Convert formatting to use ruff instead of yapf + isort (#26247)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
@@ -22,6 +22,7 @@
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
"""Inference-only GLM-4-0414 model compatible with HuggingFace weights."""
|
||||
|
||||
from collections.abc import Iterable
|
||||
from typing import Optional, Union
|
||||
|
||||
@@ -34,8 +35,7 @@ from vllm.compilation.decorators import support_torch_compile
|
||||
from vllm.config import CacheConfig, VllmConfig
|
||||
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
|
||||
from vllm.model_executor.layers.layernorm import RMSNorm
|
||||
from vllm.model_executor.layers.linear import (QKVParallelLinear,
|
||||
RowParallelLinear)
|
||||
from vllm.model_executor.layers.linear import QKVParallelLinear, RowParallelLinear
|
||||
from vllm.model_executor.layers.logits_processor import LogitsProcessor
|
||||
from vllm.model_executor.layers.quantization import QuantizationConfig
|
||||
from vllm.model_executor.layers.rotary_embedding import get_rope
|
||||
@@ -49,21 +49,22 @@ from .utils import AutoWeightsLoader, PPMissingLayer, maybe_prefix
|
||||
|
||||
|
||||
class Glm4Attention(nn.Module):
|
||||
|
||||
def __init__(self,
|
||||
config: Glm4Config,
|
||||
hidden_size: int,
|
||||
num_heads: int,
|
||||
num_kv_heads: int,
|
||||
max_position: int = 4096 * 32,
|
||||
head_dim: Optional[int] = None,
|
||||
qkv_bias: bool = False,
|
||||
rope_theta: float = 10000,
|
||||
cache_config: Optional[CacheConfig] = None,
|
||||
quant_config: Optional[QuantizationConfig] = None,
|
||||
rope_scaling: Optional[tuple] = None,
|
||||
prefix: str = "",
|
||||
attn_type: str = AttentionType.DECODER) -> None:
|
||||
def __init__(
|
||||
self,
|
||||
config: Glm4Config,
|
||||
hidden_size: int,
|
||||
num_heads: int,
|
||||
num_kv_heads: int,
|
||||
max_position: int = 4096 * 32,
|
||||
head_dim: Optional[int] = None,
|
||||
qkv_bias: bool = False,
|
||||
rope_theta: float = 10000,
|
||||
cache_config: Optional[CacheConfig] = None,
|
||||
quant_config: Optional[QuantizationConfig] = None,
|
||||
rope_scaling: Optional[tuple] = None,
|
||||
prefix: str = "",
|
||||
attn_type: str = AttentionType.DECODER,
|
||||
) -> None:
|
||||
super().__init__()
|
||||
self.hidden_size = hidden_size
|
||||
tp_size = get_tensor_model_parallel_world_size()
|
||||
@@ -112,14 +113,16 @@ class Glm4Attention(nn.Module):
|
||||
partial_rotary_factor=partial_rotary_factor,
|
||||
is_neox_style=False,
|
||||
)
|
||||
self.attn = Attention(self.num_heads,
|
||||
self.head_dim,
|
||||
self.scaling,
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
prefix=f"{prefix}.attn",
|
||||
attn_type=attn_type)
|
||||
self.attn = Attention(
|
||||
self.num_heads,
|
||||
self.head_dim,
|
||||
self.scaling,
|
||||
num_kv_heads=self.num_kv_heads,
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
prefix=f"{prefix}.attn",
|
||||
attn_type=attn_type,
|
||||
)
|
||||
|
||||
def forward(
|
||||
self,
|
||||
@@ -135,11 +138,12 @@ class Glm4Attention(nn.Module):
|
||||
|
||||
|
||||
class Glm4DecoderLayer(nn.Module):
|
||||
|
||||
def __init__(self,
|
||||
vllm_config: VllmConfig,
|
||||
prefix: str = "",
|
||||
config: Optional[Glm4Config] = None) -> None:
|
||||
def __init__(
|
||||
self,
|
||||
vllm_config: VllmConfig,
|
||||
prefix: str = "",
|
||||
config: Optional[Glm4Config] = None,
|
||||
) -> None:
|
||||
super().__init__()
|
||||
|
||||
config = config or vllm_config.model_config.hf_config
|
||||
@@ -157,8 +161,8 @@ class Glm4DecoderLayer(nn.Module):
|
||||
max_position=config.max_position_embeddings,
|
||||
num_kv_heads=config.num_key_value_heads,
|
||||
rope_theta=rope_theta,
|
||||
qkv_bias=getattr(config, 'attention_bias', False),
|
||||
head_dim=getattr(config, 'head_dim', None),
|
||||
qkv_bias=getattr(config, "attention_bias", False),
|
||||
head_dim=getattr(config, "head_dim", None),
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
rope_scaling=rope_scaling,
|
||||
@@ -172,14 +176,14 @@ class Glm4DecoderLayer(nn.Module):
|
||||
quant_config=quant_config,
|
||||
prefix=f"{prefix}.mlp",
|
||||
)
|
||||
self.input_layernorm = RMSNorm(config.hidden_size,
|
||||
eps=config.rms_norm_eps)
|
||||
self.post_attention_layernorm = RMSNorm(config.hidden_size,
|
||||
eps=config.rms_norm_eps)
|
||||
self.post_self_attn_layernorm = RMSNorm(config.hidden_size,
|
||||
eps=config.rms_norm_eps)
|
||||
self.post_mlp_layernorm = RMSNorm(config.hidden_size,
|
||||
eps=config.rms_norm_eps)
|
||||
self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
||||
self.post_attention_layernorm = RMSNorm(
|
||||
config.hidden_size, eps=config.rms_norm_eps
|
||||
)
|
||||
self.post_self_attn_layernorm = RMSNorm(
|
||||
config.hidden_size, eps=config.rms_norm_eps
|
||||
)
|
||||
self.post_mlp_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
||||
|
||||
def forward(
|
||||
self,
|
||||
@@ -192,8 +196,7 @@ class Glm4DecoderLayer(nn.Module):
|
||||
residual = hidden_states
|
||||
hidden_states = self.input_layernorm(hidden_states)
|
||||
else:
|
||||
hidden_states, residual = self.input_layernorm(
|
||||
hidden_states, residual)
|
||||
hidden_states, residual = self.input_layernorm(hidden_states, residual)
|
||||
hidden_states = self.self_attn(
|
||||
positions=positions,
|
||||
hidden_states=hidden_states,
|
||||
@@ -202,8 +205,7 @@ class Glm4DecoderLayer(nn.Module):
|
||||
hidden_states = self.post_self_attn_layernorm(hidden_states)
|
||||
|
||||
# Fully Connected
|
||||
hidden_states, residual = self.post_attention_layernorm(
|
||||
hidden_states, residual)
|
||||
hidden_states, residual = self.post_attention_layernorm(hidden_states, residual)
|
||||
hidden_states = self.mlp(hidden_states)
|
||||
hidden_states = self.post_mlp_layernorm(hidden_states)
|
||||
|
||||
@@ -221,13 +223,13 @@ ALL_DECODER_LAYER_TYPES = {
|
||||
"positions": -1,
|
||||
"intermediate_tensors": 0,
|
||||
"inputs_embeds": 0,
|
||||
})
|
||||
}
|
||||
)
|
||||
class Glm4Model(LlamaModel):
|
||||
|
||||
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
|
||||
super().__init__(vllm_config=vllm_config,
|
||||
prefix=prefix,
|
||||
layer_type=Glm4DecoderLayer)
|
||||
super().__init__(
|
||||
vllm_config=vllm_config, prefix=prefix, layer_type=Glm4DecoderLayer
|
||||
)
|
||||
|
||||
|
||||
class Glm4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
@@ -253,25 +255,28 @@ class Glm4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
self.lora_config = lora_config
|
||||
|
||||
self.quant_config = quant_config
|
||||
self.model = Glm4Model(vllm_config=vllm_config,
|
||||
prefix=maybe_prefix(prefix, "model"))
|
||||
self.model = Glm4Model(
|
||||
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
|
||||
)
|
||||
|
||||
if get_pp_group().is_last_rank:
|
||||
if config.tie_word_embeddings:
|
||||
self.lm_head = self.model.embed_tokens
|
||||
else:
|
||||
self.lm_head = ParallelLMHead(config.vocab_size,
|
||||
config.hidden_size,
|
||||
quant_config=quant_config,
|
||||
prefix=maybe_prefix(
|
||||
prefix, "lm_head"))
|
||||
self.lm_head = ParallelLMHead(
|
||||
config.vocab_size,
|
||||
config.hidden_size,
|
||||
quant_config=quant_config,
|
||||
prefix=maybe_prefix(prefix, "lm_head"),
|
||||
)
|
||||
else:
|
||||
self.lm_head = PPMissingLayer()
|
||||
|
||||
self.logits_processor = LogitsProcessor(config.vocab_size)
|
||||
|
||||
self.make_empty_intermediate_tensors = (
|
||||
self.model.make_empty_intermediate_tensors)
|
||||
self.model.make_empty_intermediate_tensors
|
||||
)
|
||||
|
||||
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
|
||||
return self.model.get_input_embeddings(input_ids)
|
||||
@@ -283,8 +288,9 @@ class Glm4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
intermediate_tensors: Optional[IntermediateTensors] = None,
|
||||
inputs_embeds: Optional[torch.Tensor] = None,
|
||||
) -> Union[torch.Tensor, IntermediateTensors]:
|
||||
hidden_states = self.model(input_ids, positions, intermediate_tensors,
|
||||
inputs_embeds)
|
||||
hidden_states = self.model(
|
||||
input_ids, positions, intermediate_tensors, inputs_embeds
|
||||
)
|
||||
return hidden_states
|
||||
|
||||
def compute_logits(
|
||||
@@ -294,11 +300,9 @@ class Glm4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
logits = self.logits_processor(self.lm_head, hidden_states)
|
||||
return logits
|
||||
|
||||
def load_weights(self, weights: Iterable[tuple[str,
|
||||
torch.Tensor]]) -> set[str]:
|
||||
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
|
||||
loader = AutoWeightsLoader(
|
||||
self,
|
||||
skip_prefixes=(["lm_head."]
|
||||
if self.config.tie_word_embeddings else None),
|
||||
skip_prefixes=(["lm_head."] if self.config.tie_word_embeddings else None),
|
||||
)
|
||||
return loader.load_weights(weights)
|
||||
|
||||
Reference in New Issue
Block a user