llama : initial Mamba-2 support (#9126)

* llama : initial Mamba-2 support

* ggml : SIMD ggml_ssm_scan for Mamba-2

* ggml : improve ggml_mul speed when masking recurrent states

* llama : support running Mamba-Codestral-7B-v0.1

* llama : fix Mamba-2 conv state saving

* ggml : make the ggml_mul fast broadcast path more consistently formatted

* llama : remove unused variable

* llama : add missing break

* convert_hf : prefer SentencePiece tokenizer for Mamba-2 when present

The tokenzier.json of Mamba-Codestral-7B-v0.1 otherwise requires
workarounds to work correctly.

* llama : avoid redundant state copy for Mamba 1 and 2

* metal : attempt to adapt SSM_SCAN for Mamba-2

* metal : fix SSM_SCAN pipeline scope

* metal : use log and exp instead of log1pf and expf in SSM_SCAN

* metal : remove unused arguments for SSM_SCAN

The max index is 31, so trimming the arguments is necessary.

* metal : add back n_seqs to SSM_SCAN args

Whoops, this is needed for the offset in the concatenated output.

* metal : fix SSM_SCAN state head offset

* metal : fix wrong number of tokens per sequence in SSM_SCAN

* ggml : remove unused fast broadcast path in GGML_MUL

This was initially added because states were masked with ggml_mul,
but this is no longer done and so this "optimisation" is no longer
necessary, or at least not worth the additional code complexity.

* ggml : avoid multiply by D in GGML_OP_SSM_SCAN

This makes the weight buft detection in src/llama.cpp simpler.

* convert : transpose Mamba-2 A, D and reshape SSM_NORM

This breaks existing conversions of Mamba-2 models
to avoid some reshapes.

Not sure if it's a good idea,
but it makes the graph slightly cleaner.

* llama : more appropriate SSM_SCAN and SSM_CONV buft support checks

* convert : fix flake8 lint

* metal : fix confusion between ; and ,

* metal : add missing args for nb references in ssm_scan_f32_group

* metal : single-user mamba2 inference works

* kv-cache : remove const_cast when setting inputs for s_copy

And also fix multi-user inference for recurrent models
by using cell_id instead of i as the kv cell index
when populating s_copy.

* convert : avoid AutoConfig for Mamba and Mamba2 hparams

* kv-cache : allow context shift for recurrent models

* graph : fix recurrent state copies when avoiding copies

Works, but using lambda functions might not be that clean.

* ggml : fix mamba2 ssm scan when compiled with SVE

* ggml-cpu : reorder SVE FMA for consistency with other SIMD arches

* cuda : implement ssm scan for Mamba2

There is still room for improvement, but it works!

* cuda : adapt Mamba1 ssm scan to shape changes from Mamba2

* mamba : fix mismatched new and delete size for llm_build_mamba

Subclasses of llm_graph_context cannot have extra fields,
because the called destructor is not the one from the subclass.
This otherwise would cause problems when runnning Mamba-(1|2) inference
when compiled -DGGML_SANITIZE_ADDRESS=ON

* cuda : graceful fallback for Mamba-1 models with weird embd size
This commit is contained in:
compilade
2025-07-02 13:10:24 -04:00
committed by GitHub
parent e17991c466
commit 5d46babdc2
24 changed files with 1075 additions and 311 deletions

View File

@ -2084,28 +2084,58 @@ struct test_ssm_scan : public test_case {
const ggml_type type;
const int64_t d_state;
const int64_t d_inner;
const int64_t head_dim;
const int64_t n_head;
const int64_t n_group;
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);
return VARS_TO_STR7(type, d_state, head_dim, n_head, n_group, n_seq_tokens, n_seqs);
}
test_ssm_scan(ggml_type type = GGML_TYPE_F32,
int64_t d_state = 32, int64_t d_inner = 32, int64_t n_seq_tokens = 32, int64_t n_seqs = 32)
: type(type), d_state(d_state), d_inner(d_inner), n_seq_tokens(n_seq_tokens), n_seqs(n_seqs) {}
int64_t d_state = 32,
int64_t head_dim = 1, // non-zero for Mamba-2
int64_t n_head = 32,
int64_t n_group = 1,
int64_t n_seq_tokens = 32,
int64_t n_seqs = 32)
: type(type), d_state(d_state), head_dim(head_dim), n_head(n_head), n_group(n_group), n_seq_tokens(n_seq_tokens), n_seqs(n_seqs) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * s = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, d_inner, n_seqs, 1 }.data());
ggml_tensor * x = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_inner, n_seq_tokens, n_seqs, 1 }.data());
ggml_tensor * dt = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_inner, n_seq_tokens, n_seqs, 1 }.data());
ggml_tensor * A = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, d_inner, 1 , 1 }.data());
ggml_tensor * B = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, n_seq_tokens, n_seqs, 1 }.data());
ggml_tensor * C = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, n_seq_tokens, n_seqs, 1 }.data());
ggml_tensor * out = ggml_ssm_scan(ctx, s, x, dt, A, B, C);
ggml_tensor * s = ggml_new_tensor_4d(ctx, type, d_state, head_dim, n_head, n_seqs);
ggml_tensor * x = ggml_new_tensor_4d(ctx, type, head_dim, n_head, n_seq_tokens, n_seqs);
ggml_tensor * dt = ggml_new_tensor_3d(ctx, type, n_head, n_seq_tokens, n_seqs);
ggml_tensor * A = ggml_new_tensor_2d(ctx, type, (head_dim > 1) ? 1 : d_state, n_head);
ggml_tensor * B = ggml_new_tensor_4d(ctx, type, d_state, n_group, n_seq_tokens, n_seqs);
ggml_tensor * C = ggml_new_tensor_4d(ctx, type, d_state, n_group, n_seq_tokens, n_seqs);
ggml_tensor * ids = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, n_seqs);
ggml_tensor * out = ggml_ssm_scan(ctx, s, x, dt, A, B, C, ids);
return out;
}
// similar to test_mul_mat_id
void initialize_tensors(ggml_context * ctx) override {
std::random_device rd;
std::default_random_engine rng(rd());
for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != NULL; t = ggml_get_next_tensor(ctx, t)) {
if (t->type == GGML_TYPE_I32) {
if (ggml_is_view_op(t->op)) { continue; }
// ids
for (int64_t r = 0; r < ggml_nrows(t); r++) {
std::vector<int32_t> data(t->ne[0]);
for (int i = 0; i < t->ne[0]; i++) {
data[i] = i;
}
std::shuffle(data.begin(), data.end(), rng);
ggml_backend_tensor_set(t, data.data(), r * t->nb[1], t->ne[0] * sizeof(int32_t));
}
} else {
init_tensor_uniform(t);
}
}
}
};
// GGML_OP_RWKV_WKV6
@ -4498,7 +4528,8 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_ssm_conv(GGML_TYPE_F32, {8, 1536, 1, 1}, {4, 1536, 1, 1}));
test_cases.emplace_back(new test_ssm_conv(GGML_TYPE_F32, {4, 1536, 4, 1}, {4, 1536, 1, 1}));
test_cases.emplace_back(new test_ssm_scan(GGML_TYPE_F32, 16, 1024, 32, 4));
test_cases.emplace_back(new test_ssm_scan(GGML_TYPE_F32, 16, 1, 1024, 1, 32, 4)); // Mamba-1
test_cases.emplace_back(new test_ssm_scan(GGML_TYPE_F32, 128, 64, 16, 2, 32, 4)); // Mamba-2
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));