Skip to content

Commit d2faf38

Browse files
gabe-l-hartgabe-l-hart
committed
feat: Add conversion for Bamba models
This is borrowed and adapted from the original implementation ggml-org#10810 Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
1 parent 673f1e3 commit d2faf38

File tree

4 files changed

+155
-10
lines changed

4 files changed

+155
-10
lines changed

‎convert_hf_to_gguf.py

Lines changed: 105 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -4688,6 +4688,9 @@ def __init__(self, dir_model: Path, *args, **kwargs):
46884688
with open(dir_model / "config.json", "r", encoding="utf-8") as f:
46894689
hparams = json.load(f)
46904690
super().__init__(dir_model, *args, hparams=hparams, **kwargs)
4691+
self.d_model = self.find_hparam(["hidden_size", "d_model", "dim"])
4692+
self.d_inner = self.find_hparam(["intermediate_size", "d_inner"], optional=True) or 2 * self.d_model
4693+
self.n_group = self.hparams.get("n_groups", 1)
46914694

46924695
def set_vocab(self):
46934696
vocab_size = self.hparams["vocab_size"]
@@ -4758,10 +4761,7 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
47584761
# (D is also unsqueezed, but for more straightforward broadcast internally)
47594762
data_torch = data_torch.reshape((*data_torch.shape, 1))
47604763
elif self.match_model_tensor_name(new_name, gguf.MODEL_TENSOR.SSM_NORM, bid):
4761-
d_model = self.find_hparam(["hidden_size", "d_model", "dim"])
4762-
d_inner = self.find_hparam(["intermediate_size", "d_inner"], optional=True) or 2 * d_model
4763-
n_group = self.hparams.get("n_groups", 1)
4764-
data_torch = data_torch.reshape((n_group, d_inner // n_group))
4764+
data_torch = data_torch.reshape((self.n_group, self.d_inner // self.n_group))
47654765

47664766
if name.endswith(".A_log"):
47674767
logger.debug("A_log --> A ==> " + new_name)
@@ -4770,6 +4770,107 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
47704770
yield (new_name, data_torch)
47714771

47724772

4773+
@ModelBase.register("BambaForCausalLM")
4774+
class BambaModel(Mamba2Model):
4775+
"""Bamba is a hybrid SSM + Attention model that uses Mamba2 SSM layers"""
4776+
model_arch = gguf.MODEL_ARCH.BAMBA
4777+
undo_permute = True
4778+
4779+
def __init__(self, *args, **kwargs):
4780+
4781+
# Hybrid mamba models use a prefix for the mamba-specific params.
4782+
# TODO: Extend this if the prefix(es) need to be configurable
4783+
self.hparam_prefixes = ["mamba"]
4784+
4785+
super().__init__(*args, **kwargs)
4786+
4787+
# Use Llama conversion for attention
4788+
self._transformer_model_class: type[TextModel] = LlamaModel
4789+
4790+
# Lists of which layers use ssm vs attention
4791+
self._attn_layers = self.hparams.get("attn_layer_indices", [])
4792+
if not self._attn_layers:
4793+
attn_period = self.hparams.get("attn_layer_period")
4794+
assert attn_period, "Didn't find attn_layer_indices or attn_layer_period"
4795+
attn_offset = self.hparams.get("attn_layer_offset")
4796+
assert attn_offset is not None, "No attention layer offset set with attn_layer_period"
4797+
self._attn_layers = [
4798+
i for i in range(self.block_count)
4799+
if i % attn_period == attn_offset
4800+
]
4801+
self._ssm_layers = [
4802+
i for i in range(self.block_count)
4803+
if i not in self._attn_layers
4804+
]
4805+
4806+
# n_group and d_inner are used during reshape_tensors for mamaba2
4807+
self.d_model = self.find_hparam(["hidden_size", "d_model"])
4808+
self.n_group = self.find_hparam(["n_groups"])
4809+
self.d_inner = self.find_hparam(["expand"]) * self.d_model
4810+
4811+
def find_hparam(self, keys: Iterable[str], *args, **kwargs) -> Any:
4812+
prefixed = []
4813+
for pfx in self.hparam_prefixes:
4814+
prefixed.extend(
4815+
"_".join([pfx, k])
4816+
for k in keys
4817+
)
4818+
keys = list(keys) + prefixed
4819+
return super().find_hparam(keys, *args, **kwargs)
4820+
4821+
def set_gguf_parameters(self):
4822+
4823+
## General Params ##
4824+
self.gguf_writer.add_embedding_length(self.d_model)
4825+
self.gguf_writer.add_block_count(self.block_count)
4826+
self.gguf_writer.add_context_length(self.hparams.get("max_position_embeddings", 0))
4827+
self.gguf_writer.add_vocab_size(self.hparams["vocab_size"])
4828+
self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"])
4829+
4830+
## Mamba mixer params ##
4831+
self.gguf_writer.add_ssm_conv_kernel(self.find_hparam(["conv_kernel", "d_conv"]))
4832+
self.gguf_writer.add_ssm_state_size(self.find_hparam(["state_size", "d_state"]))
4833+
self.gguf_writer.add_ssm_group_count(self.n_group)
4834+
self.gguf_writer.add_ssm_inner_size(self.d_inner)
4835+
# NOTE: The mamba_dt_rank is _not_ the right field for how this is used
4836+
# in llama.cpp
4837+
self.gguf_writer.add_ssm_time_step_rank(self.find_hparam(["n_heads"]))
4838+
4839+
## Attention params ##
4840+
self.gguf_writer.add_attn_layer_indices(self._attn_layers)
4841+
self.gguf_writer.add_rope_dimension_count(self.hparams["attn_rotary_emb"])
4842+
self.gguf_writer.add_head_count(self.hparams["num_attention_heads"])
4843+
self.gguf_writer.add_head_count_kv(self.find_hparam(["num_key_value_heads", "n_head_kv"]))
4844+
4845+
## Feed Forward Params ##
4846+
self.gguf_writer.add_layer_norm_rms_eps(
4847+
self.find_hparam(["layer_norm_epsilon", "rms_norm_eps"], optional=True) or 1e-5
4848+
)
4849+
4850+
## Validation ##
4851+
d_head = self.find_hparam(["d_head"], optional=True) or 64
4852+
assert self.hparams.get("hidden_act") in [None, "silu"], "Only SILU activation supported"
4853+
assert self.d_inner % d_head == 0, f"SSM inner size {self.d_inner} not a multiple of head dim {d_head}"
4854+
4855+
def modify_tensors(
4856+
self, data_torch: Tensor, name: str, bid: int | None
4857+
) -> Iterable[tuple[str, Tensor]]:
4858+
4859+
# Determine whether this is a mamaba layer or an attention layer
4860+
if bid in self._ssm_layers:
4861+
for mamba_new_name, data_torch in super().modify_tensors(
4862+
data_torch, name, bid
4863+
):
4864+
yield mamba_new_name, data_torch
4865+
elif bid in self._attn_layers:
4866+
for llama_new_name, data_torch in self._transformer_model_class.modify_tensors(
4867+
self, data_torch, name, bid
4868+
):
4869+
yield llama_new_name, data_torch
4870+
else:
4871+
yield self.map_tensor_name(name), data_torch
4872+
4873+
47734874
@ModelBase.register("CohereForCausalLM")
47744875
class CommandR2Model(TextModel):
47754876
model_arch = gguf.MODEL_ARCH.COMMAND_R

‎gguf-py/gguf/constants.py

Lines changed: 30 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -167,6 +167,9 @@ class SSM:
167167
GROUP_COUNT = "{arch}.ssm.group_count"
168168
DT_B_C_RMS = "{arch}.ssm.dt_b_c_rms"
169169

170+
class HybridAttention:
171+
ATTN_LAYER_INDICES = "{arch}.attention.layer_indices"
172+
170173
class WKV:
171174
HEAD_SIZE = "{arch}.wkv.head_size"
172175

@@ -320,6 +323,7 @@ class MODEL_ARCH(IntEnum):
320323
ARWKV7 = auto()
321324
MAMBA = auto()
322325
MAMBA2 = auto()
326+
BAMBA = auto()
323327
XVERSE = auto()
324328
COMMAND_R = auto()
325329
COHERE2 = auto()
@@ -602,6 +606,7 @@ class MODEL_TENSOR(IntEnum):
602606
MODEL_ARCH.ARWKV7: "arwkv7",
603607
MODEL_ARCH.MAMBA: "mamba",
604608
MODEL_ARCH.MAMBA2: "mamba2",
609+
MODEL_ARCH.BAMBA: "bamba",
605610
MODEL_ARCH.XVERSE: "xverse",
606611
MODEL_ARCH.COMMAND_R: "command-r",
607612
MODEL_ARCH.COHERE2: "cohere2",
@@ -1636,6 +1641,31 @@ class MODEL_TENSOR(IntEnum):
16361641
MODEL_TENSOR.SSM_NORM,
16371642
MODEL_TENSOR.SSM_OUT,
16381643
],
1644+
MODEL_ARCH.BAMBA: [
1645+
MODEL_TENSOR.TOKEN_EMBD,
1646+
MODEL_TENSOR.OUTPUT_NORM,
1647+
MODEL_TENSOR.OUTPUT,
1648+
MODEL_TENSOR.ATTN_NORM,
1649+
MODEL_TENSOR.SSM_IN,
1650+
MODEL_TENSOR.SSM_CONV1D,
1651+
MODEL_TENSOR.SSM_DT,
1652+
MODEL_TENSOR.SSM_A,
1653+
MODEL_TENSOR.SSM_D,
1654+
MODEL_TENSOR.SSM_NORM,
1655+
MODEL_TENSOR.SSM_OUT,
1656+
MODEL_TENSOR.ATTN_Q,
1657+
MODEL_TENSOR.ATTN_K,
1658+
MODEL_TENSOR.ATTN_V,
1659+
MODEL_TENSOR.ATTN_OUT,
1660+
MODEL_TENSOR.FFN_NORM,
1661+
MODEL_TENSOR.FFN_GATE,
1662+
MODEL_TENSOR.FFN_DOWN,
1663+
MODEL_TENSOR.FFN_UP,
1664+
MODEL_TENSOR.FFN_GATE_INP,
1665+
MODEL_TENSOR.FFN_GATE_EXP,
1666+
MODEL_TENSOR.FFN_DOWN_EXP,
1667+
MODEL_TENSOR.FFN_UP_EXP,
1668+
],
16391669
MODEL_ARCH.XVERSE: [
16401670
MODEL_TENSOR.TOKEN_EMBD,
16411671
MODEL_TENSOR.OUTPUT_NORM,

‎gguf-py/gguf/gguf_writer.py

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -849,6 +849,9 @@ def add_ssm_group_count(self, value: int) -> None:
849849
def add_ssm_dt_b_c_rms(self, value: bool) -> None:
850850
self.add_bool(Keys.SSM.DT_B_C_RMS.format(arch=self.arch), value)
851851

852+
def add_attn_layer_indices(self, values: list[int]) -> None:
853+
self.add_array(Keys.HybridAttention.ATTN_LAYER_INDICES.format(arch=self.arch), values)
854+
852855
def add_tokenizer_model(self, model: str) -> None:
853856
self.add_string(Keys.Tokenizer.MODEL, model)
854857

‎gguf-py/gguf/tensor_mapping.py

Lines changed: 17 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -13,7 +13,7 @@ class TensorNameMap:
1313
"transformer.wte", # gpt2 gpt-j mpt refact qwen dbrx jais exaone
1414
"transformer.word_embeddings", # falcon
1515
"word_embeddings", # bloom
16-
"model.embed_tokens", # llama-hf nemotron olmoe olmo2 rwkv6qwen2 glm4-0414
16+
"model.embed_tokens", # llama-hf nemotron olmoe olmo2 rwkv6qwen2 glm4-0414 bamba
1717
"tok_embeddings", # llama-pth
1818
"embeddings.word_embeddings", # bert nomic-bert
1919
"language_model.embedding.word_embeddings", # persimmon
@@ -117,7 +117,7 @@ class TensorNameMap:
117117
"transformer.h.{bid}.input_layernorm", # falcon7b
118118
"h.{bid}.input_layernorm", # bloom
119119
"transformer.h.{bid}.ln_mlp", # falcon40b
120-
"model.layers.{bid}.input_layernorm", # llama-hf nemotron olmoe phimoe
120+
"model.layers.{bid}.input_layernorm", # llama-hf nemotron olmoe phimoe bamba
121121
"layers.{bid}.attention_norm", # llama-pth
122122
"language_model.encoder.layers.{bid}.input_layernorm", # persimmon
123123
"model.layers.{bid}.ln1", # yi
@@ -275,7 +275,8 @@ class TensorNameMap:
275275
"transformer.decoder_layer.{bid}.rms_norm_2", # Grok
276276
"encoder.layers.{bid}.post_attention_layernorm", # chatglm
277277
"transformer.layers.{bid}.ffn_norm", # openelm
278-
"model.layers.{bid}.post_attention_layernorm", # llama4
278+
"language_model.model.layers.{bid}.post_attention_layernorm", # llama4
279+
"model.layers.{bid}.pre_ff_layernorm", # bamba
279280
),
280281

281282
# Post feed-forward norm
@@ -339,7 +340,8 @@ class TensorNameMap:
339340
"model.layers.{bid}.residual_mlp.w3", # arctic
340341
"encoder.layers.{bid}.mlp.dense_h_to_4h", # chatglm
341342
"transformer.h.{bid}.mlp.c_fc_1", # exaone
342-
"model.layers.{bid}.feed_forward.up_proj", # llama4
343+
"language_model.model.layers.{bid}.feed_forward.up_proj", # llama4
344+
"model.layers.{bid}.feed_forward.up_proj", # bamba
343345
),
344346

345347
MODEL_TENSOR.FFN_UP_EXP: (
@@ -376,7 +378,8 @@ class TensorNameMap:
376378
"transformer.h.{bid}.mlp.linear_1", # refact
377379
"model.layers.{bid}.residual_mlp.w1", # arctic
378380
"transformer.h.{bid}.mlp.c_fc_0", # exaone
379-
"model.layers.{bid}.feed_forward.gate_proj", # llama4
381+
"language_model.model.layers.{bid}.feed_forward.gate_proj", # llama4
382+
"model.layers.{bid}.feed_forward.gate_proj", # bamba
380383
),
381384

382385
MODEL_TENSOR.FFN_GATE_EXP: (
@@ -421,7 +424,8 @@ class TensorNameMap:
421424
"encoder.layer.{bid}.mlp.down_layer", # jina-bert-v2
422425
"encoder.layers.{bid}.mlp.dense_4h_to_h", # chatglm
423426
"model.layers.h.{bid}.mlp.c_proj", # exaone
424-
"model.layers.{bid}.feed_forward.down_proj", # llama4
427+
"language_model.model.layers.{bid}.feed_forward.down_proj", # llama4
428+
"model.layers.{bid}.feed_forward.down_proj", # bamba
425429
),
426430

427431
MODEL_TENSOR.FFN_DOWN_EXP: (
@@ -476,11 +480,13 @@ class TensorNameMap:
476480
MODEL_TENSOR.SSM_IN: (
477481
"model.layers.{bid}.in_proj",
478482
"backbone.layers.{bid}.mixer.in_proj",
483+
"model.layers.{bid}.mamba.in_proj", # bamba
479484
),
480485

481486
MODEL_TENSOR.SSM_CONV1D: (
482487
"model.layers.{bid}.conv1d",
483488
"backbone.layers.{bid}.mixer.conv1d",
489+
"model.layers.{bid}.mamba.conv1d", # bamba
484490
),
485491

486492
MODEL_TENSOR.SSM_X: (
@@ -491,25 +497,30 @@ class TensorNameMap:
491497
MODEL_TENSOR.SSM_DT: (
492498
"model.layers.{bid}.dt_proj",
493499
"backbone.layers.{bid}.mixer.dt_proj",
500+
"model.layers.{bid}.mamba.dt_proj", # bamba
494501
),
495502

496503
MODEL_TENSOR.SSM_A: (
497504
"model.layers.{bid}.A_log",
498505
"backbone.layers.{bid}.mixer.A_log",
506+
"model.layers.{bid}.mamba.A_log", # bamba
499507
),
500508

501509
MODEL_TENSOR.SSM_D: (
502510
"model.layers.{bid}.D",
503511
"backbone.layers.{bid}.mixer.D",
512+
"model.layers.{bid}.mamba.D", # bamba
504513
),
505514

506515
MODEL_TENSOR.SSM_NORM: (
507516
"backbone.layers.{bid}.mixer.norm", # mamba2
517+
"model.layers.{bid}.mamba.norm", # bamba
508518
),
509519

510520
MODEL_TENSOR.SSM_OUT: (
511521
"model.layers.{bid}.out_proj",
512522
"backbone.layers.{bid}.mixer.out_proj",
523+
"model.layers.{bid}.mamba.out_proj", # bamba
513524
),
514525

515526
MODEL_TENSOR.TIME_MIX_W0: (

0 commit comments

Comments
 (0)