c0c2eb614e
### What this PR does / why we need it?
This PR extends original `rope_triton_forward` and
`split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as
inputs. This fully aligns to vLLM RoPE api interface. Compared with
earlier implementation for RoPE, the benefits are:
1. avoiding pre-computation of `cos` `sin` before model execution, which
helps to remove redundant codes.
2. allowing eagle3 draft model to have different rope parameters with
main model (see #6612 ). This help to recover accept rate && accuracy in
that case.
In addition, this kernel change only introduces very small performance
degradation. Those `index_select` or `chunk` operations are now changed
into simple memory access in triton kernel (For example,
https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81).
**Highlights**
- **RoPE Cache Unification**: Replaced separate _sin and _cos global
tensors with a unified cos_sin_cache and explicit positions tensor for
Rotary Positional Embeddings (RoPE), streamlining data handling.
- **Triton Kernel Integration**: Updated Triton kernels
(split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the
cos_sin_cache and positions for more efficient and integrated RoPE
calculations.
- **Custom Operation Registration**: Registered `rope_forward_oot` as a
new custom operation, allowing its use in fused compilation passes and
providing a dedicated entry point for the new RoPE implementation.
- **Refactored RoPE Forward Pass**: Modified the rope_forward_oot
function to accept the new cos_sin_cache and positions arguments,
enabling a more flexible and integrated RoPE application within the
system.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
5326c89803
Additional test on Qwen3-235b accuracy:
| Aime2024 | GSM8K | Livecodebench |
| -------- | -------- | -------- |
| 83.33 | 96.26 | 70.23 |
---------
Signed-off-by: Angazenn <supperccell@163.com>
243 lines
9.8 KiB
Python
243 lines
9.8 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
# This file is a part of the vllm-ascend project.
|
|
#
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
#
|
|
import torch
|
|
import torch._inductor.pattern_matcher as pm
|
|
from torch._inductor.pattern_matcher import PatternMatcherPass, PatternPrettyPrinter
|
|
from vllm.config import VllmConfig, get_layers_from_vllm_config
|
|
from vllm.config.compilation import Range
|
|
from vllm.logger import logger
|
|
|
|
from vllm_ascend.utils import vllm_version_is
|
|
|
|
if vllm_version_is("v0.15.0"):
|
|
from vllm.attention.layer import Attention # type: ignore
|
|
from vllm.compilation.vllm_inductor_pass import VllmInductorPass # type: ignore
|
|
else:
|
|
from vllm.compilation.passes.vllm_inductor_pass import VllmInductorPass
|
|
from vllm.model_executor.layers.attention import Attention
|
|
|
|
|
|
class QKNormRopeFusionPattern:
|
|
def __init__(self, vllm_config, head_dim, num_heads, num_kv_heads, eps=1e-6):
|
|
self.vllm_config = vllm_config
|
|
self.head_dim = head_dim
|
|
self.num_heads = num_heads
|
|
self.num_kv_heads = num_kv_heads
|
|
self.q_size = self.num_heads * self.head_dim
|
|
self.kv_size = self.num_kv_heads * self.head_dim
|
|
self.eps = eps
|
|
self.device = vllm_config.device_config.device if vllm_config.device_config else None
|
|
|
|
def get_inputs(self):
|
|
T = 5
|
|
max_position_embeddings = 16384
|
|
qkv = torch.empty(T, self.q_size + 2 * self.kv_size, dtype=torch.bfloat16, device="npu")
|
|
q_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
k_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
cos_sin_cache = torch.empty(max_position_embeddings, self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
positions = torch.ones(T, dtype=torch.int64, device="npu")
|
|
return [qkv, q_weight, k_weight, cos_sin_cache, positions]
|
|
|
|
def register(self, pm_pass: PatternMatcherPass):
|
|
def pattern(
|
|
qkv: torch.Tensor,
|
|
q_weight: torch.Tensor,
|
|
k_weight: torch.Tensor,
|
|
cos_sin_cache: torch.Tensor,
|
|
positions: torch.Tensor,
|
|
):
|
|
q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
|
|
|
|
q_by_head = q.view(*q.shape[:-1], q.shape[-1] // self.head_dim, self.head_dim)
|
|
q_norm_out, _ = torch.ops.npu.npu_rms_norm(q_by_head, q_weight, self.eps)
|
|
|
|
k_by_head = k.view(*k.shape[:-1], k.shape[-1] // self.head_dim, self.head_dim)
|
|
k_norm_out, _ = torch.ops.npu.npu_rms_norm(k_by_head, k_weight, self.eps)
|
|
|
|
q_flat = q_norm_out.view(q.shape)
|
|
k_flat = k_norm_out.view(k.shape)
|
|
q_rope, k_rope = torch.ops.vllm.npu_rotary_embedding(
|
|
positions, q_flat, k_flat, cos_sin_cache, self.head_dim, self.head_dim, True
|
|
)
|
|
|
|
return q_rope, k_rope, v
|
|
|
|
def replacement(
|
|
qkv: torch.Tensor,
|
|
q_weight: torch.Tensor,
|
|
k_weight: torch.Tensor,
|
|
cos_sin_cache: torch.Tensor,
|
|
positions: torch.Tensor,
|
|
):
|
|
results = torch.ops.vllm.qkv_rmsnorm_rope(
|
|
input=qkv,
|
|
q_weight=q_weight,
|
|
k_weight=k_weight,
|
|
q_hidden_size=self.q_size,
|
|
kv_hidden_size=self.kv_size,
|
|
head_dim=self.head_dim,
|
|
eps=self.eps,
|
|
q_bias=None,
|
|
k_bias=None,
|
|
cos_sin_cache=cos_sin_cache,
|
|
positions=positions,
|
|
)
|
|
|
|
return results
|
|
|
|
pm.register_replacement(pattern, replacement, self.get_inputs(), pm.fwd_only, pm_pass)
|
|
|
|
|
|
class QKNormRopeFusionPatternWithBias:
|
|
def __init__(self, vllm_config, head_dim, num_heads, num_kv_heads, eps=1e-6):
|
|
self.head_dim = head_dim
|
|
self.num_heads = num_heads
|
|
self.num_kv_heads = num_kv_heads
|
|
self.q_size = self.num_heads * self.head_dim
|
|
self.kv_size = self.num_kv_heads * self.head_dim
|
|
self.eps = eps
|
|
self.vllm_config = vllm_config
|
|
self.device = vllm_config.device_config.device if vllm_config.device_config else None
|
|
|
|
def get_inputs(self):
|
|
T = 5
|
|
max_position_embeddings = 16384
|
|
qkv = torch.empty(T, self.q_size + 2 * self.kv_size, dtype=torch.bfloat16, device="npu")
|
|
q_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
k_weight = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
q_bias = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
k_bias = torch.empty(self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
cos_sin_cache = torch.empty(max_position_embeddings, self.head_dim, dtype=torch.bfloat16, device="npu")
|
|
positions = torch.ones(T, dtype=torch.int64, device="npu")
|
|
|
|
return [qkv, q_weight, k_weight, q_bias, k_bias, cos_sin_cache, positions]
|
|
|
|
def register(self, pm_pass: PatternMatcherPass):
|
|
def pattern(
|
|
qkv: torch.Tensor,
|
|
q_weight: torch.Tensor,
|
|
k_weight: torch.Tensor,
|
|
q_bias: torch.Tensor,
|
|
k_bias: torch.Tensor,
|
|
cos_sin_cache: torch.Tensor,
|
|
positions: torch.Tensor,
|
|
):
|
|
q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
|
|
|
|
q_by_head = q.view(*q.shape[:-1], q.shape[-1] // self.head_dim, self.head_dim)
|
|
q_norm_out, _ = torch.ops.npu.npu_rms_norm(q_by_head, q_weight, self.eps)
|
|
q_normed = q_norm_out + q_bias
|
|
|
|
k_by_head = k.view(*k.shape[:-1], k.shape[-1] // self.head_dim, self.head_dim)
|
|
k_norm_out, _ = torch.ops.npu.npu_rms_norm(k_by_head, k_weight, self.eps)
|
|
k_normed = k_norm_out + k_bias
|
|
|
|
q_flat = q_normed.view(q.shape)
|
|
k_flat = k_normed.view(k.shape)
|
|
q_rope, k_rope = torch.ops.vllm.npu_rotary_embedding(
|
|
positions, q_flat, k_flat, cos_sin_cache, self.head_dim, self.head_dim, True
|
|
)
|
|
|
|
return q_rope, k_rope, v
|
|
|
|
def replacement(
|
|
qkv: torch.Tensor,
|
|
q_weight: torch.Tensor,
|
|
k_weight: torch.Tensor,
|
|
q_bias: torch.Tensor,
|
|
k_bias: torch.Tensor,
|
|
cos_sin_cache: torch.Tensor,
|
|
positions: torch.Tensor,
|
|
):
|
|
results = torch.ops.vllm.qkv_rmsnorm_rope(
|
|
input=qkv,
|
|
q_weight=q_weight,
|
|
k_weight=k_weight,
|
|
q_hidden_size=self.q_size,
|
|
kv_hidden_size=self.kv_size,
|
|
head_dim=self.head_dim,
|
|
eps=self.eps,
|
|
q_bias=q_bias,
|
|
k_bias=k_bias,
|
|
cos_sin_cache=cos_sin_cache,
|
|
positions=positions,
|
|
)
|
|
return results
|
|
|
|
pm.register_replacement(pattern, replacement, self.get_inputs(), pm.fwd_only, pm_pass)
|
|
|
|
|
|
class QKNormRopeFusionPass(VllmInductorPass):
|
|
"""
|
|
A pass for fusing QKV split and RMSNorm operations into a single qk_rmsnorm operator.
|
|
"""
|
|
|
|
def __init__(self, vllm_config: VllmConfig):
|
|
super().__init__(vllm_config)
|
|
self.pattern_match_passes: PatternMatcherPass = PatternMatcherPass(pass_name="qknorm_rope_fusion_pass")
|
|
|
|
dtype = vllm_config.model_config.dtype
|
|
if dtype not in (torch.bfloat16,):
|
|
logger.debug("QKNorm and Rope fusion not enabled: unsupported dtype %s", dtype)
|
|
return
|
|
|
|
# use one attn layer to get meta (such as head_dim) for QKNormRopeFusionPattern
|
|
attn_layers: dict[str, Attention] = get_layers_from_vllm_config(vllm_config, Attention)
|
|
if len(attn_layers) == 0:
|
|
logger.debug("QKNorm and Rope fusion enabled, but no Attention layers were discovered.")
|
|
return
|
|
layer = next(iter(attn_layers.values()))
|
|
for epsilon in [1e-6, 1e-5]:
|
|
if layer.head_size != 128:
|
|
logger.debug("QKNorm and Rope fusion not enabled: head_dim %d is not equal of 128", layer.head_size)
|
|
continue
|
|
QKNormRopeFusionPattern(
|
|
vllm_config=vllm_config,
|
|
head_dim=layer.head_size,
|
|
num_heads=layer.num_heads,
|
|
num_kv_heads=layer.num_kv_heads,
|
|
eps=epsilon,
|
|
).register(self.pattern_match_passes)
|
|
|
|
QKNormRopeFusionPatternWithBias(
|
|
vllm_config=vllm_config,
|
|
head_dim=layer.head_size,
|
|
num_heads=layer.num_heads,
|
|
num_kv_heads=layer.num_kv_heads,
|
|
eps=epsilon,
|
|
).register(self.pattern_match_passes)
|
|
|
|
def __call__(self, graph: torch.fx.Graph):
|
|
self.begin()
|
|
self.matched_count = self.pattern_match_passes.apply(graph)
|
|
logger.debug("Fused %s QKNorm and Rope patterns", self.matched_count)
|
|
logger.debug("Patterns registered for replacement:")
|
|
pattern_idx = 0
|
|
for pattern_entry in self.pattern_match_passes.patterns.values():
|
|
for p in pattern_entry:
|
|
p_str = PatternPrettyPrinter.run(p.pattern)
|
|
logger.debug("Pattern %d: %s", pattern_idx, p_str)
|
|
pattern_idx += 1
|
|
self.end_and_log()
|
|
|
|
def is_applicable_for_range(self, compile_range: Range) -> bool:
|
|
"""
|
|
Check if the pass is applicable for the current configuration.
|
|
"""
|
|
return True
|