llama : initial Mamba-2 support (ggml-org#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

Author: compilade

convert_hf_to_gguf.py

@@ -4781,6 +4781,14 @@ def set_gguf_parameters(self):
 class MambaModel(TextModel):
     model_arch = gguf.MODEL_ARCH.MAMBA
 
+    def __init__(self, dir_model: Path, *args, **kwargs):
+        # Avoid using AutoConfig for hparams
+        hparams = kwargs.pop("hparams", None)
+        if hparams is None:
+            with open(dir_model / "config.json", "r", encoding="utf-8") as f:
+                hparams = json.load(f)
+        super().__init__(dir_model, *args, hparams=hparams, **kwargs)
+
     def set_vocab(self):
         vocab_size = self.hparams["vocab_size"]
         # Round vocab size to next multiple of 8
@@ -4855,6 +4863,100 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
         return [(new_name, data_torch)]
 
 
+@ModelBase.register("Mamba2ForCausalLM")
+class Mamba2Model(TextModel):
+    model_arch = gguf.MODEL_ARCH.MAMBA2
+
+    def __init__(self, dir_model: Path, *args, **kwargs):
+        # Avoid using AutoConfig for hparams
+        # It wrongly assumes all Mamba2 models are Mamba-Codestral-7B-v0.1
+        hparams = kwargs.pop("hparams", None)
+        if hparams is None:
+            with open(dir_model / "config.json", "r", encoding="utf-8") as f:
+                hparams = json.load(f)
+        super().__init__(dir_model, *args, hparams=hparams, **kwargs)
+
+    def set_vocab(self):
+        vocab_size = self.hparams["vocab_size"]
+        # Round vocab size to next multiple of 16
+        pad_vocab = self.hparams.get("pad_vocab_size_multiple", 16)
+        # pad using ceiling division
+        # ref: https://stackoverflow.com/a/17511341/22827863
+        vocab_size = -(vocab_size // -pad_vocab) * pad_vocab
+        self.hparams["vocab_size"] = vocab_size
+
+        if (self.dir_model / "tokenizer.model").is_file():
+            self._set_vocab_sentencepiece()
+        elif (self.dir_model / "tokenizer.model.v3").is_file():
+            # mamba-codestral
+            raise NotImplementedError(f"Please rename {self.dir_model / 'tokenizer.model.v3'} to {self.dir_model / 'tokenizer.model'}")
+        elif (self.dir_model / "tokenizer.json").is_file():
+            self._set_vocab_gpt2()
+        else:
+            # Use the GPT-NeoX tokenizer when no tokenizer files are present
+            self._set_vocab_builtin("gpt-neox", vocab_size)
+
+    def set_gguf_parameters(self):
+        d_model = self.find_hparam(["hidden_size", "d_model", "dim"])
+        d_conv  = self.find_hparam(["conv_kernel",       "d_conv"],  optional=True) or 4
+        d_inner = self.find_hparam(["intermediate_size", "d_inner"], optional=True) or 2 * d_model
+        d_state = self.find_hparam(["state_size",        "d_state"], optional=True) or 128
+        head_dim = self.find_hparam(["head_dim"],                    optional=True) or 64
+        n_group = self.find_hparam(["n_groups"],                     optional=True) or 1
+
+        rms_norm_eps = self.find_hparam(["layer_norm_epsilon", "rms_norm_eps"], optional=True) or 1e-5
+
+        # Fail early for models which don't have a block expansion factor of 2
+        # TODO: does this really matter?
+        assert d_inner == 2 * d_model
+        assert d_inner % head_dim == 0
+
+        self.gguf_writer.add_context_length(2**20)  # arbitrary value; for those who use the default
+        self.gguf_writer.add_embedding_length(d_model)
+        self.gguf_writer.add_feed_forward_length(0)  # unused, but seemingly required when loading
+        self.gguf_writer.add_head_count(0)  # unused, but seemingly required when loading
+        self.gguf_writer.add_block_count(self.block_count)
+        self.gguf_writer.add_ssm_conv_kernel(d_conv)
+        self.gguf_writer.add_ssm_inner_size(d_inner)
+        self.gguf_writer.add_ssm_state_size(d_state)
+        self.gguf_writer.add_ssm_time_step_rank(d_inner // head_dim)
+        self.gguf_writer.add_ssm_group_count(n_group)
+        self.gguf_writer.add_layer_norm_rms_eps(rms_norm_eps)
+        self.gguf_writer.add_file_type(self.ftype)
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+
+        if name.startswith("model.backbone") or name.startswith("model.lm_head"):
+            # map Mamba-Codestral-7B-v0.1 tensor names to the names used by Mamba-2
+            name = name.removeprefix("model.")
+
+        if name.endswith(".dt_bias"):
+            name = name.rpartition(".dt_bias")[0] + ".dt_proj.bias"
+
+        new_name = self.map_tensor_name(name)
+
+        if self.match_model_tensor_name(new_name, gguf.MODEL_TENSOR.SSM_CONV1D, bid):
+            data_torch = data_torch.squeeze()
+        elif any(self.match_model_tensor_name(new_name, t, bid, suffix="") for t in [
+            gguf.MODEL_TENSOR.SSM_A,
+            gguf.MODEL_TENSOR.SSM_D,
+        ]):
+            # unsqueeze A to use similar shape semantics as Mamba-1
+            # (D is also unsqueezed, but for more straightforward broadcast internally)
+            data_torch = data_torch.reshape((*data_torch.shape, 1))
+        elif self.match_model_tensor_name(new_name, gguf.MODEL_TENSOR.SSM_NORM, bid):
+            d_model = self.find_hparam(["hidden_size", "d_model", "dim"])
+            d_inner = self.find_hparam(["intermediate_size", "d_inner"], optional=True) or 2 * d_model
+            n_group = self.hparams.get("n_groups", 1)
+            data_torch = data_torch.reshape((n_group, d_inner // n_group))
+
+        if name.endswith(".A_log"):
+            logger.debug("A_log --> A ==> " + new_name)
+            data_torch = -torch.exp(data_torch)
+
+        yield (new_name, data_torch)
+
+
 @ModelBase.register("CohereForCausalLM")
 class CommandR2Model(TextModel):
     model_arch = gguf.MODEL_ARCH.COMMAND_R
@@ -6615,12 +6717,20 @@ def get_model_architecture(hparams: dict[str, Any], model_type: ModelType) -> st
     # maybe we should fallback to text model's arch in that case, since not many models have both
     text_config = hparams.get("text_config", {})
     vision_config = hparams.get("vision_config", {})
-    arch = hparams["architectures"][0]
+    arch = None
+    if (arches := hparams.get("architectures")) is not None and len(arches) > 0:
+        arch = arches[0]
+    elif "ssm_cfg" in hparams:
+        # For non-hf Mamba and Mamba2 models
+        arch = hparams["ssm_cfg"].get("layer", "Mamba") + "ForCausalLM"
+
     # if "architectures" is found in the sub-config, use that instead
     if model_type == ModelType.TEXT and text_config.get("architectures") is not None:
         arch = text_config["architectures"][0]
     elif model_type == ModelType.MMPROJ and vision_config.get("architectures") is not None:
         arch = vision_config["architectures"][0]
+    if arch is None:
+        raise ValueError("Failed to detect model architecture")
     return arch
 
 

ggml/include/ggml.h

@@ -2031,7 +2031,8 @@ extern "C" {
             struct ggml_tensor  * dt,
             struct ggml_tensor  * A,
             struct ggml_tensor  * B,
-            struct ggml_tensor  * C);
+            struct ggml_tensor  * C,
+            struct ggml_tensor  * ids);
 
     // partition into non-overlapping windows with padding if needed
     // example: