CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : initial Mamba-2 support (#9126)

Browse Source 
* 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
Download Patch File Download Diff File Expand all files Collapse all files

20
src/llama-arch.cpp
View File

@@ -45,6 +45,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
{ LLM_ARCH_GEMMA3N, "gemma3n" },
{ LLM_ARCH_STARCODER2, "starcoder2" },
{ LLM_ARCH_MAMBA, "mamba" },
{ LLM_ARCH_MAMBA2, "mamba2" },
{ LLM_ARCH_XVERSE, "xverse" },
{ LLM_ARCH_COMMAND_R, "command-r" },
{ LLM_ARCH_COHERE2, "cohere2" },
@@ -170,6 +171,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
{ LLM_KV_SSM_INNER_SIZE, "%s.ssm.inner_size" },
{ LLM_KV_SSM_STATE_SIZE, "%s.ssm.state_size" },
{ LLM_KV_SSM_TIME_STEP_RANK, "%s.ssm.time_step_rank" },
{ LLM_KV_SSM_GROUP_COUNT, "%s.ssm.group_count" },
{ LLM_KV_SSM_DT_B_C_RMS, "%s.ssm.dt_b_c_rms" },
{ LLM_KV_WKV_HEAD_SIZE, "%s.wkv.head_size" },
@@ -1004,6 +1006,22 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
{ LLM_TENSOR_SSM_OUT, "blk.%d.ssm_out" },
},
},
{
LLM_ARCH_MAMBA2,
{
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" },
{ LLM_TENSOR_OUTPUT_NORM, "output_norm" },
{ LLM_TENSOR_OUTPUT, "output" },
{ LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" },
{ LLM_TENSOR_SSM_IN, "blk.%d.ssm_in" },
{ LLM_TENSOR_SSM_CONV1D, "blk.%d.ssm_conv1d" },
{ LLM_TENSOR_SSM_DT, "blk.%d.ssm_dt" },
{ LLM_TENSOR_SSM_A, "blk.%d.ssm_a" },
{ LLM_TENSOR_SSM_D, "blk.%d.ssm_d" },
{ LLM_TENSOR_SSM_NORM, "blk.%d.ssm_norm" },
{ LLM_TENSOR_SSM_OUT, "blk.%d.ssm_out" },
},
},
{
LLM_ARCH_XVERSE,
{
@@ -1761,6 +1779,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
{LLM_TENSOR_SSM_CONV1D, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_CONV}},
{LLM_TENSOR_SSM_A, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_SCAN}},
{LLM_TENSOR_SSM_D, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_SSM_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_TIME_MIX_LERP_X, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_TIME_MIX_LN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_CHANNEL_MIX_LERP_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
@@ -1894,6 +1913,7 @@ const llm_tensor_info & llm_tensor_info_for(llm_tensor tensor) {
bool llm_arch_is_recurrent(const llm_arch & arch) {
switch (arch) {
case LLM_ARCH_MAMBA:
case LLM_ARCH_MAMBA2:
case LLM_ARCH_RWKV6:
case LLM_ARCH_RWKV6QWEN2:
case LLM_ARCH_RWKV7: