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

Update rope_scaling to rope_parameters in preparation for Transformers v5 (#28542)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-11-19 18:06:36 +01:00
committed by GitHub
parent d44e9df7d4
commit a8b70304d6
104 changed files with 542 additions and 910 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
vllm/model_executor/models/lfm2.py
View File

@@ -2,7 +2,6 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from collections.abc import Iterable
from itertools import islice
from typing import Any
import torch
import torch.nn as nn
@@ -96,8 +95,6 @@ class Lfm2Attention(nn.Module):
hidden_size: int,
num_heads: int,
num_kv_heads: int,
rope_theta: float = 10000,
rope_scaling: dict[str, Any] | None = None,
max_position_embeddings: int = 8192,
cache_config: CacheConfig | None = None,
quant_config: QuantizationConfig | None = None,
@@ -126,7 +123,6 @@ class Lfm2Attention(nn.Module):
self.q_size = self.num_heads * self.head_dim
self.kv_size = self.num_kv_heads * self.head_dim
self.scaling = self.head_dim**-0.5
self.rope_theta = rope_theta
self.max_position_embeddings = max_position_embeddings
self.qkv_proj = QKVParallelLinear(
@@ -149,8 +145,7 @@ class Lfm2Attention(nn.Module):
self.head_dim,
rotary_dim=self.head_dim,
max_position=self.max_position_embeddings,
base=self.rope_theta,
rope_scaling=rope_scaling,
rope_parameters=config.rope_parameters,
is_neox_style=True,
)
self.attn = Attention(
@@ -199,14 +194,6 @@ class Lfm2AttentionDecoderLayer(nn.Module):
self.config = config
self.layer_idx = layer_idx
rope_theta = getattr(config, "rope_theta", 10000)
rope_scaling = getattr(config, "rope_scaling", None)
if rope_scaling is not None and getattr(
config, "original_max_position_embeddings", None
):
rope_scaling["original_max_position_embeddings"] = (
config.original_max_position_embeddings
)
max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
self.self_attn = Lfm2Attention(
@@ -215,8 +202,6 @@ class Lfm2AttentionDecoderLayer(nn.Module):
hidden_size=config.hidden_size,
num_heads=config.num_attention_heads,
num_kv_heads=config.num_key_value_heads,
rope_theta=rope_theta,
rope_scaling=rope_scaling,
max_position_embeddings=max_position_embeddings,
cache_config=cache_config,
quant_config=quant_config,