vllm.model_executor.models.bailing_moe
Inference-only BailingMoE model compatible with HuggingFace weights.
BailingAttention
¶
Bases: Module
Source code in vllm/model_executor/models/bailing_moe.py
attn
instance-attribute
¶
attn = Attention(
num_heads,
head_dim,
scale,
num_kv_heads=num_kv_heads,
cache_config=cache_config,
prefix=f"{prefix}.attn",
)
dense
instance-attribute
¶
dense = RowParallelLinear(
total_num_heads * head_dim,
hidden_size,
bias=use_bias,
quant_config=quant_config,
prefix=f"{prefix}.dense",
)
query_key_value
instance-attribute
¶
query_key_value = QKVParallelLinear(
hidden_size,
head_dim,
total_num_heads,
total_kv_heads,
bias=use_bias or use_qkv_bias,
quant_config=quant_config,
prefix=f"{prefix}.query_key_value",
)
rotary_emb
instance-attribute
¶
rotary_emb = get_rope(
head_dim,
rotary_dim=head_dim,
max_position=max_position_embeddings,
base=rope_theta,
is_neox_style=True,
rope_scaling=rope_scaling,
)
__init__
¶
__init__(
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
)
Source code in vllm/model_executor/models/bailing_moe.py
forward
¶
Source code in vllm/model_executor/models/bailing_moe.py
BailingMLP
¶
Bases: Module
Source code in vllm/model_executor/models/bailing_moe.py
down_proj
instance-attribute
¶
down_proj = RowParallelLinear(
intermediate_size,
hidden_size,
bias=use_bias,
quant_config=quant_config,
reduce_results=reduce_results,
prefix=f"{prefix}.down_proj",
)
gate_up_proj
instance-attribute
¶
gate_up_proj = MergedColumnParallelLinear(
hidden_size,
[intermediate_size] * 2,
bias=use_bias,
quant_config=quant_config,
prefix=f"{prefix}.gate_up_proj",
)
__init__
¶
__init__(
intermediate_size: int,
config: PretrainedConfig,
quant_config: Optional[QuantizationConfig] = None,
reduce_results: Optional[bool] = True,
prefix: str = "",
) -> None
Source code in vllm/model_executor/models/bailing_moe.py
BailingMoE
¶
Bases: Module
Source code in vllm/model_executor/models/bailing_moe.py
experts
instance-attribute
¶
experts = FusedMoE(
num_experts=num_experts,
top_k=top_k,
hidden_size=hidden_size,
intermediate_size=moe_intermediate_size,
reduce_results=False,
renormalize=norm_expert_prob,
quant_config=quant_config,
prefix=f"{prefix}.experts",
)
gate
instance-attribute
¶
gate = ReplicatedLinear(
hidden_size, num_experts, bias=False, quant_config=None
)
shared_experts
instance-attribute
¶
shared_experts = BailingMLP(
intermediate_size=intermediate_size,
config=config,
quant_config=quant_config,
reduce_results=False,
prefix=f"{prefix}.shared_experts",
)
__init__
¶
__init__(
intermediate_size: int,
config: PretrainedConfig,
quant_config: Optional[QuantizationConfig] = None,
reduce_results: Optional[bool] = True,
prefix: str = "",
)
Source code in vllm/model_executor/models/bailing_moe.py
forward
¶
Source code in vllm/model_executor/models/bailing_moe.py
BailingMoeBlock
¶
Bases: Module
Source code in vllm/model_executor/models/bailing_moe.py
attention
instance-attribute
¶
attention = BailingAttention(
config,
cache_config,
quant_config,
prefix=f"{prefix}.attention",
)
mlp
instance-attribute
¶
mlp = BailingMoE(
intermediate_size,
config,
quant_config,
True,
prefix=f"{prefix}.mlp",
)
post_attention_layernorm
instance-attribute
¶
post_attention_layernorm = RMSNorm(
hidden_size, eps=rms_norm_eps
)
__init__
¶
__init__(
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
)
Source code in vllm/model_executor/models/bailing_moe.py
forward
¶
Source code in vllm/model_executor/models/bailing_moe.py
BailingMoeForCausalLM
¶
Bases: Module
, SupportsPP
Source code in vllm/model_executor/models/bailing_moe.py
lm_head
instance-attribute
¶
lm_head = (
word_embeddings
if tie_word_embeddings
else ParallelLMHead(
vocab_size, hidden_size, quant_config=quant_config
)
)
make_empty_intermediate_tensors
instance-attribute
¶
model
instance-attribute
¶
model = BailingMoeModel(
vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"),
)
packed_modules_mapping
class-attribute
instance-attribute
¶
packed_modules_mapping = {
"query_key_value": ["query_key_value"],
"gate_up_proj": ["gate_proj", "up_proj"],
}
__init__
¶
__init__(
*, vllm_config: VllmConfig, prefix: str = ""
) -> None
Source code in vllm/model_executor/models/bailing_moe.py
compute_logits
¶
compute_logits(
hidden_states: Tensor,
sampling_metadata: SamplingMetadata,
) -> Optional[Tensor]
Source code in vllm/model_executor/models/bailing_moe.py
forward
¶
forward(
input_ids: Tensor,
positions: Tensor,
intermediate_tensors: Optional[
IntermediateTensors
] = None,
inputs_embeds: Optional[Tensor] = None,
) -> Union[Tensor, IntermediateTensors]
Source code in vllm/model_executor/models/bailing_moe.py
get_input_embeddings
¶
load_weights
¶
Source code in vllm/model_executor/models/bailing_moe.py
BailingMoeModel
¶
Bases: Module
Source code in vllm/model_executor/models/bailing_moe.py
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 |
|
make_empty_intermediate_tensors
instance-attribute
¶
make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(
["hidden_states", "residual"], hidden_size
)
)
word_embeddings
instance-attribute
¶
word_embeddings = VocabParallelEmbedding(
vocab_size, embed_dim
)
__init__
¶
__init__(*, vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/bailing_moe.py
forward
¶
forward(
input_ids: Tensor,
position_ids: Tensor,
intermediate_tensors: Optional[IntermediateTensors],
inputs_embeds: Optional[Tensor] = None,
) -> Union[Tensor, IntermediateTensors]