Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-17 08:19:49 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	.github/workflows/build.yml
#	Makefile
#	Package.swift
#	ci/run.sh
#	docs/backend/SYCL.md
#	examples/llama-bench/llama-bench.cpp
#	examples/server/CMakeLists.txt
#	examples/server/README.md
#	ggml/CMakeLists.txt
#	ggml/src/CMakeLists.txt
#	grammars/README.md
#	scripts/sync-ggml-am.sh
#	scripts/sync-ggml.last
#	scripts/sync-ggml.sh
#	tests/run-json-schema-to-grammar.mjs
#	tests/test-backend-ops.cpp
This commit is contained in:
Concedo
2024-11-09 13:36:47 +08:00
parent 47968e610e ec450d3bbf
commit a244b1ffd2
55 changed files with 31704 additions and 2916 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama.cpp
+4 -4
View File
@@ -7052,7 +7052,7 @@ static const std::map<llm_tensor, llm_tensor_info> llm_tensor_info_mapping = {
{LLM_TENSOR_TIME_MIX_LERP_R, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
{LLM_TENSOR_TIME_MIX_LERP_G, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
{LLM_TENSOR_TIME_MIX_DECAY, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
{LLM_TENSOR_TIME_MIX_FIRST, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_RWKV_WKV}},
{LLM_TENSOR_TIME_MIX_FIRST, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_RWKV_WKV6}},
{LLM_TENSOR_ATTN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_ATTN_NORM_2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_ATTN_OUT_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
@@ -7168,7 +7168,7 @@ static bool weight_buft_supported(const llama_hparams & hparams, ggml_tensor * w
ggml_tensor * C = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_state, n_seq_tokens, n_seqs);
op_tensor = ggml_ssm_scan(ctx, s, x, dt, w, B, C);
} break;
case GGML_OP_RWKV_WKV:
case GGML_OP_RWKV_WKV6:
{
// FIXME
const int64_t S = 123;
@@ -7181,7 +7181,7 @@ static bool weight_buft_supported(const llama_hparams & hparams, ggml_tensor * w
ggml_tensor * tf = w;
ggml_tensor * td = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 1, S, H, n_tokens);
ggml_tensor * state = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, S, n_seqs, S, H);
op_tensor = ggml_rwkv_wkv(ctx, k, v, r, tf, td, state);
op_tensor = ggml_rwkv_wkv6(ctx, k, v, r, tf, td, state);
} break;
default:
GGML_ABORT("%s: missing test for op %s for tensor %s", __func__, ggml_op_name(op), w->name);
@@ -10145,7 +10145,7 @@ static struct ggml_tensor * llm_build_rwkv6_time_mix(
v = ggml_transpose(ctx, v);
r = ggml_transpose(ctx, r);
struct ggml_tensor * wkv_output = ggml_rwkv_wkv(ctx, k, v, r, layer->time_mix_first, w, *wkv_state);
struct ggml_tensor * wkv_output = ggml_rwkv_wkv6(ctx, k, v, r, layer->time_mix_first, w, *wkv_state);
cur = ggml_view_1d(ctx, wkv_output, n_embd * n_tokens, 0);
*wkv_state = ggml_view_1d(ctx, wkv_output, n_embd * head_size * n_seqs, n_embd * n_tokens * sizeof(float));