[Models] Add remaining model PP support (#7168)

Signed-off-by: Muralidhar Andoorveedu <muralidhar.andoorveedu@centml.ai>
Signed-off-by: Murali Andoorveedu <muralidhar.andoorveedu@centml.ai>
Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Murali Andoorveedu
2024-10-04 10:56:58 +08:00
committed by GitHub
co-authored by DarkLight1337
parent 303d44790a
commit 0f6d7a9a34
69 changed files with 2585 additions and 1344 deletions
+40 -20
View File
@@ -1,5 +1,5 @@
import math
from typing import Iterable, List, Optional, Tuple
from typing import Iterable, List, Optional, Tuple, Union
import torch
from torch import nn
@@ -7,14 +7,13 @@ from transformers.configuration_utils import PretrainedConfig
from vllm.attention import Attention, AttentionMetadata
from vllm.config import CacheConfig, LoRAConfig
from vllm.distributed import (get_tensor_model_parallel_rank,
from vllm.distributed import (get_pp_group, get_tensor_model_parallel_rank,
get_tensor_model_parallel_world_size)
from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
QKVParallelLinear,
RowParallelLinear)
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.quantization.base_config import (
QuantizationConfig)
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.rotary_embedding import get_rope
from vllm.model_executor.layers.sampler import Sampler, SamplerOutput
from vllm.model_executor.layers.vocab_parallel_embedding import (
@@ -23,6 +22,10 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import IntermediateTensors
from .interfaces import SupportsPP
from .utils import (is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers)
def load_column_parallel_weight(param: torch.nn.Parameter,
loaded_weight: torch.Tensor):
@@ -301,20 +304,25 @@ class Phi3SmallModel(nn.Module):
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
):
super().__init__()
self.config = config
self.embed_tokens = VocabParallelEmbedding(config.vocab_size,
config.hidden_size)
self.mup_embedding_multiplier = config.mup_embedding_multiplier
self.layers = nn.ModuleList([
Phi3SmallDecoderLayer(config, layer_idx, cache_config,
quant_config)
for layer_idx in range(config.num_hidden_layers)
])
self.start_layer, self.end_layer, self.layers = make_layers(
config.num_hidden_layers,
lambda prefix: Phi3SmallDecoderLayer(config,
int(prefix.split('.')[-1]),
cache_config, quant_config),
prefix=f"{prefix}.layers")
self.final_layernorm = nn.LayerNorm(config.hidden_size,
eps=config.layer_norm_epsilon)
self.make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(["hidden_states"],
config.hidden_size))
def get_input_embeddings(self):
return self.embed_tokens
@@ -327,30 +335,37 @@ class Phi3SmallModel(nn.Module):
input_ids: torch.LongTensor,
positions: Optional[torch.LongTensor],
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata = None,
):
hidden_states = self.embed_tokens(input_ids)
if (self.mup_embedding_multiplier is not None
and self.mup_embedding_multiplier > 0.0):
hidden_states = hidden_states * self.mup_embedding_multiplier
for i in range(len(self.layers)):
attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors],
) -> Union[torch.Tensor, IntermediateTensors]:
if get_pp_group().is_first_rank:
hidden_states = self.embed_tokens(input_ids)
if (self.mup_embedding_multiplier is not None
and self.mup_embedding_multiplier > 0.0):
hidden_states = hidden_states * self.mup_embedding_multiplier
else:
assert intermediate_tensors
hidden_states = intermediate_tensors["hidden_states"]
for i in range(self.start_layer, self.end_layer):
layer = self.layers[i]
hidden_states = layer(
positions,
hidden_states,
kv_caches[i],
kv_caches[i - self.start_layer],
attn_metadata,
)
if not get_pp_group().is_last_rank:
return IntermediateTensors({"hidden_states": hidden_states})
hidden_states = self.final_layernorm(hidden_states)
return hidden_states
class Phi3SmallForCausalLM(nn.Module):
class Phi3SmallForCausalLM(nn.Module, SupportsPP):
_tied_weights_keys = ["lm_head.weight"]
def __init__(
self,
config,
config: PretrainedConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
lora_config: Optional[LoRAConfig] = None,
@@ -372,6 +387,8 @@ class Phi3SmallForCausalLM(nn.Module):
self.lm_head.weight = self.model.embed_tokens.weight
self.logits_processor = LogitsProcessor(config.vocab_size)
self.sampler = Sampler()
self.make_empty_intermediate_tensors = (
self.model.make_empty_intermediate_tensors)
# tokens in tiktoken but not used
if hasattr(config, 'dummy_token_indices'):
@@ -419,12 +436,13 @@ class Phi3SmallForCausalLM(nn.Module):
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors] = None,
) -> torch.Tensor:
) -> Union[torch.Tensor, IntermediateTensors]:
output_hidden_states = self.model(
input_ids=input_ids,
positions=positions,
kv_caches=kv_caches,
attn_metadata=attn_metadata,
intermediate_tensors=intermediate_tensors,
)
output_hidden_states = output_hidden_states
return output_hidden_states
@@ -447,6 +465,8 @@ class Phi3SmallForCausalLM(nn.Module):
continue
if name.endswith(".bias") and name not in params_dict:
continue
if is_pp_missing_parameter(name, self):
continue
param = params_dict[name]
weight_loader = getattr(param, "weight_loader",
default_weight_loader)