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

[Model] Clean up MiniCPMV (#10751)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2024-11-29 12:47:06 +08:00
committed by GitHub
parent c83919c7a6
commit fa6ecb9aa7
7 changed files with 149 additions and 215 deletions
Download Patch File Download Diff File Expand all files Collapse all files

153
vllm/model_executor/models/minicpm.py
View File

@@ -52,7 +52,7 @@ from vllm.model_executor.utils import set_weight_attrs
from vllm.sequence import IntermediateTensors
from .interfaces import SupportsLoRA, SupportsPP
from .utils import (is_pp_missing_parameter,
from .utils import (AutoWeightsLoader, is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers,
maybe_prefix)
@@ -378,6 +378,7 @@ class MiniCPMModel(nn.Module):
config.hidden_size,
org_num_embeddings=config.vocab_size,
)
self.num_experts = getattr(self.config, "num_experts", 0)
self._init_layers(prefix, config, cache_config, quant_config)
self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.make_empty_intermediate_tensors = (
@@ -437,6 +438,73 @@ class MiniCPMModel(nn.Module):
hidden_states = self.norm(hidden_states)
return hidden_states
def load_weights(self, weights: Iterable[Tuple[str,
torch.Tensor]]) -> Set[str]:
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),
]
expert_params_mapping = [
# (param_name, weight_name, expert_id)
("ws" if weight_name in ["w1", "w3"] else "w2s",
f"experts.{expert_id}.{weight_name}.weight", expert_id)
for expert_id in range(self.num_experts)
for weight_name in ["w1", "w2", "w3"]
]
params_dict = dict(self.named_parameters())
loaded_params: Set[str] = set()
for name, loaded_weight in weights:
if "rotary_emb.inv_freq" in name:
continue
if ("rotary_emb.cos_cached" in name
or "rotary_emb.sin_cached" in name):
# Models trained using ColossalAI may include these tensors in
# the checkpoint. Skip them.
continue
for (param_name, weight_name, shard_id) in stacked_params_mapping:
if weight_name not in name:
continue
name = name.replace(weight_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:
for param_name, weight_name, expert_id in expert_params_mapping:
if weight_name not in name:
continue
name = name.replace(weight_name, param_name)
if is_pp_missing_parameter(name, self):
continue
param = params_dict[name]
weight_loader = param.weight_loader
weight_loader(param,
loaded_weight,
weight_name,
expert_id=expert_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)
return loaded_params
class MiniCPMForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
packed_modules_mapping = {
@@ -480,8 +548,9 @@ class MiniCPMForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
self.cache_config = cache_config
self.quant_config = quant_config
self.num_experts = getattr(self.config, "num_experts", 0)
self._init_model(vllm_config=vllm_config, prefix=prefix)
self.model = self._init_model(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
unpadded_vocab_size = config.vocab_size
if lora_config:
unpadded_vocab_size += lora_config.lora_extra_vocab_size
@@ -506,8 +575,7 @@ class MiniCPMForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
self.model.make_empty_intermediate_tensors)
def _init_model(self, *, vllm_config: VllmConfig, prefix: str = ""):
self.model = MiniCPMModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
return MiniCPMModel(vllm_config=vllm_config, prefix=prefix)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.model.get_input_embeddings(input_ids)
@@ -546,72 +614,9 @@ class MiniCPMForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
def load_weights(self, weights: Iterable[Tuple[str,
torch.Tensor]]) -> Set[str]:
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),
]
expert_params_mapping = [
# (param_name, weight_name, expert_id)
("ws" if weight_name in ["w1", "w3"] else "w2s",
f"experts.{expert_id}.{weight_name}.weight", expert_id)
for expert_id in range(self.num_experts)
for weight_name in ["w1", "w2", "w3"]
]
params_dict = dict(self.named_parameters())
loaded_params: Set[str] = set()
for name, loaded_weight in weights:
if "rotary_emb.inv_freq" in name:
continue
if ("rotary_emb.cos_cached" in name
or "rotary_emb.sin_cached" in name):
# Models trained using ColossalAI may include these tensors in
# the checkpoint. Skip them.
continue
# With tie_word_embeddings, we can skip lm_head.weight
# The weight might appear unnecessarily in the files if the model is
# processed with quantization, LoRA, fine-tuning, etc.
if self.config.tie_word_embeddings and "lm_head.weight" in name:
continue
for (param_name, weight_name, shard_id) in stacked_params_mapping:
if weight_name not in name:
continue
name = name.replace(weight_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:
for param_name, weight_name, expert_id in expert_params_mapping:
if weight_name not in name:
continue
name = name.replace(weight_name, param_name)
if is_pp_missing_parameter(name, self):
continue
param = params_dict[name]
weight_loader = param.weight_loader
weight_loader(param,
loaded_weight,
weight_name,
expert_id=expert_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)
return loaded_params
loader = AutoWeightsLoader(
self,
skip_prefixes=(["lm_head."]
if self.config.tie_word_embeddings else None),
)
return loader.load_weights(weights)