CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

model : add BailingMoeV2 support (#16063)

Browse Source 
* add BailingMoeV2 support

* update llm types

* undo

* undo

* update llm types

* add model collection link

* update

* almost working

* correct group selection and rename n_group_exp

* avoid large top_k and use argmax instead for now

if we had something like argmax2 that would be equivalent, but this works fine until then

* poke

* skip group selection when there are no tokens

* fix 1T conversion

* hopefully fixed expert group selection

third time's the charm?

* make expert group selection generally available

The new LLaDA2Moe model uses this method too, make it generally available regardless of architecture.

* allow n_expert_groups to be 1 (Kimi K2)

* address review suggestions
This commit is contained in:
Sigbjørn Skjæret
2025-10-20 21:38:20 +02:00
committed by GitHub
parent c9c1972e2c
commit 84bf3c6778
15 changed files with 521 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

101
convert_hf_to_gguf.py
View File

@@ -892,8 +892,8 @@ class TextModel(ModelBase):
# ref: https://huggingface.co/JetBrains/Mellum-4b-base
res = "mellum"
if chkhsh == "9b1be57e70d20d9501b2b3186e792d81181ae36ada3903c26f9fea418cf87206":
# ref: https://huggingface.co/inclusionAI/LLaDA-MoE-7B-A1B-Base
res = "llada-moe"
# ref: https://huggingface.co/inclusionAI/Ling-mini-base-2.0
res = "bailingmoe2"
if chkhsh == "53e325976a6e142379c19b09afcae354f2f496f147afa8f9e189a33fe4e3024e":
# ref: https://huggingface.co/ibm-granite/granite-docling-258M
res = "granite-docling"
@@ -8055,6 +8055,103 @@ class BailingMoeModel(TextModel):
raise ValueError(f"Unprocessed experts: {experts}")
@ModelBase.register("BailingMoeV2ForCausalLM")
class BailingMoeV2Model(TextModel):
model_arch = gguf.MODEL_ARCH.BAILINGMOE2
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
if nextn_layers := self.hparams.get("num_nextn_predict_layers", 0):
self.block_count = self.hparams["num_hidden_layers"] + nextn_layers
self.tensor_map = gguf.get_tensor_name_map(self.model_arch, self.block_count)
def set_vocab(self):
self._set_vocab_gpt2()
def set_gguf_parameters(self):
super().set_gguf_parameters()
hparams = self.hparams
if (rope_dim := hparams.get("head_dim")) is None:
rope_dim = hparams["hidden_size"] // hparams["num_attention_heads"]
self.gguf_writer.add_rope_dimension_count(int(rope_dim * self.hparams.get("partial_rotary_factor", 0.5)))
rope_scaling = self.hparams.get("rope_scaling") or {}
if rope_scaling.get("rope_type", rope_scaling.get("type")) == "yarn" and "factor" in rope_scaling:
self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.YARN)
self.gguf_writer.add_rope_scaling_factor(rope_scaling["factor"])
self.gguf_writer.add_rope_scaling_orig_ctx_len(rope_scaling["original_max_position_embeddings"])
else:
self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.NONE)
self.gguf_writer.add_leading_dense_block_count(hparams["first_k_dense_replace"])
self.gguf_writer.add_vocab_size(hparams["vocab_size"])
self.gguf_writer.add_expert_feed_forward_length(hparams["moe_intermediate_size"])
self.gguf_writer.add_expert_shared_feed_forward_length(hparams.get("moe_shared_expert_intermediate_size", hparams["moe_intermediate_size"] * hparams["num_shared_experts"]))
self.gguf_writer.add_expert_weights_scale(hparams["routed_scaling_factor"])
self.gguf_writer.add_expert_count(hparams["num_experts"])
self.gguf_writer.add_expert_shared_count(hparams["num_shared_experts"])
self.gguf_writer.add_expert_group_count(hparams["n_group"])
self.gguf_writer.add_expert_group_used_count(hparams["topk_group"])
self.gguf_writer.add_expert_weights_norm(hparams["norm_topk_prob"])
if hparams["score_function"] == "sigmoid":
self.gguf_writer.add_expert_gating_func(gguf.ExpertGatingFuncType.SIGMOID)
elif hparams["score_function"] == "softmax":
self.gguf_writer.add_expert_gating_func(gguf.ExpertGatingFuncType.SOFTMAX)
else:
raise ValueError(f"Unsupported score_function value: {hparams['score_function']}")
if (nextn_layers := self.hparams.get("num_nextn_predict_layers")) is not None:
self.gguf_writer.add_nextn_predict_layers(nextn_layers)
_experts: list[dict[str, Tensor]] | None = None
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
if "mlp.experts" in name:
n_experts = self.hparams["num_experts"]
assert bid is not None
tensors: list[tuple[str, Tensor]] = []
if self._experts is None:
self._experts = [{} for _ in range(self.block_count)]
self._experts[bid][name] = data_torch
if len(self._experts[bid]) >= n_experts * 3:
# merge the experts into a single 3d tensor
for w_name in ["down_proj", "gate_proj", "up_proj"]:
datas: list[Tensor] = []
for xid in range(n_experts):
ename = f"model.layers.{bid}.mlp.experts.{xid}.{w_name}.weight"
datas.append(self._experts[bid][ename])
del self._experts[bid][ename]
data_torch = torch.stack(datas, dim=0)
merged_name = f"model.layers.{bid}.mlp.experts.{w_name}.weight"
new_name = self.map_tensor_name(merged_name)
tensors.append((new_name, data_torch))
return tensors
if name.endswith(".expert_bias"):
name = name.replace(".expert_bias", ".expert_bias.bias")
return [(self.map_tensor_name(name), data_torch)]
def prepare_tensors(self):
super().prepare_tensors()
if self._experts is not None:
# flatten `list[dict[str, Tensor]]` into `list[str]`
experts = [k for d in self._experts for k in d.keys()]
if len(experts) > 0:
raise ValueError(f"Unprocessed experts: {experts}")
@ModelBase.register("GroveMoeForCausalLM", "modeling_grove_moe.GroveMoeForCausalLM")
class GroveMoeModel(TextModel):
model_arch = gguf.MODEL_ARCH.GROVEMOE