vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Optimize RWKV6 Operator Naming and Implement Multi-core CPU/ SYCL Acceleration (#10133)

Browse source 
* rwkv6: rename to wkv6

* rwkv6: support avx2 avx512 armv8 armv9

* rwkv6: update cuda file name

* rwkv6: rename params

* wkv on sycl

* sycl: add some ops

* sycl: Enhance OP support judgment

* wkv6: drop armv9 and tranfer to GGML style

ggml-ci

* sync : ggml

* update the function to use appropriate types

* fix define error

* Update ggml/src/ggml-cpu.c

* add appropriate asserts

* move element-wise functions outside

* put the declaration outside the loop

* rewrite to be more inline with the common pattern for distributing threads

* use recommended way GGML_TENSOR_LOCALS

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Diego Devesa <slarengh@gmail.com>
Co-authored-by: Plamen Minev <pacominev@gmail.com>
Co-authored-by: Yuri Khrustalev <ykhrustalev@users.noreply.github.com>
Co-authored-by: Meng, Hengyu <airdldl@163.com>
This commit is contained in:
Zhiyuan Li 2024-11-07 18:19:10 +11:00 • committed by GitHub
parent 5c333e0140
commit 3bcd40b3c5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
22 changed files with 1977 additions and 1027 deletions
Download patch file Download diff file Expand all files Collapse all files

16
tests/test-backend-ops.cpp
View file

@ -1614,8 +1614,8 @@ struct test_ssm_scan : public test_case {
}
};
// GGML_OP_RWKV_WKV
struct test_rwkv_wkv : public test_case {
// GGML_OP_RWKV_WKV6
struct test_rwkv_wkv6 : public test_case {
const ggml_type type;
const int64_t head_count;
@ -1627,7 +1627,7 @@ struct test_rwkv_wkv : public test_case {
return VARS_TO_STR5(type, head_count, head_size, n_seq_tokens, n_seqs);
}
test_rwkv_wkv(ggml_type type = GGML_TYPE_F32,
test_rwkv_wkv6(ggml_type type = GGML_TYPE_F32,
int64_t head_count = 32, int64_t head_size = 64, int64_t n_seq_tokens = 32, int64_t n_seqs = 32)
: type(type), head_count(head_count), head_size(head_size), n_seq_tokens(n_seq_tokens), n_seqs(n_seqs) {}
@ -1639,7 +1639,7 @@ struct test_rwkv_wkv : public test_case {
ggml_tensor * tf = ggml_new_tensor(ctx, type, 2, std::vector<int64_t>{ head_size, head_count }.data());
ggml_tensor * td = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ 1, head_size, head_count, n_tokens }.data());
ggml_tensor * s = ggml_new_tensor(ctx, type, 2, std::vector<int64_t>{ head_size * head_size * head_count, n_seqs }.data());
ggml_tensor * out = ggml_rwkv_wkv(ctx, k, v, r, tf, td, s);
ggml_tensor * out = ggml_rwkv_wkv6(ctx, k, v, r, tf, td, s);
return out;
}
};
@ -3499,10 +3499,10 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_ssm_scan(GGML_TYPE_F32, 16, 1024, 32, 4));
test_cases.emplace_back(new test_rwkv_wkv(GGML_TYPE_F32, 32, 64, 1, 1));
test_cases.emplace_back(new test_rwkv_wkv(GGML_TYPE_F32, 32, 64, 32, 1));
test_cases.emplace_back(new test_rwkv_wkv(GGML_TYPE_F32, 32, 64, 32, 4));
test_cases.emplace_back(new test_rwkv_wkv(GGML_TYPE_F32, 32, 64, 128, 4));
test_cases.emplace_back(new test_rwkv_wkv6(GGML_TYPE_F32, 32, 64, 1, 1));
test_cases.emplace_back(new test_rwkv_wkv6(GGML_TYPE_F32, 32, 64, 32, 1));
test_cases.emplace_back(new test_rwkv_wkv6(GGML_TYPE_F32, 32, 64, 32, 4));
test_cases.emplace_back(new test_rwkv_wkv6(GGML_TYPE_F32, 32, 64, 128, 4));
#if 1
for (ggml_type type_a : base_types) {