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

Remove unused kwargs from model definitions (#13555)

Browse Source
This commit is contained in:
Harry Mellor
2025-02-25 01:13:52 +00:00
committed by GitHub
parent f61528d46d
commit cdc1fa12eb
104 changed files with 436 additions and 1654 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
vllm/model_executor/models/molmo.py
View File

@@ -16,7 +16,7 @@ from transformers import (BatchFeature, PretrainedConfig, ProcessorMixin,
from transformers.image_utils import ImageInput
from transformers.tokenization_utils_base import TextInput
from vllm.attention import Attention, AttentionMetadata
from vllm.attention import Attention
from vllm.attention.layer import MultiHeadAttention
from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, VllmConfig
@@ -460,15 +460,13 @@ class MolmoAttention(nn.Module):
self,
positions: torch.Tensor,
hidden_states: torch.Tensor,
kv_cache: torch.Tensor,
attn_metadata: AttentionMetadata,
) -> torch.Tensor:
qkv, _ = self.qkv_proj(hidden_states)
q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
if self.q_norm is not None and self.k_norm is not None:
q, k = self._apply_qk_norm(q, k)
q, k = self.rotary_emb(positions, q, k)
attn_output = self.attn(q, k, v, kv_cache, attn_metadata)
attn_output = self.attn(q, k, v)
output, _ = self.o_proj(attn_output)
return output
@@ -580,8 +578,6 @@ class MolmoDecoderLayer(nn.Module):
self,
positions: torch.Tensor,
hidden_states: torch.Tensor,
kv_cache: torch.Tensor,
attn_metadata: AttentionMetadata,
residual: Optional[torch.Tensor],
) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor, torch.Tensor]]]:
# Self Attention
@@ -594,8 +590,6 @@ class MolmoDecoderLayer(nn.Module):
hidden_states = self.self_attn(
positions=positions,
hidden_states=hidden_states,
kv_cache=kv_cache,
attn_metadata=attn_metadata,
)
hidden_states, residual = self.post_attention_layernorm(
@@ -610,8 +604,6 @@ class MolmoDecoderNormAfterLayer(MolmoDecoderLayer):
self,
positions: torch.Tensor,
hidden_states: torch.Tensor,
kv_cache: torch.Tensor,
attn_metadata: AttentionMetadata,
residual: Optional[torch.Tensor],
) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor, torch.Tensor]]]:
# Self Attention
@@ -619,8 +611,6 @@ class MolmoDecoderNormAfterLayer(MolmoDecoderLayer):
hidden_states = self.self_attn(
positions=positions,
hidden_states=hidden_states,
kv_cache=kv_cache,
attn_metadata=attn_metadata,
)
hidden_states = self.input_layernorm(hidden_states)
@@ -841,8 +831,6 @@ class MolmoModel(nn.Module, SupportsQuant):
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
) -> torch.Tensor:
@@ -858,13 +846,10 @@ class MolmoModel(nn.Module, SupportsQuant):
residual = intermediate_tensors["residual"]
# Apply blocks one-by-one.
for i in range(self.start_layer, self.end_layer):
layer = self.layers[i]
for layer in self.layers[self.start_layer:self.end_layer]:
hidden_states, residual = layer(
positions,
hidden_states,
kv_caches[i - self.start_layer],
attn_metadata,
residual,
)
if not get_pp_group().is_last_rank:
@@ -1643,8 +1628,6 @@ class MolmoForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA,
self,
input_ids: torch.LongTensor,
positions: torch.LongTensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
**kwargs: object,
@@ -1663,8 +1646,6 @@ class MolmoForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA,
hidden_states = self.model(input_ids,
positions,
kv_caches,
attn_metadata,
intermediate_tensors,
inputs_embeds=inputs_embeds)