activation.py 20.3 KB
Newer Older
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
"""Custom activation functions."""
4
import math
5
6
from typing import Optional

Woosuk Kwon's avatar
Woosuk Kwon committed
7
8
import torch
import torch.nn as nn
9
import torch.nn.functional as F
Woosuk Kwon's avatar
Woosuk Kwon committed
10

11
12
from vllm.distributed import (divide, get_tensor_model_parallel_rank,
                              get_tensor_model_parallel_world_size)
13
from vllm.logger import init_logger
14
from vllm.model_executor.custom_op import CustomOp
15
from vllm.model_executor.utils import set_weight_attrs
16
from vllm.platforms import current_platform
17
from vllm.utils import LazyDict
Woosuk Kwon's avatar
Woosuk Kwon committed
18

19
20
logger = init_logger(__name__)

Woosuk Kwon's avatar
Woosuk Kwon committed
21

22
@CustomOp.register("fatrelu_and_mul")
23
24
class FatreluAndMul(CustomOp):
    """An activation function for FATReLU.
25

26
27
28
29
30
31
32
33
34
35
36
37
    The function computes x -> FATReLU(x[:d]) * x[d:] where
    d = x.shape[-1] // 2.
    This is used in openbmb/MiniCPM-S-1B-sft.

    Shapes:
        x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
        return: (num_tokens, d) or (batch_size, seq_len, d)
    """

    def __init__(self, threshold: float = 0.):
        super().__init__()
        self.threshold = threshold
38
        if current_platform.is_cuda_alike():
39
            self.op = torch.ops._C.fatrelu_and_mul
40
41
        elif current_platform.is_cpu():
            self._forward_method = self.forward_native
42
43
44
45
46
47
48
49
50

    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
        d = x.shape[-1] // 2
        x1 = x[..., :d]
        x2 = x[..., d:]
        x1 = F.threshold(x1, self.threshold, 0.0)
        return x1 * x2

    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
51
52
53
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
54
        self.op(out, x, self.threshold)
55
        return out
56
57


58
@CustomOp.register("silu_and_mul")
59
class SiluAndMul(CustomOp):
60
61
    """An activation function for SwiGLU.

62
    The function computes x -> silu(x[:d]) * x[d:] where d = x.shape[-1] // 2.
Woosuk Kwon's avatar
Woosuk Kwon committed
63

64
    Shapes:
65
66
        x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
        return: (num_tokens, d) or (batch_size, seq_len, d)
67
    """
Woosuk Kwon's avatar
Woosuk Kwon committed
68

69
70
    def __init__(self):
        super().__init__()
71
        if current_platform.is_cuda_alike():
72
73
            self.op = torch.ops._C.silu_and_mul
        elif current_platform.is_xpu():
74
75
            from vllm._ipex_ops import ipex_ops
            self.op = ipex_ops.silu_and_mul
76
77
        elif current_platform.is_cpu():
            self._forward_method = self.forward_native
78

79
    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
80
81
82
83
        """PyTorch-native implementation equivalent to forward()."""
        d = x.shape[-1] // 2
        return F.silu(x[..., :d]) * x[..., d:]

84
    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
85
86
87
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
88
        self.op(out, x)
Woosuk Kwon's avatar
Woosuk Kwon committed
89
        return out
90

91
92
93
94
    def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
95
        self.op(out, x)
96
97
        return out

98
99
100
101
102
103
104
    def forward_neuron(self, x: torch.Tensor) -> torch.Tensor:
        d = x.shape[-1] // 2
        x_reshaped = x.view(-1, x.shape[-1])
        s = x_reshaped[:, :d] * F.sigmoid(x_reshaped[:, :d])
        result = s * x_reshaped[:, d:]
        return result.view(*x.shape[:-1], d)

105

106
107
108
109
110
111
112
113
114
115
116
117
118
@CustomOp.register("mul_and_silu")
class MulAndSilu(CustomOp):
    """An activation function for SwiGLU.

    The function computes x -> x[:d] * silu(x[d:]) where d = x.shape[-1] // 2.

    Shapes:
        x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
        return: (num_tokens, d) or (batch_size, seq_len, d)
    """

    def __init__(self):
        super().__init__()
119
        if current_platform.is_cuda_alike():
120
121
122
123
            self.op = torch.ops._C.mul_and_silu
        elif current_platform.is_xpu():
            from vllm._ipex_ops import ipex_ops
            self.op = ipex_ops.silu_and_mul
124
125
        elif current_platform.is_cpu():
            self._forward_method = self.forward_native
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142

    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
        """PyTorch-native implementation equivalent to forward()."""
        d = x.shape[-1] // 2
        return x[..., :d] * F.silu(x[..., d:])

    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
        self.op(out, x)
        return out

    # TODO implement forward_xpu for MulAndSilu
    # def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:


Robert Shaw's avatar
Robert Shaw committed
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
@CustomOp.register("gelu_and_mul_sparse")
class GeluAndMulSparse(CustomOp):
    """An activation function for GeluAndMulSparse.
    This activation function is used in Gemma3n. It computes:
        up_proj = self.up_proj(x)
        gate_proj = self.gate_proj(x)
        gate_proj = self._gaussian_topk(gate_proj) # sparsity
        activations = self.act_fn(gate_proj) # gelu
        down_proj = self.down_proj(activations * up_proj)
    Shapes:
        x: (num_tokens, 2 * d) or (batch_size, seq_len, 2 * d)
        return: (num_tokens, d) or (batch_size, seq_len, d)
    """

    def __init__(self, activation_sparsity: float, approximate: str = "none"):
        super().__init__()
        # Gelu.
        self.approximate = approximate
        if approximate not in ("none", "tanh"):
            raise ValueError(f"Unknown approximate mode: {approximate}")

        # Sparsity.
        if activation_sparsity == 0.0:
            raise ValueError(
                "activation_sparsity is 0.0. Please use GeluAndMul.")
        target_sparsity_tensor = torch.tensor(activation_sparsity,
                                              dtype=torch.float32)
        normal_dist = torch.distributions.normal.Normal(0, 1)
        self.std_multiplier = normal_dist.icdf(target_sparsity_tensor)

    def _gaussian_topk(self, x: torch.Tensor) -> torch.Tensor:
        """Get % sparse percentile of the Gaussian distribution."""
        # NOTE(rob): for TP>1, we could all-gather to get the means/std.
        # But we do not do this because in expectation they are the same
        # and in practice the eval scores are good without gathering.
        mean = torch.mean(x, dim=-1, keepdim=True)
        std = torch.std(x, dim=-1, keepdim=True, unbiased=False)
        cutoff_x = mean + std * self.std_multiplier
        return nn.functional.relu(x - cutoff_x)

    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
        """PyTorch-native implementation equivalent to forward()."""
        d = x.shape[-1] // 2
        out = self._gaussian_topk(x[..., :d])
        out = F.gelu(out, approximate=self.approximate)
        return out * x[..., d:]

    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
        return self.forward_native(x)


194
@CustomOp.register("gelu_and_mul")
195
class GeluAndMul(CustomOp):
196
197
198
199
200
201
202
203
204
    """An activation function for GeGLU.

    The function computes x -> GELU(x[:d]) * x[d:] where d = x.shape[-1] // 2.

    Shapes:
        x: (batch_size, seq_len, 2 * d) or (num_tokens, 2 * d)
        return: (batch_size, seq_len, d) or (num_tokens, d)
    """

205
206
207
208
209
    def __init__(self, approximate: str = "none"):
        super().__init__()
        self.approximate = approximate
        if approximate not in ("none", "tanh"):
            raise ValueError(f"Unknown approximate mode: {approximate}")
210
211
212
213
214
215
216
217
218
219
220
        if current_platform.is_cuda_alike() or current_platform.is_cpu():
            if approximate == "none":
                self.op = torch.ops._C.gelu_and_mul
            elif approximate == "tanh":
                self.op = torch.ops._C.gelu_tanh_and_mul
        elif current_platform.is_xpu():
            from vllm._ipex_ops import ipex_ops
            if approximate == "none":
                self.op = ipex_ops.gelu_and_mul
            else:
                self.op = ipex_ops.gelu_tanh_and_mul
221

222
    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
223
224
        """PyTorch-native implementation equivalent to forward()."""
        d = x.shape[-1] // 2
225
        return F.gelu(x[..., :d], approximate=self.approximate) * x[..., d:]
226

227
    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
228
229
230
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
231
        self.op(out, x)
232
233
        return out

234
235
236
237
    def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
238
        self.op(out, x)
239
240
        return out

241
242
243
    def extra_repr(self) -> str:
        return f'approximate={repr(self.approximate)}'

244

245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
@CustomOp.register("swigluoai_and_mul")
class SwigluOAIAndMul(CustomOp):
    # https://github.com/huggingface/transformers/blob/v4.55.0/src/transformers/models/gpt_oss/modeling_gpt_oss.py#L106-L110
    def __init__(self, alpha: float = 1.702, limit: float = 7.0):
        super().__init__()
        self.alpha = alpha
        self.limit = limit

    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
        """PyTorch-native implementation equivalent to forward()."""

        gate, up = x[..., ::2], x[..., 1::2]
        gate = gate.clamp(min=None, max=self.limit)
        up = up.clamp(min=-self.limit, max=self.limit)
        glu = gate * torch.sigmoid(gate * self.alpha)
        gated_output = (up + 1) * glu
        return gated_output

    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
        d = x.shape[-1] // 2
        output_shape = (x.shape[:-1] + (d, ))
        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
        torch.ops._C.swigluoai_and_mul(out, x, self.alpha, self.limit)
        return out

    def extra_repr(self) -> str:
        return f"alpha={repr(self.alpha)}, limit={repr(self.limit)}"


274
@CustomOp.register("gelu_new")
275
class NewGELU(CustomOp):
276

277
278
279
280
281
282
283
284
    def __init__(self):
        super().__init__()
        if current_platform.is_cuda_alike() or current_platform.is_cpu():
            self.op = torch.ops._C.gelu_new
        elif current_platform.is_xpu():
            from vllm._ipex_ops import ipex_ops
            self.op = ipex_ops.gelu_new

285
    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
286
287
288
289
290
        """PyTorch-native implementation equivalent to forward()."""
        c = math.sqrt(2.0 / math.pi)
        return 0.5 * x * (1.0 + torch.tanh(c *
                                           (x + 0.044715 * torch.pow(x, 3.0))))

291
    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
292
        out = torch.empty_like(x)
293
        self.op(out, x)
294
295
        return out

296
    def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
297
        return self.op(x)
298

299

300
@CustomOp.register("gelu_fast")
301
class FastGELU(CustomOp):
302

303
304
305
306
307
308
309
310
    def __init__(self):
        super().__init__()
        if current_platform.is_cuda_alike() or current_platform.is_cpu():
            self.op = torch.ops._C.gelu_fast
        elif current_platform.is_xpu():
            from vllm._ipex_ops import ipex_ops
            self.op = ipex_ops.gelu_fast

311
    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
312
313
314
315
        """PyTorch-native implementation equivalent to forward()."""
        return 0.5 * x * (1.0 + torch.tanh(x * 0.7978845608 *
                                           (1.0 + 0.044715 * x * x)))

316
    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
317
        out = torch.empty_like(x)
318
        self.op(out, x)
319
320
        return out

321
    def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
322
        return self.op(x)
323

324

325
@CustomOp.register("quick_gelu")
326
327
class QuickGELU(CustomOp):
    # https://github.com/huggingface/transformers/blob/main/src/transformers/activations.py#L90
328
329
330
331
332
333
334
335
    def __init__(self):
        super().__init__()
        if current_platform.is_cuda_alike() or current_platform.is_cpu():
            self.op = torch.ops._C.gelu_quick
        elif current_platform.is_xpu():
            from vllm._ipex_ops import ipex_ops
            self.op = ipex_ops.gelu_quick

336
337
338
339
340
341
    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
        """PyTorch-native implementation equivalent to forward()."""
        return x * torch.sigmoid(1.702 * x)

    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
        out = torch.empty_like(x)
342
        self.op(out, x)
343
344
        return out

345
346
    def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
        out = torch.empty_like(x)
347
        self.op(out, x)
348
349
        return out

350
351
352
    # TODO implement forward_xpu for QuickGELU
    # def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:

353

354
@CustomOp.register("relu2")
355
356
357
358
359
360
361
class ReLUSquaredActivation(CustomOp):
    """
    Applies the relu^2 activation introduced in https://arxiv.org/abs/2109.08668v2
    """

    def forward_native(self, x: torch.Tensor) -> torch.Tensor:
        """PyTorch-native implementation equivalent to forward()."""
362
        return torch.square(F.relu(x))
363
364

    def forward_cuda(self, x: torch.Tensor) -> torch.Tensor:
365
        #TODO : implement cuda kenrels
366
367
368
        return self.forward_native(x)


369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
@CustomOp.register("xielu")
class XIELU(CustomOp):
    """
    Applies the xIELU activation function introduced in https://arxiv.org/abs/2411.13010
    If the user has installed the nickjbrowning/XIELU, we import xIELU CUDA
    Otherwise, we emit a single warning and use xIELU Python
    """

    def __init__(
        self,
        alpha_p_init: float = 0.8,
        alpha_n_init: float = 0.8,
        beta: float = 0.5,
        eps: float = -1e-6,
        dtype: torch.dtype = torch.bfloat16,
        with_vector_loads: bool = False,
    ):
        super().__init__()
        self.alpha_p = nn.Parameter(
            torch.log(torch.exp(torch.tensor(alpha_p_init, dtype=dtype)) -
                      1).unsqueeze(0))
        self.alpha_n = nn.Parameter(
            torch.log(
                torch.exp(torch.tensor(alpha_n_init - beta, dtype=dtype)) -
                1).unsqueeze(0))
        self.register_buffer("beta", torch.tensor(beta, dtype=dtype))
        self.register_buffer("eps", torch.tensor(eps, dtype=dtype))
        self.with_vector_loads = with_vector_loads
        # Temporary until xIELU CUDA fully implemented
        self._beta_scalar = float(self.beta.detach().cpu().float().item())
        self._eps_scalar = float(self.eps.detach().cpu().float().item())

        self._xielu_cuda_obj = None
        try:
            import xielu.ops  # noqa: F401

            self._xielu_cuda_obj = torch.classes.xielu.XIELU()
            msg = "Using experimental xIELU CUDA."
            try:
                from torch._dynamo import allow_in_graph

                self._xielu_cuda_fn = allow_in_graph(self._xielu_cuda)
                msg += " Enabled torch._dynamo for xIELU CUDA."
            except Exception as err:
                msg += (f" Could not enable torch._dynamo for xIELU ({err}) - "
                        "this may result in slower performance.")
                self._xielu_cuda_fn = self._xielu_cuda
            logger.warning_once(msg)
        except Exception as err:
            logger.warning_once(
                "CUDA-fused xIELU not available (%s) –"
                " falling back to a Python version.\n"
                "For CUDA xIELU (experimental), `pip install git+https://github.com/nickjbrowning/XIELU`",
                str(err),
            )

    def _xielu_python(self, x: torch.Tensor) -> torch.Tensor:
        alpha_p = nn.functional.softplus(self.alpha_p)
        alpha_n = self.beta + nn.functional.softplus(self.alpha_n)
        return torch.where(
            x > 0,
            alpha_p * x * x + self.beta * x,
            (torch.expm1(torch.min(x, self.eps)) - x) * alpha_n +
            self.beta * x,
        )

    def _xielu_cuda(self, x: torch.Tensor) -> torch.Tensor:
        """Firewall function to prevent torch.compile from seeing .item()"""
        assert self._xielu_cuda_obj is not None, (
            "XIELU CUDA object must not be None")
        original_shape = x.shape
        # CUDA kernel expects 3D tensors, reshape if needed
        while x.dim() < 3:
            x = x.unsqueeze(0)
        if x.dim() > 3:
            x = x.view(-1, 1, x.size(-1))
        if original_shape != x.shape:
            logger.warning_once(
                "Warning: xIELU input tensor expects 3 dimensions"
                " but got (shape: %s). Reshaping to (shape: %s).",
                original_shape,
                x.shape,
            )
        result = self._xielu_cuda_obj.forward(
            x,
            self.alpha_p,
            self.alpha_n,
            # Temporary until xIELU CUDA fully implemented ->
            # self.{beta,eps}.item()
            self._beta_scalar,
            self._eps_scalar,
            self.with_vector_loads,
        )
        return result.view(original_shape)

    def forward(self, input: torch.Tensor) -> torch.Tensor:
        if self._xielu_cuda_obj is not None and input.is_cuda:
            if not torch._dynamo.is_compiling():
                return self._xielu_cuda_fn(input)
            else:
                logger.warning_once(
                    "torch._dynamo is compiling, using Python version of xIELU."
                )
        return self._xielu_python(input)


475
476
477
478
479
480
481
482
483
class ScaledActivation(nn.Module):
    """An activation function with post-scale parameters.

    This is used for some quantization methods like AWQ.
    """

    def __init__(
        self,
        act_module: nn.Module,
484
485
486
        intermediate_size: int,
        input_is_parallel: bool = True,
        params_dtype: Optional[torch.dtype] = None,
487
488
489
    ):
        super().__init__()
        self.act = act_module
490
        self.input_is_parallel = input_is_parallel
491
492
493
494
495
496
497
498
        if input_is_parallel:
            tp_size = get_tensor_model_parallel_world_size()
            intermediate_size_per_partition = divide(intermediate_size,
                                                     tp_size)
        else:
            intermediate_size_per_partition = intermediate_size
        if params_dtype is None:
            params_dtype = torch.get_default_dtype()
499
        self.scales = nn.Parameter(
500
            torch.empty(intermediate_size_per_partition, dtype=params_dtype))
501
        set_weight_attrs(self.scales, {"weight_loader": self.weight_loader})
502

503
    def forward(self, x: torch.Tensor) -> torch.Tensor:
504
505
        return self.act(x) / self.scales

506
507
    def weight_loader(self, param: nn.Parameter, loaded_weight: torch.Tensor):
        param_data = param.data
508
509
510
511
512
        if self.input_is_parallel:
            tp_rank = get_tensor_model_parallel_rank()
            shard_size = param_data.shape[0]
            start_idx = tp_rank * shard_size
            loaded_weight = loaded_weight.narrow(0, start_idx, shard_size)
513
514
515
        assert param_data.shape == loaded_weight.shape
        param_data.copy_(loaded_weight)

516

517
518
519
520
521
522
523
524
525
526
527
528
529
_ACTIVATION_REGISTRY = LazyDict({
    "gelu":
    lambda: nn.GELU(),
    "gelu_fast":
    lambda: FastGELU(),
    "gelu_new":
    lambda: NewGELU(),
    "gelu_pytorch_tanh":
    lambda: nn.GELU(approximate="tanh"),
    "relu":
    lambda: nn.ReLU(),
    "relu2":
    lambda: ReLUSquaredActivation(),
530
531
    "silu":
    lambda: nn.SiLU(),
532
533
    "quick_gelu":
    lambda: QuickGELU(),
534
535
536
537
    "tanh":
    lambda: nn.Tanh(),
    "sigmoid":
    lambda: nn.Sigmoid(),
538
539
    "xielu":
    lambda: XIELU(),
540
})
541
542


543
def get_act_fn(act_fn_name: str) -> nn.Module:
544
    """Get an activation function by name."""
545
    act_fn_name = act_fn_name.lower()
546
547
548
549
550
551
552

    if act_fn_name.startswith("torch.nn.modules."):
        activation_name = act_fn_name.split(".")[-1]
        if activation_name == "identity":
            return nn.Identity()
        act_fn_name = activation_name

553
554
555
556
    if act_fn_name not in _ACTIVATION_REGISTRY:
        raise ValueError(
            f"Activation function {act_fn_name!r} is not supported.")

557
    return _ACTIVATION_REGISTRY[act_fn_name]
558
559
560


_ACTIVATION_AND_MUL_REGISTRY = LazyDict({
561
562
563
564
565
566
567
568
    "gelu":
    lambda: GeluAndMul(),
    "silu":
    lambda: SiluAndMul(),
    "geglu":
    lambda: GeluAndMul(),
    "swigluoai":
    lambda *args, **kwargs: SwigluOAIAndMul(*args, **kwargs),
569
570
571
})


572
def get_act_and_mul_fn(act_fn_name: str) -> nn.Module:
573
574
575
576
577
578
    """Get an activation-and-mul (i.e. SiluAndMul) function by name."""
    act_fn_name = act_fn_name.lower()
    if act_fn_name not in _ACTIVATION_AND_MUL_REGISTRY:
        raise ValueError(
            f"Activation function {act_fn_name!r} is not supported.")

579
    return _ACTIVATION_AND_MUL_REGISTRY[act_fn_name]