root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-11 13:30:35 +00:00
Code Issues Actions 11 Packages Projects Releases Wiki Activity

resolve test-backend-ops conflicts

Browse Source
This commit is contained in:
pidack 2024-08-27 17:31:40 +08:00
parent 40f47872b3
commit 1928967874
1 changed files with 2 additions and 62 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

64
tests/test-backend-ops.cpp
View File

@ -474,8 +474,8 @@ struct test_case {
if (memcmp(t1_data.data(), t2_data.data(), ggml_nbytes(t1)) != 0) { if (memcmp(t1_data.data(), t2_data.data(), ggml_nbytes(t1)) != 0) {
printf("sentinel mismatch: %s ", t1->name); printf("sentinel mismatch: %s ", t1->name);
// ud->ok = false; ud->ok = false;
// return true; return true;
} }
} }
@ -1694,64 +1694,6 @@ struct test_leaky_relu : public test_case {
} }
}; };
// GGML_OP_SSM_CONV
struct test_ssm_conv : public test_case {
const ggml_type type;
const int64_t d_conv;
const int64_t d_inner;
const int64_t n_seq_tokens;
const int64_t n_seqs;
std::string vars() override {
return VARS_TO_STR5(type, d_conv, d_inner, n_seq_tokens, n_seqs);
}
test_ssm_conv(ggml_type type = GGML_TYPE_F32,
int64_t d_conv = 4,
int64_t d_inner = 1536,
int64_t n_seq_tokens = 7,
int64_t n_seqs = 2)
: type(type), d_conv(d_conv), d_inner(d_inner), n_seq_tokens(n_seq_tokens), n_seqs(n_seqs) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * sx = ggml_new_tensor_3d(ctx, type, d_conv - 1 + n_seq_tokens, d_inner, n_seqs);
ggml_tensor * c = ggml_new_tensor_2d(ctx, type, d_conv, d_inner);
ggml_tensor * out = ggml_ssm_conv(ctx, sx, c);
return out;
}
};
// GGML_OP_SSM_SCAN
struct test_ssm_scan : public test_case {
const ggml_type type;
const int64_t d_state;
const int64_t d_inner;
const int64_t n_seq_tokens;
const int64_t n_seqs;
std::string vars() override {
return VARS_TO_STR5(type, d_state, d_inner, n_seq_tokens, n_seqs);
}
test_ssm_scan(ggml_type type = GGML_TYPE_F32,
int64_t d_state = 16,
int64_t d_inner = 1536,
int64_t n_seq_tokens = 7,
int64_t n_seqs = 2)
: type(type), d_state(d_state), d_inner(d_inner), n_seq_tokens(n_seq_tokens), n_seqs(n_seqs) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * s = ggml_new_tensor_3d(ctx, type, d_state, d_inner, n_seqs);
ggml_tensor * x = ggml_new_tensor_3d(ctx, type, d_inner, n_seq_tokens, n_seqs);
ggml_tensor * dt = ggml_new_tensor_3d(ctx, type, d_inner, n_seq_tokens, n_seqs);
ggml_tensor * A = ggml_new_tensor_2d(ctx, type, d_state, d_inner);
ggml_tensor * B = ggml_new_tensor_3d(ctx, type, d_state, n_seq_tokens, n_seqs);
ggml_tensor * C = ggml_new_tensor_3d(ctx, type, d_state, n_seq_tokens, n_seqs);
ggml_tensor * out = ggml_ssm_scan(ctx, s, x, dt, A, B, C);
return out;
}
};
// GGML_OP_FLASH_ATTN_EXT // GGML_OP_FLASH_ATTN_EXT
struct test_flash_attn_ext : public test_case { struct test_flash_attn_ext : public test_case {
const int64_t hs; // head size const int64_t hs; // head size
@ -2549,8 +2491,6 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
test_cases.emplace_back(new test_arange()); test_cases.emplace_back(new test_arange());
test_cases.emplace_back(new test_timestep_embedding()); test_cases.emplace_back(new test_timestep_embedding());
test_cases.emplace_back(new test_leaky_relu()); test_cases.emplace_back(new test_leaky_relu());
test_cases.emplace_back(new test_ssm_conv());
test_cases.emplace_back(new test_ssm_scan());
for (int hs : { 64, 80, 128, 256, }) { for (int hs : { 64, 80, 128, 256, }) {
for (bool mask : { true, false } ) { for (bool mask : { true, false } ) {