biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] PP support for embedding models and update docs (#9090)

Browse Source 
Co-authored-by: Roger Wang <136131678+ywang96@users.noreply.github.com>
This commit is contained in:
Cyrus Leung
2024-10-06 16:35:27 +08:00
committed by GitHub
parent f22619fe96
commit b22b798471
12 changed files with 612 additions and 451 deletions
Download Patch File Download Diff File Expand all files Collapse all files

100
vllm/model_executor/models/gemma2.py
View File

@@ -40,7 +40,7 @@ from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import IntermediateTensors
from .interfaces import SupportsLoRA, SupportsPP
from .utils import (is_pp_missing_parameter,
from .utils import (group_weights_with_prefix, is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers)
logger = init_logger(__name__)
@@ -273,7 +273,7 @@ class Gemma2Model(nn.Module):
def forward(
self,
input_ids: torch.Tensor,
input_ids: Optional[torch.Tensor],
positions: torch.Tensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
@@ -308,6 +308,49 @@ class Gemma2Model(nn.Module):
hidden_states, _ = self.norm(hidden_states, residual)
return hidden_states
def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
stacked_params_mapping = [
# (param_name, shard_name, shard_id)
("qkv_proj", "q_proj", "q"),
("qkv_proj", "k_proj", "k"),
("qkv_proj", "v_proj", "v"),
("gate_up_proj", "gate_proj", 0),
("gate_up_proj", "up_proj", 1),
]
params_dict = dict(self.named_parameters())
loaded_params: Set[str] = set()
for name, loaded_weight in weights:
for (param_name, shard_name, shard_id) in stacked_params_mapping:
if shard_name not in name:
continue
name = name.replace(shard_name, param_name)
# Skip loading extra bias for GPTQ models.
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 = param.weight_loader
weight_loader(param, loaded_weight, shard_id)
break
else:
# Skip loading extra bias for GPTQ models.
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)
weight_loader(param, loaded_weight)
loaded_params.add(name)
unloaded_params = params_dict.keys() - loaded_params
if unloaded_params:
logger.warning(
"Some weights are not initialized from checkpoints: %s",
unloaded_params)
class Gemma2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
packed_modules_mapping = {
@@ -391,48 +434,19 @@ class Gemma2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
return next_tokens
def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
stacked_params_mapping = [
# (param_name, shard_name, shard_id)
("qkv_proj", "q_proj", "q"),
("qkv_proj", "k_proj", "k"),
("qkv_proj", "v_proj", "v"),
("gate_up_proj", "gate_proj", 0),
("gate_up_proj", "up_proj", 1),
]
params_dict = dict(self.named_parameters())
loaded_params: Set[str] = set()
for name, loaded_weight in weights:
for (param_name, shard_name, shard_id) in stacked_params_mapping:
if shard_name not in name:
weights_group = group_weights_with_prefix(weights)
self.model.load_weights(weights_group["model"])
if not self.config.tie_word_embeddings:
# NOTE: For now self.lm_head is not defined because
# tie_word_embeddings is assumed to the False
lm_head_dict = dict(self.lm_head.named_parameters())
for name, loaded_weight in weights_group["lm_head"]:
if is_pp_missing_parameter(name, self.lm_head):
continue
name = name.replace(shard_name, param_name)
# Skip loading extra bias for GPTQ models.
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 = param.weight_loader
weight_loader(param, loaded_weight, shard_id)
break
else:
# lm_head is not used in vllm as it is tied with embed_token.
# To prevent errors, skip loading lm_head.weight.
if "lm_head.weight" in name:
continue
# Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict:
continue
if is_pp_missing_parameter(name, self):
continue
param = params_dict[name]
param = lm_head_dict[name]
weight_loader = getattr(param, "weight_loader",
default_weight_loader)
weight_loader(param, loaded_weight)
loaded_params.add(name)
unloaded_params = params_dict.keys() - loaded_params
if unloaded_params:
logger.warning(
"Some weights are not initialized from checkpoints: %s",
unloaded_params)