Remove even more unnecessary load_weights methods (#48496)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
@@ -14,10 +14,7 @@ from vllm.config import CacheConfig, ModelConfig, VllmConfig
|
||||
from vllm.distributed import get_tensor_model_parallel_world_size
|
||||
from vllm.distributed.parallel_state import get_pp_group
|
||||
from vllm.model_executor.layers.attention import Attention
|
||||
from vllm.model_executor.layers.fused_moe import (
|
||||
FusedMoE,
|
||||
fused_moe_make_expert_params_mapping,
|
||||
)
|
||||
from vllm.model_executor.layers.fused_moe import FusedMoE
|
||||
from vllm.model_executor.layers.layernorm import RMSNorm
|
||||
from vllm.model_executor.layers.linear import (
|
||||
QKVParallelLinear,
|
||||
@@ -38,7 +35,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
ParallelLMHead,
|
||||
VocabParallelEmbedding,
|
||||
)
|
||||
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
|
||||
from vllm.model_executor.models.llama import LlamaMLP as JambaMLP
|
||||
from vllm.sequence import IntermediateTensors
|
||||
|
||||
@@ -52,7 +48,6 @@ from .interfaces import (
|
||||
from .utils import (
|
||||
AutoWeightsLoader,
|
||||
WeightsMapper,
|
||||
is_pp_missing_parameter,
|
||||
make_empty_intermediate_tensors_factory,
|
||||
make_layers,
|
||||
maybe_prefix,
|
||||
@@ -378,87 +373,6 @@ class JambaModel(nn.Module):
|
||||
hidden_states, _ = self.final_layernorm(hidden_states, residual)
|
||||
return hidden_states
|
||||
|
||||
def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
|
||||
# Params for weights, fp8 weight scales, fp8 activation scales
|
||||
# (param_name, weight_name, expert_id, shard_id)
|
||||
return fused_moe_make_expert_params_mapping(
|
||||
self,
|
||||
ckpt_gate_proj_name="gate_proj",
|
||||
ckpt_down_proj_name="down_proj",
|
||||
ckpt_up_proj_name="up_proj",
|
||||
num_experts=self.config.num_experts,
|
||||
)
|
||||
|
||||
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),
|
||||
]
|
||||
|
||||
params_dict = dict(self.named_parameters())
|
||||
loaded_params: set[str] = set()
|
||||
expert_params_mapping = self.get_expert_mapping()
|
||||
for name, loaded_weight in weights:
|
||||
if "rotary_emb.inv_freq" in name:
|
||||
continue
|
||||
for param_name, weight_name, shard_id in stacked_params_mapping:
|
||||
if weight_name not in name:
|
||||
continue
|
||||
if "experts" 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
|
||||
# Skip layers on other devices.
|
||||
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,
|
||||
shard_id,
|
||||
) in expert_params_mapping:
|
||||
if weight_name not in name:
|
||||
continue
|
||||
|
||||
if is_pp_missing_parameter(name, self):
|
||||
continue
|
||||
name = name.replace(weight_name, param_name)
|
||||
param = params_dict[name]
|
||||
weight_loader = param.weight_loader
|
||||
weight_loader(
|
||||
param,
|
||||
loaded_weight,
|
||||
name,
|
||||
shard_id=shard_id,
|
||||
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 JambaForCausalLM(
|
||||
nn.Module,
|
||||
@@ -470,6 +384,14 @@ class JambaForCausalLM(
|
||||
):
|
||||
hf_to_vllm_mapper = WeightsMapper(
|
||||
orig_to_new_substr={".self_attn.": ".", ".A_log": ".A"},
|
||||
orig_to_new_stacked={
|
||||
".q_proj": (".qkv_proj", "q"),
|
||||
".k_proj": (".qkv_proj", "k"),
|
||||
".v_proj": (".qkv_proj", "v"),
|
||||
# .experts.gate_up_proj must be handled by MoERunner.load_weights for EP
|
||||
".feed_forward.gate_proj": (".feed_forward.gate_up_proj", 0),
|
||||
".feed_forward.up_proj": (".feed_forward.gate_up_proj", 1),
|
||||
},
|
||||
)
|
||||
packed_modules_mapping = {
|
||||
"qkv_proj": [
|
||||
@@ -577,9 +499,6 @@ class JambaForCausalLM(
|
||||
loader = AutoWeightsLoader(self)
|
||||
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
|
||||
|
||||
def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
|
||||
return self.model.get_expert_mapping()
|
||||
|
||||
|
||||
class JambaForSequenceClassification(JambaForCausalLM):
|
||||
is_pooling_model = True
|
||||
|
||||
Reference in New Issue
Block a user