+ // output
|
|
|
|
|
+ {
|
|
|
|
|
+ output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
|
|
|
|
|
+ output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ // if output is NULL, init from the input tok embed, duplicated to allow offloading
|
|
|
|
|
+ if (output == NULL) {
|
|
|
|
|
+ output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
|
|
|
|
|
+ }
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ for (int i = 0; i < n_layer; ++i) {
|
|
|
|
|
+ auto & layer = layers[i];
|
|
|
|
|
+
|
|
|
|
|
+ // norm
|
|
|
|
|
+ layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
|
|
|
|
|
+
|
|
|
|
|
+ if (hparams.is_recurrent(i)) {
|
|
|
|
|
+ // ssm layers
|
|
|
|
|
+ layer.ssm_in = create_tensor(tn(LLM_TENSOR_SSM_IN, "weight", i), {n_embd, d_in_proj}, 0);
|
|
|
|
|
+
|
|
|
|
|
+ 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}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+
|
|
|
|
|
+ layer.ssm_dt_b = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), {n_ssm_head}, 0);
|
|
|
|
|
+
|
|
|
|
|
+ // no "weight" suffix for these
|
|
|
|
|
+ layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), {1, n_ssm_head}, 0);
|
|
|
|
|
+ layer.ssm_d = create_tensor(tn(LLM_TENSOR_SSM_D, i), {1, n_ssm_head}, 0);
|
|
|
|
|
+
|
|
|
|
|
+ layer.ssm_norm = create_tensor(tn(LLM_TENSOR_SSM_NORM, "weight", i), {d_inner / n_group, n_group}, 0);
|
|
|
|
|
+
|
|
|
|
|
+ // out_proj
|
|
|
|
|
+ layer.ssm_out = create_tensor(tn(LLM_TENSOR_SSM_OUT, "weight", i), {d_inner, n_embd}, 0);
|
|
|
|
|
+ } else {
|
|
|
|
|
+ // attention layers (with optional bias)
|
|
|
|
|
+ const int64_t n_head_i = hparams.n_head(i);
|
|
|
|
|
+ const int64_t n_embd_k_gqa_i = hparams.n_embd_k_gqa(i);
|
|
|
|
|
+ const int64_t n_embd_v_gqa_i = hparams.n_embd_v_gqa(i);
|
|
|
|
|
+ layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head_i}, 0);
|
|
|
|
|
+ layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa_i}, 0);
|
|
|
|
|
+ layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa_i}, 0);
|
|
|
|
|
+ layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head_i, n_embd}, 0);
|
|
|
|
|
+ layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_k_gqa_i}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_v_gqa_i}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ // feed forward (w/ optional biases)
|
|
|
|
|
+ if (n_expert > 0) {
|
|
|
|
|
+ // MoE FFN
|
|
|
|
|
+ layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
|
|
|
|
|
+ layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 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);
|
|
|
|
|
+
|
|
|
|
|
+ // For Granite MoE Shared
|
|
|
|
|
+ if (hparams.n_ff_shexp > 0) {
|
|
|
|
|
+ layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, hparams.n_ff_shexp}, 0);
|
|
|
|
|
+ layer.ffn_up_shexp = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP, "weight", i), {n_embd, hparams.n_ff_shexp}, 0);
|
|
|
|
|
+ layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {hparams.n_ff_shexp, n_embd}, 0);
|
|
|
|
|
+ }
|
|
|
|
|
+ } else {
|
|
|
|
|
+ layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
|
|
|
|
|
+ layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
|
|
|
|
|
+ layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
|
|
|
|
|
+ layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0);
|
|
|
|
|
+ layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);
|
|
|
|
|
+ layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);
|
|
|
|
|
+ }
|
|
|
|
|
+ }
|
|
|
|
|
+ } break;
|
|
|
case LLM_ARCH_XVERSE:
|
|
case LLM_ARCH_XVERSE:
|
|
|
{
|
|
{
|
|
|
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
|
|
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
|
|
@@ -5026,7 +5158,8 @@ void llama_model::print_info() const {
|
|
|
if (arch == LLM_ARCH_MAMBA ||
|
|
if (arch == LLM_ARCH_MAMBA ||
|
|
|
arch == LLM_ARCH_MAMBA2 ||
|
|
arch == LLM_ARCH_MAMBA2 ||
|
|
|
arch == LLM_ARCH_JAMBA ||
|
|
arch == LLM_ARCH_JAMBA ||
|
|
|
- arch == LLM_ARCH_FALCON_H1) {
|
|
|
|
|
|
|
+ arch == LLM_ARCH_FALCON_H1 ||
|
|
|
|
|
+ arch == LLM_ARCH_GRANITE_HYBRID) {
|
|
|
LLAMA_LOG_INFO("%s: ssm_d_conv = %u\n", __func__, hparams.ssm_d_conv);
|
|
LLAMA_LOG_INFO("%s: ssm_d_conv = %u\n", __func__, hparams.ssm_d_conv);
|
|
|
LLAMA_LOG_INFO("%s: ssm_d_inner = %u\n", __func__, hparams.ssm_d_inner);
|
|
LLAMA_LOG_INFO("%s: ssm_d_inner = %u\n", __func__, hparams.ssm_d_inner);
|
|
|
LLAMA_LOG_INFO("%s: ssm_d_state = %u\n", __func__, hparams.ssm_d_state);
|
|
LLAMA_LOG_INFO("%s: ssm_d_state = %u\n", __func__, hparams.ssm_d_state);
|
|
@@ -5081,7 +5214,8 @@ void llama_model::print_info() const {
|
|
|
|
|
|
|
|
if (arch == LLM_ARCH_MINICPM ||
|
|
if (arch == LLM_ARCH_MINICPM ||
|
|
|
arch == LLM_ARCH_GRANITE ||
|
|
arch == LLM_ARCH_GRANITE ||
|
|
|
- arch == LLM_ARCH_GRANITE_MOE) {
|
|
|
|
|
|
|
+ arch == LLM_ARCH_GRANITE_MOE ||
|
|
|
|
|
+ arch == LLM_ARCH_GRANITE_HYBRID) {
|
|
|
LLAMA_LOG_INFO("%s: f_embedding_scale = %f\n", __func__, hparams.f_embedding_scale);
|
|
LLAMA_LOG_INFO("%s: f_embedding_scale = %f\n", __func__, hparams.f_embedding_scale);
|
|
|
LLAMA_LOG_INFO("%s: f_residual_scale = %f\n", __func__, hparams.f_residual_scale);
|
|
LLAMA_LOG_INFO("%s: f_residual_scale = %f\n", __func__, hparams.f_residual_scale);
|
|
|
LLAMA_LOG_INFO("%s: f_attention_scale = %f\n", __func__, hparams.f_attention_scale);
|
|
LLAMA_LOG_INFO("%s: f_attention_scale = %f\n", __func__, hparams.f_attention_scale);
|
|
@@ -13797,13 +13931,11 @@ struct llm_build_arwkv7 : public llm_build_rwkv7_base {
|
|
|
}
|
|
}
|
|
|
};
|
|
};
|
|
|
|
|
|
|
|
-
|
|
|
|
|
struct llm_build_granite : public llm_graph_context {
|
|
struct llm_build_granite : public llm_graph_context {
|
|
|
llm_build_granite(
|
|
llm_build_granite(
|
|
|
const llama_model & model,
|
|
const llama_model & model,
|
|
|
const llm_graph_params & params,
|
|
const llm_graph_params & params,
|
|
|
- ggml_cgraph * gf,
|
|
|
|
|
- const bool use_rope = true)
|
|
|
|
|
|
|
+ ggml_cgraph * gf)
|
|
|
: llm_graph_context(params) {
|
|
: llm_graph_context(params) {
|
|
|
|
|
|
|
|
const int64_t n_embd_head = hparams.n_embd_head_v;
|
|
const int64_t n_embd_head = hparams.n_embd_head_v;
|
|
@@ -13818,14 +13950,12 @@ struct llm_build_granite : public llm_graph_context {
|
|
|
|
|
|
|
|
// inp_pos - built only if rope enabled
|
|
// inp_pos - built only if rope enabled
|
|
|
ggml_tensor * inp_pos = nullptr;
|
|
ggml_tensor * inp_pos = nullptr;
|
|
|
- if (use_rope) {
|
|
|
|
|
|
|
+ if (hparams.rope_finetuned) {
|
|
|
inp_pos = build_inp_pos();
|
|
inp_pos = build_inp_pos();
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
auto * inp_attn = build_attn_inp_kv_unified();
|
|
auto * inp_attn = build_attn_inp_kv_unified();
|
|
|
|
|
|
|
|
- const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
|
|
|
|
|
-
|
|
|
|
|
ggml_tensor * inp_out_ids = build_inp_out_ids();
|
|
ggml_tensor * inp_out_ids = build_inp_out_ids();
|
|
|
|
|
|
|
|
for (int il = 0; il < n_layer; ++il) {
|
|
for (int il = 0; il < n_layer; ++il) {
|
|
@@ -13838,128 +13968,237 @@ struct llm_build_granite : public llm_graph_context {
|
|
|
cb(cur, "attn_norm", il);
|
|
cb(cur, "attn_norm", il);
|
|
|
|
|
|
|
|
// self-attention
|
|
// self-attention
|
|
|
- {
|
|
|
|
|
- // compute Q and K and (optionally) RoPE them
|
|
|
|
|
- ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
|
|
|
|
|
- cb(Qcur, "Qcur", il);
|
|
|
|
|
- if (model.layers[il].bq) {
|
|
|
|
|
- Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
|
|
|
|
|
- cb(Qcur, "Qcur", il);
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ cur = build_attention_layer(
|
|
|
|
|
+ gf, cur, inp_pos, inp_attn,
|
|
|
|
|
+ model, n_embd_head, il);
|
|
|
|
|
|
|
|
- ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
|
|
|
|
|
- cb(Kcur, "Kcur", il);
|
|
|
|
|
- if (model.layers[il].bk) {
|
|
|
|
|
- Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
|
|
|
|
|
- cb(Kcur, "Kcur", il);
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ if (il == n_layer - 1 && inp_out_ids) {
|
|
|
|
|
+ cur = ggml_get_rows(ctx0, cur, inp_out_ids);
|
|
|
|
|
+ inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
|
|
|
|
|
+ }
|
|
|
|
|
|
|
|
- ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
|
|
|
|
|
- cb(Vcur, "Vcur", il);
|
|
|
|
|
- if (model.layers[il].bv) {
|
|
|
|
|
- Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
|
|
|
|
|
- cb(Vcur, "Vcur", il);
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ // ffn
|
|
|
|
|
+ cur = build_layer_ffn(cur, inpSA, model, il);
|
|
|
|
|
|
|
|
- Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
|
|
|
|
|
- Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
|
|
|
|
|
- Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
|
|
|
|
|
|
|
+ // input for next layer
|
|
|
|
|
+ inpL = cur;
|
|
|
|
|
+ }
|
|
|
|
|
|
|
|
- if (use_rope) {
|
|
|
|
|
- ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
|
|
|
|
|
- Qcur = ggml_rope_ext(
|
|
|
|
|
- ctx0, Qcur, inp_pos, rope_factors,
|
|
|
|
|
- n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
|
|
|
|
|
- ext_factor, attn_factor, beta_fast, beta_slow
|
|
|
|
|
- );
|
|
|
|
|
|
|
+ cur = inpL;
|
|
|
|
|
|
|
|
- Kcur = ggml_rope_ext(
|
|
|
|
|
- ctx0, Kcur, inp_pos, rope_factors,
|
|
|
|
|
- n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
|
|
|
|
|
- ext_factor, attn_factor, beta_fast, beta_slow
|
|
|
|
|
- );
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ cur = build_norm(cur,
|
|
|
|
|
+ model.output_norm, NULL,
|
|
|
|
|
+ LLM_NORM_RMS, -1);
|
|
|
|
|
|
|
|
- cb(Qcur, "Qcur", il);
|
|
|
|
|
- cb(Kcur, "Kcur", il);
|
|
|
|
|
- cb(Vcur, "Vcur", il);
|
|
|
|
|
|
|
+ cb(cur, "result_norm", -1);
|
|
|
|
|
+ res->t_embd = cur;
|
|
|
|
|
|
|
|
- cur = build_attn(inp_attn, gf,
|
|
|
|
|
- model.layers[il].wo, model.layers[il].bo,
|
|
|
|
|
- Qcur, Kcur, Vcur, nullptr, nullptr, kq_scale, il);
|
|
|
|
|
|
|
+ // lm_head
|
|
|
|
|
+ cur = build_lora_mm(model.output, cur);
|
|
|
|
|
+
|
|
|
|
|
+ // For Granite architectures - scale logits
|
|
|
|
|
+ cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_logit_scale);
|
|
|
|
|
+ cb(cur, "result_output", -1);
|
|
|
|
|
+ res->t_logits = cur;
|
|
|
|
|
+
|
|
|
|
|
+ ggml_build_forward_expand(gf, cur);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * build_attention_layer(
|
|
|
|
|
+ ggml_cgraph * gf,
|
|
|
|
|
+ ggml_tensor * cur,
|
|
|
|
|
+ ggml_tensor * inp_pos,
|
|
|
|
|
+ llm_graph_input_attn_kv_unified * inp_attn,
|
|
|
|
|
+ const llama_model & model,
|
|
|
|
|
+ const int64_t n_embd_head,
|
|
|
|
|
+ const int il) {
|
|
|
|
|
+
|
|
|
|
|
+ // compute Q and K and (optionally) RoPE them
|
|
|
|
|
+ ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
|
|
|
|
|
+ cb(Qcur, "Qcur", il);
|
|
|
|
|
+ if (model.layers[il].bq) {
|
|
|
|
|
+ Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
|
|
|
|
|
+ cb(Qcur, "Qcur", il);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
|
|
|
|
|
+ cb(Kcur, "Kcur", il);
|
|
|
|
|
+ if (model.layers[il].bk) {
|
|
|
|
|
+ Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
|
|
|
|
|
+ cb(Kcur, "Kcur", il);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
|
|
|
|
|
+ cb(Vcur, "Vcur", il);
|
|
|
|
|
+ if (model.layers[il].bv) {
|
|
|
|
|
+ Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
|
|
|
|
|
+ cb(Vcur, "Vcur", il);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, hparams.n_head(il), n_tokens);
|
|
|
|
|
+ Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, hparams.n_head_kv(il), n_tokens);
|
|
|
|
|
+ Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, hparams.n_head_kv(il), n_tokens);
|
|
|
|
|
+
|
|
|
|
|
+ const bool use_rope = hparams.rope_finetuned;
|
|
|
|
|
+ if (use_rope) {
|
|
|
|
|
+ ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
|
|
|
|
|
+ Qcur = ggml_rope_ext(
|
|
|
|
|
+ ctx0, Qcur, inp_pos, rope_factors,
|
|
|
|
|
+ n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
|
|
|
|
|
+ ext_factor, attn_factor, beta_fast, beta_slow
|
|
|
|
|
+ );
|
|
|
|
|
+
|
|
|
|
|
+ Kcur = ggml_rope_ext(
|
|
|
|
|
+ ctx0, Kcur, inp_pos, rope_factors,
|
|
|
|
|
+ n_rot, 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);
|
|
|
|
|
+ cb(Vcur, "Vcur", il);
|
|
|
|
|
+
|
|
|
|
|
+ const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
|
|
|
|
|
+ cur = build_attn(inp_attn, gf,
|
|
|
|
|
+ model.layers[il].wo, model.layers[il].bo,
|
|
|
|
|
+ Qcur, Kcur, Vcur, nullptr, nullptr, kq_scale, il);
|
|
|
cb(cur, "attn_out", il);
|
|
cb(cur, "attn_out", il);
|
|
|
- }
|
|
|
|
|
|
|
+ return cur;
|
|
|
|
|
+ }
|
|
|
|
|
|
|
|
- if (il == n_layer - 1 && inp_out_ids) {
|
|
|
|
|
- cur = ggml_get_rows(ctx0, cur, inp_out_ids);
|
|
|
|
|
- inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ ggml_tensor * build_layer_ffn(
|
|
|
|
|
+ ggml_tensor * cur,
|
|
|
|
|
+ ggml_tensor * inpSA,
|
|
|
|
|
+ const llama_model & model,
|
|
|
|
|
+ const int il) {
|
|
|
|
|
|
|
|
- // For Granite architectures - scale residual
|
|
|
|
|
|
|
+ // For Granite architectures - scale residual
|
|
|
|
|
+ if (hparams.f_residual_scale) {
|
|
|
cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
|
|
cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
|
|
|
- ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
|
|
|
|
|
- cb(ffn_inp, "ffn_inp", il);
|
|
|
|
|
|
|
+ }
|
|
|
|
|
+ ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
|
|
|
|
|
+ cb(ffn_inp, "ffn_inp", il);
|
|
|
|
|
|
|
|
- // feed-forward network (non-MoE)
|
|
|
|
|
- if (model.layers[il].ffn_gate_inp == nullptr) {
|
|
|
|
|
|
|
+ // feed-forward network (non-MoE)
|
|
|
|
|
+ if (model.layers[il].ffn_gate_inp == nullptr) {
|
|
|
|
|
|
|
|
- cur = build_norm(ffn_inp,
|
|
|
|
|
- model.layers[il].ffn_norm, NULL,
|
|
|
|
|
- LLM_NORM_RMS, il);
|
|
|
|
|
- cb(cur, "ffn_norm", il);
|
|
|
|
|
|
|
+ cur = build_norm(ffn_inp,
|
|
|
|
|
+ model.layers[il].ffn_norm, NULL,
|
|
|
|
|
+ LLM_NORM_RMS, il);
|
|
|
|
|
+ cb(cur, "ffn_norm", il);
|
|
|
|
|
|
|
|
- cur = build_ffn(cur,
|
|
|
|
|
- model.layers[il].ffn_up, model.layers[il].ffn_up_b, NULL,
|
|
|
|
|
- model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
|
|
|
|
|
- model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
|
|
|
|
|
- NULL,
|
|
|
|
|
- LLM_FFN_SILU, LLM_FFN_PAR, il);
|
|
|
|
|
- cb(cur, "ffn_out", il);
|
|
|
|
|
|
|
+ cur = build_ffn(cur,
|
|
|
|
|
+ model.layers[il].ffn_up, model.layers[il].ffn_up_b, NULL,
|
|
|
|
|
+ model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
|
|
|
|
|
+ model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
|
|
|
|
|
+ NULL,
|
|
|
|
|
+ LLM_FFN_SILU, LLM_FFN_PAR, il);
|
|
|
|
|
+ cb(cur, "ffn_out", il);
|
|
|
|
|
|
|
|
- } else {
|
|
|
|
|
- // MoE branch
|
|
|
|
|
- cur = build_norm(ffn_inp,
|
|
|
|
|
- model.layers[il].ffn_norm, NULL,
|
|
|
|
|
- LLM_NORM_RMS, il);
|
|
|
|
|
- cb(cur, "ffn_norm", il);
|
|
|
|
|
|
|
+ } else {
|
|
|
|
|
+ // MoE branch
|
|
|
|
|
+ cur = build_norm(ffn_inp,
|
|
|
|
|
+ model.layers[il].ffn_norm, NULL,
|
|
|
|
|
+ LLM_NORM_RMS, il);
|
|
|
|
|
+ cb(cur, "ffn_norm", il);
|
|
|
|
|
|
|
|
- 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,
|
|
|
|
|
- model.layers[il].ffn_down_exps,
|
|
|
|
|
- nullptr,
|
|
|
|
|
- n_expert, n_expert_used,
|
|
|
|
|
- LLM_FFN_SILU, true,
|
|
|
|
|
- false, 0.0,
|
|
|
|
|
- LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
|
|
|
|
|
- il);
|
|
|
|
|
- cb(moe_out, "ffn_moe_out", il);
|
|
|
|
|
|
|
+ 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,
|
|
|
|
|
+ model.layers[il].ffn_down_exps,
|
|
|
|
|
+ nullptr,
|
|
|
|
|
+ n_expert, n_expert_used,
|
|
|
|
|
+ LLM_FFN_SILU, true,
|
|
|
|
|
+ false, 0.0,
|
|
|
|
|
+ LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
|
|
|
|
|
+ il);
|
|
|
|
|
+ cb(moe_out, "ffn_moe_out", il);
|
|
|
|
|
|
|
|
- // For Granite MoE Shared
|
|
|
|
|
- if (hparams.n_ff_shexp > 0) {
|
|
|
|
|
- ggml_tensor * ffn_shexp = build_ffn(cur,
|
|
|
|
|
- model.layers[il].ffn_up_shexp, NULL, NULL,
|
|
|
|
|
- model.layers[il].ffn_gate_shexp, NULL, NULL,
|
|
|
|
|
- model.layers[il].ffn_down_shexp, NULL, NULL,
|
|
|
|
|
- NULL,
|
|
|
|
|
- LLM_FFN_SILU, LLM_FFN_PAR, il);
|
|
|
|
|
- cb(ffn_shexp, "ffn_shexp", il);
|
|
|
|
|
|
|
+ // For Granite MoE Shared
|
|
|
|
|
+ if (hparams.n_ff_shexp > 0) {
|
|
|
|
|
+ ggml_tensor * ffn_shexp = build_ffn(cur,
|
|
|
|
|
+ model.layers[il].ffn_up_shexp, NULL, NULL,
|
|
|
|
|
+ model.layers[il].ffn_gate_shexp, NULL, NULL,
|
|
|
|
|
+ model.layers[il].ffn_down_shexp, NULL, NULL,
|
|
|
|
|
+ NULL,
|
|
|
|
|
+ LLM_FFN_SILU, LLM_FFN_PAR, il);
|
|
|
|
|
+ cb(ffn_shexp, "ffn_shexp", il);
|
|
|
|
|
|
|
|
- cur = ggml_add(ctx0, moe_out, ffn_shexp);
|
|
|
|
|
- cb(cur, "ffn_out", il);
|
|
|
|
|
- } else {
|
|
|
|
|
- cur = moe_out;
|
|
|
|
|
- }
|
|
|
|
|
|
|
+ cur = ggml_add(ctx0, moe_out, ffn_shexp);
|
|
|
|
|
+ cb(cur, "ffn_out", il);
|
|
|
|
|
+ } else {
|
|
|
|
|
+ cur = moe_out;
|
|
|
}
|
|
}
|
|
|
|
|
+ }
|
|
|
|
|
|
|
|
- // For Granite architectures - scale residual
|
|
|
|
|
|
|
+ // For Granite architectures - scale residual
|
|
|
|
|
+ if (hparams.f_residual_scale) {
|
|
|
cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
|
|
cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
|
|
|
- cur = ggml_add(ctx0, cur, ffn_inp);
|
|
|
|
|
- cb(cur, "ffn_out", il);
|
|
|
|
|
|
|
+ }
|
|
|
|
|
+ cur = ggml_add(ctx0, cur, ffn_inp);
|
|
|
|
|
+ cb(cur, "ffn_out", il);
|
|
|
|
|
|
|
|
- cur = build_cvec(cur, il);
|
|
|
|
|
- cb(cur, "l_out", il);
|
|
|
|
|
|
|
+ cur = build_cvec(cur, il);
|
|
|
|
|
+ cb(cur, "l_out", il);
|
|
|
|
|
+
|
|
|
|
|
+ return cur;
|
|
|
|
|
+ }
|
|
|
|
|
+};
|
|
|
|
|
+
|
|
|
|
|
+struct llm_build_granite_hybrid : public llm_graph_context_mamba {
|
|
|
|
|
+
|
|
|
|
|
+ llm_build_granite_hybrid(
|
|
|
|
|
+ const llama_model & model,
|
|
|
|
|
+ const llm_graph_params & params,
|
|
|
|
|
+ ggml_cgraph * gf) :
|
|
|
|
|
+ llm_graph_context_mamba(params) {
|
|
|
|
|
+
|
|
|
|
|
+ const int64_t n_embd_head = hparams.n_embd_head_v;
|
|
|
|
|
+ GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * cur;
|
|
|
|
|
+ ggml_tensor * inpL;
|
|
|
|
|
+
|
|
|
|
|
+ inpL = build_inp_embd(model.tok_embd);
|
|
|
|
|
+
|
|
|
|
|
+ auto * inp = build_inp_mem_hybrid();
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * inp_out_ids = build_inp_out_ids();
|
|
|
|
|
+
|
|
|
|
|
+ // Positional embeddings populated if rope enabled
|
|
|
|
|
+ ggml_tensor * inp_pos = nullptr;
|
|
|
|
|
+ if (hparams.rope_finetuned) {
|
|
|
|
|
+ inp_pos = build_inp_pos();
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ for (int il = 0; il < n_layer; ++il) {
|
|
|
|
|
+ struct ggml_tensor * inpSA = inpL;
|
|
|
|
|
+
|
|
|
|
|
+ // norm
|
|
|
|
|
+ cur = build_norm(inpL,
|
|
|
|
|
+ model.layers[il].attn_norm, NULL,
|
|
|
|
|
+ LLM_NORM_RMS, il);
|
|
|
|
|
+ cb(cur, "attn_norm", il);
|
|
|
|
|
+
|
|
|
|
|
+ if (hparams.is_recurrent(il)) {
|
|
|
|
|
+ // ssm layer //
|
|
|
|
|
+ cur = build_mamba2_layer(inp->get_recr(), gf, cur, model, ubatch, il);
|
|
|
|
|
+ } else {
|
|
|
|
|
+ // attention layer //
|
|
|
|
|
+ cur = build_attention_layer(
|
|
|
|
|
+ gf, cur, inp_pos, inp->get_attn(), model,
|
|
|
|
|
+ n_embd_head, il);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ if (il == n_layer - 1 && inp_out_ids) {
|
|
|
|
|
+ cur = ggml_get_rows(ctx0, cur, inp_out_ids);
|
|
|
|
|
+ inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ // ffn
|
|
|
|
|
+ cur = build_layer_ffn(cur, inpSA, model, il);
|
|
|
|
|
|
|
|
// input for next layer
|
|
// input for next layer
|
|
|
inpL = cur;
|
|
inpL = cur;
|
|
@@ -13978,12 +14217,156 @@ struct llm_build_granite : public llm_graph_context {
|
|
|
cur = build_lora_mm(model.output, cur);
|
|
cur = build_lora_mm(model.output, cur);
|
|
|
|
|
|
|
|
// For Granite architectures - scale logits
|
|
// For Granite architectures - scale logits
|
|
|
- cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_logit_scale);
|
|
|
|
|
|
|
+ if (hparams.f_logit_scale) {
|
|
|
|
|
+ cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_logit_scale);
|
|
|
|
|
+ }
|
|
|
cb(cur, "result_output", -1);
|
|
cb(cur, "result_output", -1);
|
|
|
res->t_logits = cur;
|
|
res->t_logits = cur;
|
|
|
|
|
|
|
|
ggml_build_forward_expand(gf, cur);
|
|
ggml_build_forward_expand(gf, cur);
|
|
|
}
|
|
}
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * build_attention_layer(
|
|
|
|
|
+ ggml_cgraph * gf,
|
|
|
|
|
+ ggml_tensor * cur,
|
|
|
|
|
+ ggml_tensor * inp_pos,
|
|
|
|
|
+ llm_graph_input_attn_kv_unified * inp_attn,
|
|
|
|
|
+ const llama_model & model,
|
|
|
|
|
+ const int64_t n_embd_head,
|
|
|
|
|
+ const int il) {
|
|
|
|
|
+
|
|
|
|
|
+ // compute Q and K and (optionally) RoPE them
|
|
|
|
|
+ ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
|
|
|
|
|
+ cb(Qcur, "Qcur", il);
|
|
|
|
|
+ if (model.layers[il].bq) {
|
|
|
|
|
+ Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
|
|
|
|
|
+ cb(Qcur, "Qcur", il);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
|
|
|
|
|
+ cb(Kcur, "Kcur", il);
|
|
|
|
|
+ if (model.layers[il].bk) {
|
|
|
|
|
+ Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
|
|
|
|
|
+ cb(Kcur, "Kcur", il);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
|
|
|
|
|
+ cb(Vcur, "Vcur", il);
|
|
|
|
|
+ if (model.layers[il].bv) {
|
|
|
|
|
+ Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
|
|
|
|
|
+ cb(Vcur, "Vcur", il);
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, hparams.n_head(il), n_tokens);
|
|
|
|
|
+ Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, hparams.n_head_kv(il), n_tokens);
|
|
|
|
|
+ Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, hparams.n_head_kv(il), n_tokens);
|
|
|
|
|
+
|
|
|
|
|
+ const bool use_rope = hparams.rope_finetuned;
|
|
|
|
|
+ if (use_rope) {
|
|
|
|
|
+ ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
|
|
|
|
|
+ Qcur = ggml_rope_ext(
|
|
|
|
|
+ ctx0, Qcur, inp_pos, rope_factors,
|
|
|
|
|
+ n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
|
|
|
|
|
+ ext_factor, attn_factor, beta_fast, beta_slow
|
|
|
|
|
+ );
|
|
|
|
|
+
|
|
|
|
|
+ Kcur = ggml_rope_ext(
|
|
|
|
|
+ ctx0, Kcur, inp_pos, rope_factors,
|
|
|
|
|
+ n_rot, 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);
|
|
|
|
|
+ cb(Vcur, "Vcur", il);
|
|
|
|
|
+
|
|
|
|
|
+ const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
|
|
|
|
|
+ cur = build_attn(inp_attn, gf,
|
|
|
|
|
+ model.layers[il].wo, model.layers[il].bo,
|
|
|
|
|
+ Qcur, Kcur, Vcur, nullptr, nullptr, kq_scale, il);
|
|
|
|
|
+ cb(cur, "attn_out", il);
|
|
|
|
|
+ return cur;
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ ggml_tensor * build_layer_ffn(
|
|
|
|
|
+ ggml_tensor * cur,
|
|
|
|
|
+ ggml_tensor * inpSA,
|
|
|
|
|
+ const llama_model & model,
|
|
|
|
|
+ const int il) {
|
|
|
|
|
+
|
|
|
|
|
+ // For Granite architectures - scale residual
|
|
|
|
|
+ if (hparams.f_residual_scale) {
|
|
|
|
|
+ cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
|
|
|
|
|
+ }
|
|
|
|
|
+ ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
|
|
|
|
|
+ cb(ffn_inp, "ffn_inp", il);
|
|
|
|
|
+
|
|
|
|
|
+ // feed-forward network (non-MoE)
|
|
|
|
|
+ if (model.layers[il].ffn_gate_inp == nullptr) {
|
|
|
|
|
+
|
|
|
|
|
+ cur = build_norm(ffn_inp,
|
|
|
|
|
+ model.layers[il].ffn_norm, NULL,
|
|
|
|
|
+ LLM_NORM_RMS, il);
|
|
|
|
|
+ cb(cur, "ffn_norm", il);
|
|
|
|
|
+
|
|
|
|
|
+ cur = build_ffn(cur,
|
|
|
|
|
+ model.layers[il].ffn_up, model.layers[il].ffn_up_b, NULL,
|
|
|
|
|
+ model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
|
|
|
|
|
+ model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
|
|
|
|
|
+ NULL,
|
|
|
|
|
+ LLM_FFN_SILU, LLM_FFN_PAR, il);
|
|
|
|
|
+ cb(cur, "ffn_out", il);
|
|
|
|
|
+
|
|
|
|
|
+ } else {
|
|
|
|
|
+ // MoE branch
|
|
|
|
|
+ cur = build_norm(ffn_inp,
|
|
|
|
|
+ model.layers[il].ffn_norm, NULL,
|
|
|
|
|
+ LLM_NORM_RMS, il);
|
|
|
|
|
+ cb(cur, "ffn_norm", il);
|
|
|
|
|
+
|
|
|
|
|
+ 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,
|
|
|
|
|
+ model.layers[il].ffn_down_exps,
|
|
|
|
|
+ nullptr,
|
|
|
|
|
+ n_expert, n_expert_used,
|
|
|
|
|
+ LLM_FFN_SILU, true,
|
|
|
|
|
+ false, 0.0,
|
|
|
|
|
+ LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
|
|
|
|
|
+ il);
|
|
|
|
|
+ cb(moe_out, "ffn_moe_out", il);
|
|
|
|
|
+
|
|
|
|
|
+ // For Granite MoE Shared
|
|
|
|
|
+ if (hparams.n_ff_shexp > 0) {
|
|
|
|
|
+ ggml_tensor * ffn_shexp = build_ffn(cur,
|
|
|
|
|
+ model.layers[il].ffn_up_shexp, NULL, NULL,
|
|
|
|
|
+ model.layers[il].ffn_gate_shexp, NULL, NULL,
|
|
|
|
|
+ model.layers[il].ffn_down_shexp, NULL, NULL,
|
|
|
|
|
+ NULL,
|
|
|
|
|
+ LLM_FFN_SILU, LLM_FFN_PAR, il);
|
|
|
|
|
+ cb(ffn_shexp, "ffn_shexp", il);
|
|
|
|
|
+
|
|
|
|
|
+ cur = ggml_add(ctx0, moe_out, ffn_shexp);
|
|
|
|
|
+ cb(cur, "ffn_out", il);
|
|
|
|
|
+ } else {
|
|
|
|
|
+ cur = moe_out;
|
|
|
|
|
+ }
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
+ // For Granite architectures - scale residual
|
|
|
|
|
+ if (hparams.f_residual_scale) {
|
|
|
|
|
+ cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
|
|
|
|
|
+ }
|
|
|
|
|
+ cur = ggml_add(ctx0, cur, ffn_inp);
|
|
|
|
|
+ cb(cur, "ffn_out", il);
|
|
|
|
|
+
|
|
|
|
|
+ cur = build_cvec(cur, il);
|
|
|
|
|
+ cb(cur, "l_out", il);
|
|
|
|
|
+
|
|
|
|
|
+ return cur;
|
|
|
|
|
+ }
|
|
|
};
|
|
};
|
|
|
|
|
|
|
|
// ref: https://github.com/facebookresearch/chameleon
|
|
// ref: https://github.com/facebookresearch/chameleon
|
|
@@ -15834,6 +16217,10 @@ llm_graph_result_ptr llama_model::build_graph(
|
|
|
{
|
|
{
|
|
|
llm = std::make_unique<llm_build_granite>(*this, params, gf);
|
|
llm = std::make_unique<llm_build_granite>(*this, params, gf);
|
|
|
} break;
|
|
} break;
|
|
|
|
|
+ case LLM_ARCH_GRANITE_HYBRID:
|
|
|
|
|
+ {
|
|
|
|
|
+ llm = std::make_unique<llm_build_granite_hybrid>(*this, params, gf);
|
|
|
|
|
+ } break;
|
|
|
case LLM_ARCH_CHAMELEON:
|
|
case LLM_ARCH_CHAMELEON:
|
|
|
{
|
|
{
|
|
|
llm = std::make_unique<llm_build_chameleon>(*this, params, gf);
|
|
llm = std::make_unique<llm_build_chameleon>(*this, params, gf);
|
|
@@ -16023,6 +16410,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
|
|
|
case LLM_ARCH_GLM4:
|
|
case LLM_ARCH_GLM4:
|
|
|
case LLM_ARCH_GRANITE:
|
|
case LLM_ARCH_GRANITE:
|
|
|
case LLM_ARCH_GRANITE_MOE:
|
|
case LLM_ARCH_GRANITE_MOE:
|
|
|
|
|
+ case LLM_ARCH_GRANITE_HYBRID:
|
|
|
case LLM_ARCH_CHAMELEON:
|
|
case LLM_ARCH_CHAMELEON:
|
|
|
case LLM_ARCH_BAILINGMOE:
|
|
case LLM_ARCH_BAILINGMOE:
|
|
|
case LLM_ARCH_NEO_BERT:
|
|
case LLM_ARCH_NEO_BERT:
|
Gabe Goodhart