Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 17:25:07 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	.clang-tidy
#	.devops/musa.Dockerfile
#	.github/workflows/build-linux-cross.yml
#	.github/workflows/build.yml
#	.github/workflows/docker.yml
#	.gitignore
#	CODEOWNERS
#	CONTRIBUTING.md
#	README.md
#	build-xcframework.sh
#	ci/README-MUSA.md
#	ci/run.sh
#	common/CMakeLists.txt
#	docs/docker.md
#	examples/CMakeLists.txt
#	examples/eval-callback/CMakeLists.txt
#	examples/model-conversion/Makefile
#	examples/model-conversion/README.md
#	examples/model-conversion/logits.cpp
#	examples/model-conversion/scripts/causal/compare-logits.py
#	examples/model-conversion/scripts/causal/run-org-model.py
#	examples/model-conversion/scripts/embedding/compare-embeddings-logits.sh
#	examples/model-conversion/scripts/embedding/run-converted-model.sh
#	examples/model-conversion/scripts/embedding/run-original-model.py
#	examples/model-conversion/scripts/utils/check-nmse.py
#	examples/model-conversion/scripts/utils/inspect-org-model.py
#	examples/model-conversion/scripts/utils/semantic_check.py
#	ggml/CMakeLists.txt
#	ggml/include/ggml-zdnn.h
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-opencl/kernels/set_rows.cl
#	ggml/src/ggml-rpc/ggml-rpc.cpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	ggml/src/ggml-sycl/set_rows.cpp
#	ggml/src/ggml-webgpu/ggml-webgpu.cpp
#	ggml/src/ggml-zdnn/ggml-zdnn.cpp
#	tests/CMakeLists.txt
#	tests/test-backend-ops.cpp
#	tests/test-quantize-perf.cpp
#	tests/test-tokenizers-repo.sh
#	tools/perplexity/perplexity.cpp
#	tools/server/tests/README.md
This commit is contained in:
Concedo
2025-09-27 17:09:14 +08:00
parent cbb8a3feb5 72b24d96c6
commit b120e107f9
112 changed files with 10916 additions and 4106 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+249 -7
View File
@@ -71,6 +71,7 @@ const char * llm_type_name(llm_type type) {
case LLM_TYPE_1_7B: return "1.7B";
case LLM_TYPE_1_8B: return "1.8B";
case LLM_TYPE_2B: return "2B";
case LLM_TYPE_2_6B: return "2.6B";
case LLM_TYPE_2_8B: return "2.8B";
case LLM_TYPE_2_9B: return "2.9B";
case LLM_TYPE_3B: return "3B";
@@ -679,10 +680,17 @@ void llama_model::load_hparams(llama_model_loader & ml) {
} break;
case LLM_ARCH_MINICPM:
{
// Backward-compatible defaults for older MiniCPM GGUFs
hparams.f_embedding_scale = 12.0f;
hparams.f_residual_scale = 1.4f / sqrtf(float(hparams.n_layer));
hparams.f_logit_scale = hparams.n_embd ? (256.0f / float(hparams.n_embd)) : 1.0f;
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale);
ml.get_key(LLM_KV_RESIDUAL_SCALE, hparams.f_residual_scale);
ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale);
// Optional KV reads, override defaults if present in newer GGUF exports
ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale, /*required=*/false);
ml.get_key(LLM_KV_RESIDUAL_SCALE, hparams.f_residual_scale, /*required=*/false);
ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale, /*required=*/false);
// MiniCPM uses rope by default, unlike Granite which uses it as a switch
hparams.rope_finetuned = true;
@@ -1982,10 +1990,11 @@ void llama_model::load_hparams(llama_model_loader & ml) {
for (uint32_t il = 0; il < hparams.n_layer; ++il) {
hparams.recurrent_layer_arr[il] = hparams.n_head_kv(il) == 0;
}
switch (hparams.n_embd) {
case 1024: type = LLM_TYPE_350M; break;
case 1536: type = LLM_TYPE_700M; break;
case 2048: type = LLM_TYPE_1_2B; break;
switch (hparams.n_ff()) {
case 4608: type = LLM_TYPE_350M; break;
case 6912: type = LLM_TYPE_700M; break;
case 8192: type = LLM_TYPE_1_2B; break;
case 10752: type = LLM_TYPE_2_6B; break;
default: type = LLM_TYPE_UNKNOWN;
}
} break;
@@ -2012,6 +2021,19 @@ void llama_model::load_hparams(llama_model_loader & ml) {
default: type = LLM_TYPE_UNKNOWN;
}
} break;
case LLM_ARCH_GROVEMOE:
{
ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp);
ml.get_key(LLM_KV_EXPERT_CHUNK_FEED_FORWARD_LENGTH, hparams.n_ff_chexp);
ml.get_key(LLM_KV_EXPERT_GROUP_SCALE, hparams.expert_group_scale);
ml.get_key(LLM_KV_EXPERTS_PER_GROUP, hparams.n_group_experts);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
switch (hparams.n_layer) {
case 48: type = LLM_TYPE_30B_A3B; break;
default: type = LLM_TYPE_UNKNOWN;
}
} break;
default: throw std::runtime_error("unsupported model architecture");
}
@@ -3223,6 +3245,9 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
}
// output rerank head
cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);
for (int i = 0; i < n_layer; ++i) {
auto & layer = layers[i];
@@ -5893,6 +5918,53 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert }, 0);
}
} break;
case LLM_ARCH_GROVEMOE:
{
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
if (output == NULL) {
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
}
GGML_ASSERT(n_expert > 0 && "n_expert must be > 0 for GROVEMOE");
GGML_ASSERT(n_expert_used > 0 && "n_expert_used must be > 0 for GROVEMOE");
GGML_ASSERT(hparams.n_group_experts > 0 && "n_group_experts must be > 0 for GROVEMOE");
for (int i = 0; i < n_layer; ++i) {
auto & layer = layers[i];
layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0);
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0);
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k}, 0);
layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k}, 0);
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);
// MoE branch
const int64_t n_ff_exp = hparams.n_ff_exp ? hparams.n_ff_exp : n_ff / n_expert_used;
const int64_t n_ff_chexp = hparams.n_ff_chexp ? hparams.n_ff_chexp : n_embd_head_k;
const int64_t n_chunk_expert = n_expert / hparams.n_group_experts;
layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert}, 0);
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_gate_chexps = create_tensor(tn(LLM_TENSOR_FFN_GATE_CHEXPS, "weight", i), { n_embd, n_ff_chexp, n_chunk_expert}, 0);
layer.ffn_down_chexps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_CHEXPS, "weight", i), {n_ff_chexp, n_embd, n_chunk_expert}, 0);
layer.ffn_up_chexps = create_tensor(tn(LLM_TENSOR_FFN_UP_CHEXPS, "weight", i), { n_embd, n_ff_chexp, n_chunk_expert}, 0);
}
} break;
default:
throw std::runtime_error("unknown architecture");
}
@@ -6062,6 +6134,14 @@ size_t llama_model::n_devices() const {
return devices.size();
}
std::map<ggml_backend_buffer_type_t, size_t> llama_model::memory_breakdown() const {
std::map<ggml_backend_buffer_type_t, size_t> ret;
for (const ggml_backend_buffer_ptr & buf_ptr : pimpl->bufs) {
ret[ggml_backend_buffer_get_type(buf_ptr.get())] += ggml_backend_buffer_get_size(buf_ptr.get());
}
return ret;
}
uint64_t llama_model::n_elements() const {
return pimpl->n_elements;
}
@@ -6225,6 +6305,13 @@ void llama_model::print_info() const {
LLAMA_LOG_INFO("%s: expert_gating_func = %s\n", __func__, llama_expert_gating_func_name((llama_expert_gating_func_type) hparams.expert_gating_func));
}
if (arch == LLM_ARCH_GROVEMOE) {
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
LLAMA_LOG_INFO("%s: n_ff_chexp = %d\n", __func__, hparams.n_ff_chexp);
LLAMA_LOG_INFO("%s: n_group_experts = %d\n", __func__, hparams.n_group_experts);
LLAMA_LOG_INFO("%s: expert_group_scale = %.2f\n", __func__, hparams.expert_group_scale);
}
vocab.print_info();
}
@@ -18913,6 +19000,156 @@ struct llm_build_smallthinker : public llm_graph_context{
}
};
struct llm_build_grovemoe : public llm_graph_context {
llm_build_grovemoe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_chunk_expert = n_expert / hparams.n_group_experts;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_head == hparams.n_rot);
ggml_tensor * cur;
ggml_tensor * inpL;
inpL = build_inp_embd(model.tok_embd);
// inp_pos - contains the positions
ggml_tensor * inp_pos = build_inp_pos();
auto * inp_attn = build_attn_inp_kv();
ggml_tensor * inp_out_ids = build_inp_out_ids();
for (int il = 0; il < n_layer; ++il) {
ggml_tensor * inpSA = inpL;
// norm
cur = build_norm(inpL,
model.layers[il].attn_norm, NULL,
LLM_NORM_RMS, il);
cb(cur, "attn_norm", il);
// self_attention
{
// compute Q and K and RoPE them
ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
cb(Qcur, "Qcur", il);
ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
cb(Kcur, "Kcur", il);
ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
cb(Vcur, "Vcur", 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);
Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
cb(Qcur, "Qcur_normed", il);
Qcur = ggml_rope_ext(
ctx0, Qcur, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
ext_factor, attn_factor, beta_fast, beta_slow
);
Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
cb(Kcur, "Kcur_normed", il);
Kcur = ggml_rope_ext(
ctx0, Kcur, inp_pos, nullptr,
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);
cur = build_attn(inp_attn,
model.layers[il].wo, model.layers[il].bo,
Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(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);
}
ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
cb(ffn_inp, "ffn_inp", il);
// MoE branch
cur = build_norm(ffn_inp,
model.layers[il].ffn_norm, NULL,
LLM_NORM_RMS, il);
cb(cur, "ffn_norm", il);
ggml_tensor * probs = build_lora_mm(model.layers[il].ffn_gate_inp, cur); // [n_expert, n_tokens]
cb(probs, "ffn_moe_logits", il);
ggml_tensor * moe_out =
build_moe_ffn(cur,
nullptr,
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, probs);
cb(moe_out, "ffn_moe_out", il);
cur = moe_out;
// TODO: Only do the expert selection and weights once
moe_out =
build_moe_ffn(cur,
nullptr,
model.layers[il].ffn_up_chexps,
model.layers[il].ffn_gate_chexps,
model.layers[il].ffn_down_chexps,
nullptr,
n_chunk_expert, n_expert_used > n_chunk_expert ? n_chunk_expert : n_expert_used,
LLM_FFN_SILU, true,
false, 0.0,
LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
il, probs);
cb(moe_out, "ffn_adj_moe_out", il);
cur = ggml_add(ctx0, cur, ggml_scale(ctx0, moe_out, hparams.expert_group_scale));
cb(cur, "ffn_final_moe_out", il);
cur = ggml_add(ctx0, cur, ffn_inp);
cur = build_cvec(cur, il);
cb(cur, "l_out", il);
// input for next layer
inpL = cur;
}
cur = inpL;
cur = build_norm(cur,
model.output_norm, NULL,
LLM_NORM_RMS, -1);
cb(cur, "result_norm", -1);
res->t_embd = cur;
// lm_head
cur = build_lora_mm(model.output, cur);
cb(cur, "result_output", -1);
res->t_logits = cur;
ggml_build_forward_expand(gf, cur);
}
};
llama_memory_i * llama_model::create_memory(const llama_memory_params & params, llama_cparams & cparams) const {
llama_memory_i * res;
@@ -19439,6 +19676,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
llm = std::make_unique<llm_build_smallthinker<false>>(*this, params);
}
} break;
case LLM_ARCH_GROVEMOE:
{
llm = std::make_unique<llm_build_grovemoe>(*this, params);
} break;
default:
GGML_ABORT("fatal error");
}
@@ -19644,6 +19885,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_SMALLTHINKER:
case LLM_ARCH_GLM4_MOE:
case LLM_ARCH_SEED_OSS:
case LLM_ARCH_GROVEMOE:
return LLAMA_ROPE_TYPE_NEOX;
case LLM_ARCH_QWEN2VL: