Spaces:
Running on Zero
Running on Zero
File size: 9,348 Bytes
22510af | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 | from __future__ import annotations
from typing import Callable, Iterable, TYPE_CHECKING
import torch
if TYPE_CHECKING:
from torch import Tensor
from .base import ModelBase, TextModel, gguf
@ModelBase.register("BailingMoeForCausalLM")
class BailingMoeModel(TextModel):
model_arch = gguf.MODEL_ARCH.BAILINGMOE
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(rope_dim)
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_weights_scale(1.0)
self.gguf_writer.add_expert_shared_count(hparams["num_shared_experts"])
self.gguf_writer.add_expert_weights_norm(hparams["norm_topk_prob"])
_experts: list[dict[str, Tensor]] | None = None
@staticmethod
def permute(weights: Tensor, n_head: int, n_head_kv: int | None):
if n_head_kv is not None and n_head != n_head_kv:
n_head = n_head_kv
return (weights.reshape(n_head, 2, weights.shape[0] // n_head // 2, *weights.shape[1:])
.swapaxes(1, 2)
.reshape(weights.shape))
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
n_head = self.hparams["num_attention_heads"]
n_kv_head = self.hparams.get("num_key_value_heads")
n_embd = self.hparams["hidden_size"]
if (head_dim := self.hparams.get("head_dim")) is None:
head_dim = n_embd // n_head
output_name = self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT)
if name.endswith("attention.dense.weight"):
yield from super().modify_tensors(data_torch, self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_OUT, bid), bid)
return
elif name.endswith("query_key_value.weight"):
q, k, v = data_torch.split([n_head * head_dim, n_kv_head * head_dim, n_kv_head * head_dim], dim=-2)
yield from super().modify_tensors(BailingMoeModel.permute(q, n_head, n_head), self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_Q, bid), bid)
yield from super().modify_tensors(BailingMoeModel.permute(k, n_head, n_kv_head), self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_K, bid), bid)
yield from super().modify_tensors(v,self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_V, bid), bid)
return
elif name.find("mlp.experts") != -1:
n_experts = self.find_hparam(["num_local_experts", "num_experts"])
assert bid is not None
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)
yield from super().modify_tensors(data_torch, new_name, bid)
return
new_name = self.map_tensor_name(name)
if new_name == output_name and self.hparams.get("norm_head"):
data_torch = data_torch.float()
data_torch /= torch.norm(data_torch, p=2, dim=0, keepdim=True) + 1e-7
yield from super().modify_tensors(data_torch, new_name, bid)
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("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.rope_parameters.get("partial_rotary_factor", 0.5)))
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_shared_count(hparams["num_shared_experts"])
self.gguf_writer.add_expert_weights_norm(hparams["norm_topk_prob"])
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
@classmethod
def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
name, gen = item
if name.endswith(".expert_bias"):
name = name.replace(".expert_bias", ".expert_bias.bias")
return super().filter_tensors((name, gen))
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
if "mlp.experts" in name:
n_experts = self.find_hparam(["num_local_experts", "num_experts"])
assert bid is not None
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"
yield from super().modify_tensors(data_torch, merged_name, bid)
return
yield from super().modify_tensors(data_torch, name, bid)
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("SarvamMoEForCausalLM", "modeling_sarvam_moe.SarvamMoEForCausalLM")
class SarvamMoEModel(BailingMoeV2Model):
model_arch = gguf.MODEL_ARCH.BAILINGMOE2
# Sarvam-MoE shares the BailingMoeV2 architecture; only differences:
# - full rotary (no partial_rotary_factor)
# - expert bias is zero-mean normalized at load time
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"]
# Override the partial-rotary value written by BailingMoeV2 with the full rotary dim
self.gguf_writer.add_rope_dimension_count(rope_dim)
@classmethod
def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
name, gen = item
if name.endswith(".expert_bias"):
# Sarvam normalizes expert bias to zero mean
inner = gen
def gen():
t = inner()
return t - t.mean()
return super().filter_tensors((name, gen))
|