File size: 7,451 Bytes
4dd73bb
e944e94
c3b3e31
 
 
 
 
 
 
 
 
 
 
4dd73bb
 
 
 
 
 
e944e94
015983b
e944e94
 
c3b3e31
 
 
 
 
 
 
 
e944e94
c3b3e31
 
 
 
 
 
 
 
 
 
 
 
e944e94
c3b3e31
e944e94
 
c3b3e31
 
 
 
4dd73bb
 
c3b3e31
 
 
4dd73bb
 
e944e94
c3b3e31
e944e94
4dd73bb
 
e944e94
 
015983b
 
 
 
e944e94
 
 
 
 
 
 
c3b3e31
 
 
e944e94
 
 
c3b3e31
e944e94
 
 
4dd73bb
 
 
e944e94
 
 
c3b3e31
4dd73bb
e944e94
 
 
 
 
4dd73bb
e944e94
4dd73bb
e944e94
c3b3e31
e944e94
 
 
 
4dd73bb
 
e944e94
 
 
 
 
 
4dd73bb
 
e944e94
 
 
 
4dd73bb
 
 
 
e944e94
 
4dd73bb
 
e944e94
 
 
4dd73bb
 
 
e944e94
4dd73bb
 
 
 
e944e94
4dd73bb
 
 
 
e944e94
4dd73bb
 
 
 
 
 
 
 
 
e944e94
4dd73bb
 
e944e94
4dd73bb
 
 
 
 
e944e94
 
4dd73bb
 
 
 
e944e94
 
 
 
4dd73bb
e944e94
4dd73bb
 
e944e94
 
 
4dd73bb
e944e94
 
4dd73bb
e944e94
4dd73bb
e944e94
4dd73bb
c3b3e31
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
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
"""
Paired with a good language model. Thanks!

ZeroGPU note:
FlashAttention-3 (kernels-community/vllm-flash-attn3) has no binary for
Blackwell (sm_120), which is what ZeroGPU currently attaches. Loading the
kernel at import time under CUDA *emulation* reports a fake capability, so
FA3 gets enabled and then crashes on the real GPU with:

    CUDA error: no kernel image is available for execution on the device

Kernel init is therefore deferred until a real CUDA device is present, and
sm_120+ always falls back to PyTorch SDPA.
"""

import torch
from typing import Optional, Tuple
from diffusers.models.transformers.transformer_qwenimage import apply_rotary_emb_qwen

import torch.nn.functional as F

_flash_attn_func = None
_kernels_err = None
_fa3_inited = False


def _ensure_fa3() -> None:
    global _flash_attn_func, _kernels_err, _fa3_inited
    if _fa3_inited:
        return
    _fa3_inited = True

    if not torch.cuda.is_available():
        _kernels_err = RuntimeError("CUDA not available; using SDPA fallback.")
        return

    try:
        cap = torch.cuda.get_device_capability()
    except Exception as e:
        _kernels_err = e
        return

    # Blackwell (sm_120+) is not supported by the vllm-flash-attn3 kernel binary.
    if cap >= (12, 0):
        _kernels_err = RuntimeError(
            f"GPU compute capability sm_{cap[0]}{cap[1]} (Blackwell+) is not supported "
            "by the kernels-community/vllm-flash-attn3 binary; using SDPA fallback."
        )
        return

    try:
        from kernels import get_kernel
        _k = get_kernel("kernels-community/vllm-flash-attn3")
        _flash_attn_func = _k.flash_attn_func
    except Exception as e:
        _flash_attn_func = None
        _kernels_err = e


def _fa3_available() -> bool:
    _ensure_fa3()
    return _flash_attn_func is not None


def _sdpa_fallback(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor) -> torch.Tensor:
    # q/k/v: (B, S, H, D_h) → SDPA expects (B, H, S, D_h)
    q = q.transpose(1, 2)
    k = k.transpose(1, 2)
    v = v.transpose(1, 2)
    out = F.scaled_dot_product_attention(q, k, v, is_causal=False)
    return out.transpose(1, 2)  # back to (B, S, H, D_h)


@torch.library.custom_op("flash::flash_attn_func", mutates_args=())
def flash_attn_func(
    q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, causal: bool = False
) -> torch.Tensor:
    _ensure_fa3()
    if _flash_attn_func is None:
        return _sdpa_fallback(q, k, v)
    outputs, lse = _flash_attn_func(q, k, v, causal=causal)
    return outputs


@flash_attn_func.register_fake
def _(q, k, v, **kwargs):
    return torch.empty_like(q).contiguous()


class QwenDoubleStreamAttnProcessorFA3:
    """
    FA3-based attention processor for Qwen double-stream architecture.
    Computes joint attention over concatenated [text, image] streams using vLLM FlashAttention-3
    accessed via Hugging Face `kernels`, with an automatic SDPA fallback.

    Notes / limitations:
    - General attention masks are not supported here (FA3 path). `is_causal=False` and no arbitrary mask.
    - Optional windowed attention / sink tokens / softcap can be plumbed through if you use those features.
    - Expects an available `apply_rotary_emb_qwen` in scope (same as your non-FA3 processor).
    """

    _attention_backend = "fa3"  # for parity with your other processors, not used internally

    def __init__(self):
        _ensure_fa3()
        if not _fa3_available():
            print(f"[QwenDoubleStreamAttnProcessorFA3] FA3 unavailable, using SDPA fallback. Reason: {_kernels_err}")

    @torch.no_grad()
    def __call__(
        self,
        attn,  # Attention module with to_q/to_k/to_v/add_*_proj, norms, to_out, to_add_out, and .heads
        hidden_states: torch.FloatTensor,                 # (B, S_img, D_model)  image stream
        encoder_hidden_states: torch.FloatTensor = None,  # (B, S_txt, D_model)  text stream
        encoder_hidden_states_mask: torch.FloatTensor = None,  # unused in FA3 path
        attention_mask: Optional[torch.FloatTensor] = None,    # unused in FA3 path
        image_rotary_emb: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  # (img_freqs, txt_freqs)
    ) -> Tuple[torch.FloatTensor, torch.FloatTensor]:
        if encoder_hidden_states is None:
            raise ValueError("QwenDoubleStreamAttnProcessorFA3 requires encoder_hidden_states (text stream).")
        if attention_mask is not None:
            # FA3 kernel path here does not consume arbitrary masks; fail fast to avoid silent correctness issues.
            raise NotImplementedError("attention_mask is not supported in this FA3 implementation.")

        B, S_img, _ = hidden_states.shape
        S_txt = encoder_hidden_states.shape[1]

        # ---- QKV projections (image/sample stream) ----
        img_q = attn.to_q(hidden_states)   # (B, S_img, D)
        img_k = attn.to_k(hidden_states)
        img_v = attn.to_v(hidden_states)

        # ---- QKV projections (text/context stream) ----
        txt_q = attn.add_q_proj(encoder_hidden_states)  # (B, S_txt, D)
        txt_k = attn.add_k_proj(encoder_hidden_states)
        txt_v = attn.add_v_proj(encoder_hidden_states)

        # ---- Reshape to (B, S, H, D_h) ----
        H = attn.heads
        img_q = img_q.unflatten(-1, (H, -1))
        img_k = img_k.unflatten(-1, (H, -1))
        img_v = img_v.unflatten(-1, (H, -1))

        txt_q = txt_q.unflatten(-1, (H, -1))
        txt_k = txt_k.unflatten(-1, (H, -1))
        txt_v = txt_v.unflatten(-1, (H, -1))

        # ---- Q/K normalization (per your module contract) ----
        if getattr(attn, "norm_q", None) is not None:
            img_q = attn.norm_q(img_q)
        if getattr(attn, "norm_k", None) is not None:
            img_k = attn.norm_k(img_k)
        if getattr(attn, "norm_added_q", None) is not None:
            txt_q = attn.norm_added_q(txt_q)
        if getattr(attn, "norm_added_k", None) is not None:
            txt_k = attn.norm_added_k(txt_k)

        # ---- RoPE (Qwen variant) ----
        if image_rotary_emb is not None:
            img_freqs, txt_freqs = image_rotary_emb
            # expects tensors shaped (B, S, H, D_h)
            img_q = apply_rotary_emb_qwen(img_q, img_freqs, use_real=False)
            img_k = apply_rotary_emb_qwen(img_k, img_freqs, use_real=False)
            txt_q = apply_rotary_emb_qwen(txt_q, txt_freqs, use_real=False)
            txt_k = apply_rotary_emb_qwen(txt_k, txt_freqs, use_real=False)

        # ---- Joint attention over [text, image] along sequence axis ----
        # Shapes: (B, S_total, H, D_h)
        q = torch.cat([txt_q, img_q], dim=1)
        k = torch.cat([txt_k, img_k], dim=1)
        v = torch.cat([txt_v, img_v], dim=1)

        if _fa3_available():
            out = flash_attn_func(q, k, v, causal=False)  # out: (B, S_total, H, D_h)
        else:
            out = _sdpa_fallback(q, k, v)  # out: (B, S_total, H, D_h)

        # ---- Back to (B, S, D_model) ----
        out = out.flatten(2, 3).to(q.dtype)

        # Split back to text / image segments
        txt_attn_out = out[:, :S_txt, :]
        img_attn_out = out[:, S_txt:, :]

        # ---- Output projections ----
        img_attn_out = attn.to_out[0](img_attn_out)
        if len(attn.to_out) > 1:
            img_attn_out = attn.to_out[1](img_attn_out)  # dropout if present

        txt_attn_out = attn.to_add_out(txt_attn_out)

        return img_attn_out, txt_attn_out