import math import torch import torch.nn.functional as F from torch import nn, einsum from einops import rearrange, repeat from typing import Optional, Any import logging from .diffusionmodules.util import checkpoint, AlphaBlender, timestep_embedding from .sub_quadratic_attention import efficient_dot_product_attention from comfy import model_management if model_management.xformers_enabled(): import xformers import xformers.ops from comfy.cli_args import args import comfy.ops ops = comfy.ops.disable_weight_init # CrossAttn precision handling if args.dont_upcast_attention: logging.info("disabling upcasting of attention") _ATTN_PRECISION = "fp16" else: _ATTN_PRECISION = "fp32" def exists(val): return val is not None def uniq(arr): return{el: True for el in arr}.keys() def default(val, d): if exists(val): return val return d def max_neg_value(t): return -torch.finfo(t.dtype).max def init_(tensor): dim = tensor.shape[-1] std = 1 / math.sqrt(dim) tensor.uniform_(-std, std) return tensor # feedforward class GEGLU(nn.Module): def __init__(self, dim_in, dim_out, dtype=None, device=None, operations=ops): super().__init__() self.proj = operations.Linear(dim_in, dim_out * 2, dtype=dtype, device=device) def forward(self, x): x, gate = self.proj(x).chunk(2, dim=-1) return x * F.gelu(gate) class FeedForward(nn.Module): def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0., dtype=None, device=None, operations=ops): super().__init__() inner_dim = int(dim * mult) dim_out = default(dim_out, dim) project_in = nn.Sequential( operations.Linear(dim, inner_dim, dtype=dtype, device=device), nn.GELU() ) if not glu else GEGLU(dim, inner_dim, dtype=dtype, device=device, operations=operations) self.net = nn.Sequential( project_in, nn.Dropout(dropout), operations.Linear(inner_dim, dim_out, dtype=dtype, device=device) ) def forward(self, x): return self.net(x) def Normalize(in_channels, dtype=None, device=None): return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True, dtype=dtype, device=device) def attention_basic(q, k, v, heads, mask=None): b, _, dim_head = q.shape dim_head //= heads scale = dim_head ** -0.5 h = heads q, k, v = map( lambda t: t.unsqueeze(3) .reshape(b, -1, heads, dim_head) .permute(0, 2, 1, 3) .reshape(b * heads, -1, dim_head) .contiguous(), (q, k, v), ) # force cast to fp32 to avoid overflowing if _ATTN_PRECISION =="fp32": sim = einsum('b i d, b j d -> b i j', q.float(), k.float()) * scale else: sim = einsum('b i d, b j d -> b i j', q, k) * scale del q, k if exists(mask): if mask.dtype == torch.bool: mask = rearrange(mask, 'b ... -> b (...)') #TODO: check if this bool part matches pytorch attention max_neg_value = -torch.finfo(sim.dtype).max mask = repeat(mask, 'b j -> (b h) () j', h=h) sim.masked_fill_(~mask, max_neg_value) else: if len(mask.shape) == 2: bs = 1 else: bs = mask.shape[0] mask = mask.reshape(bs, -1, mask.shape[-2], mask.shape[-1]).expand(b, heads, -1, -1).reshape(-1, mask.shape[-2], mask.shape[-1]) sim.add_(mask) # attention, what we cannot get enough of sim = sim.softmax(dim=-1) out = einsum('b i j, b j d -> b i d', sim.to(v.dtype), v) out = ( out.unsqueeze(0) .reshape(b, heads, -1, dim_head) .permute(0, 2, 1, 3) .reshape(b, -1, heads * dim_head) ) return out def attention_sub_quad(query, key, value, heads, mask=None): b, _, dim_head = query.shape dim_head //= heads scale = dim_head ** -0.5 query = query.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head) value = value.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head) key = key.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 3, 1).reshape(b * heads, dim_head, -1) dtype = query.dtype upcast_attention = _ATTN_PRECISION =="fp32" and query.dtype != torch.float32 if upcast_attention: bytes_per_token = torch.finfo(torch.float32).bits//8 else: bytes_per_token = torch.finfo(query.dtype).bits//8 batch_x_heads, q_tokens, _ = query.shape _, _, k_tokens = key.shape qk_matmul_size_bytes = batch_x_heads * bytes_per_token * q_tokens * k_tokens mem_free_total, mem_free_torch = model_management.get_free_memory(query.device, True) kv_chunk_size_min = None kv_chunk_size = None query_chunk_size = None for x in [4096, 2048, 1024, 512, 256]: count = mem_free_total / (batch_x_heads * bytes_per_token * x * 4.0) if count >= k_tokens: kv_chunk_size = k_tokens query_chunk_size = x break if query_chunk_size is None: query_chunk_size = 512 if mask is not None: if len(mask.shape) == 2: bs = 1 else: bs = mask.shape[0] mask = mask.reshape(bs, -1, mask.shape[-2], mask.shape[-1]).expand(b, heads, -1, -1).reshape(-1, mask.shape[-2], mask.shape[-1]) hidden_states = efficient_dot_product_attention( query, key, value, query_chunk_size=query_chunk_size, kv_chunk_size=kv_chunk_size, kv_chunk_size_min=kv_chunk_size_min, use_checkpoint=False, upcast_attention=upcast_attention, mask=mask, ) hidden_states = hidden_states.to(dtype) hidden_states = hidden_states.unflatten(0, (-1, heads)).transpose(1,2).flatten(start_dim=2) return hidden_states def attention_split(q, k, v, heads, mask=None): b, _, dim_head = q.shape dim_head //= heads scale = dim_head ** -0.5 h = heads q, k, v = map( lambda t: t.unsqueeze(3) .reshape(b, -1, heads, dim_head) .permute(0, 2, 1, 3) .reshape(b * heads, -1, dim_head) .contiguous(), (q, k, v), ) r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device, dtype=q.dtype) mem_free_total = model_management.get_free_memory(q.device) if _ATTN_PRECISION =="fp32": element_size = 4 else: element_size = q.element_size() gb = 1024 ** 3 tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * element_size modifier = 3 mem_required = tensor_size * modifier steps = 1 if mem_required > mem_free_total: steps = 2**(math.ceil(math.log(mem_required / mem_free_total, 2))) # print(f"Expected tensor size:{tensor_size/gb:0.1f}GB, cuda free:{mem_free_cuda/gb:0.1f}GB " # f"torch free:{mem_free_torch/gb:0.1f} total:{mem_free_total/gb:0.1f} steps:{steps}") if steps > 64: max_res = math.floor(math.sqrt(math.sqrt(mem_free_total / 2.5)) / 8) * 64 raise RuntimeError(f'Not enough memory, use lower resolution (max approx. {max_res}x{max_res}). ' f'Need: {mem_required/64/gb:0.1f}GB free, Have:{mem_free_total/gb:0.1f}GB free') if mask is not None: if len(mask.shape) == 2: bs = 1 else: bs = mask.shape[0] mask = mask.reshape(bs, -1, mask.shape[-2], mask.shape[-1]).expand(b, heads, -1, -1).reshape(-1, mask.shape[-2], mask.shape[-1]) # print("steps", steps, mem_required, mem_free_total, modifier, q.element_size(), tensor_size) first_op_done = False cleared_cache = False while True: try: slice_size = q.shape[1] // steps if (q.shape[1] % steps) == 0 else q.shape[1] for i in range(0, q.shape[1], slice_size): end = i + slice_size if _ATTN_PRECISION =="fp32": with torch.autocast(enabled=False, device_type = 'cuda'): s1 = einsum('b i d, b j d -> b i j', q[:, i:end].float(), k.float()) * scale else: s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k) * scale if mask is not None: if len(mask.shape) == 2: s1 += mask[i:end] else: s1 += mask[:, i:end] s2 = s1.softmax(dim=-1).to(v.dtype) del s1 first_op_done = True r1[:, i:end] = einsum('b i j, b j d -> b i d', s2, v) del s2 break except model_management.OOM_EXCEPTION as e: if first_op_done == False: model_management.soft_empty_cache(True) if cleared_cache == False: cleared_cache = True logging.warning("out of memory error, emptying cache and trying again") continue steps *= 2 if steps > 64: raise e logging.warning("out of memory error, increasing steps and trying again {}".format(steps)) else: raise e del q, k, v r1 = ( r1.unsqueeze(0) .reshape(b, heads, -1, dim_head) .permute(0, 2, 1, 3) .reshape(b, -1, heads * dim_head) ) return r1 BROKEN_XFORMERS = False try: x_vers = xformers.__version__ #I think 0.0.23 is also broken (q with bs bigger than 65535 gives CUDA error) BROKEN_XFORMERS = x_vers.startswith("0.0.21") or x_vers.startswith("0.0.22") or x_vers.startswith("0.0.23") except: pass def attention_xformers(q, k, v, heads, mask=None): b, _, dim_head = q.shape dim_head //= heads if BROKEN_XFORMERS: if b * heads > 65535: return attention_pytorch(q, k, v, heads, mask) q, k, v = map( lambda t: t.unsqueeze(3) .reshape(b, -1, heads, dim_head) .permute(0, 2, 1, 3) .reshape(b * heads, -1, dim_head) .contiguous(), (q, k, v), ) if mask is not None: pad = 8 - q.shape[1] % 8 mask_out = torch.empty([q.shape[0], q.shape[1], q.shape[1] + pad], dtype=q.dtype, device=q.device) mask_out[:, :, :mask.shape[-1]] = mask mask = mask_out[:, :, :mask.shape[-1]] out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask) out = ( out.unsqueeze(0) .reshape(b, heads, -1, dim_head) .permute(0, 2, 1, 3) .reshape(b, -1, heads * dim_head) ) return out def attention_pytorch(q, k, v, heads, mask=None): b, _, dim_head = q.shape dim_head //= heads q, k, v = map( lambda t: t.view(b, -1, heads, dim_head).transpose(1, 2), (q, k, v), ) out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False) out = ( out.transpose(1, 2).reshape(b, -1, heads * dim_head) ) return out optimized_attention = attention_basic if model_management.xformers_enabled(): logging.info("Using xformers cross attention") optimized_attention = attention_xformers elif model_management.pytorch_attention_enabled(): logging.info("Using pytorch cross attention") optimized_attention = attention_pytorch else: if args.use_split_cross_attention: logging.info("Using split optimization for cross attention") optimized_attention = attention_split else: logging.info("Using sub quadratic optimization for cross attention, if you have memory or speed issues try using: --use-split-cross-attention") optimized_attention = attention_sub_quad optimized_attention_masked = optimized_attention def optimized_attention_for_device(device, mask=False, small_input=False): if small_input: if model_management.pytorch_attention_enabled(): return attention_pytorch #TODO: need to confirm but this is probably slightly faster for small inputs in all cases else: return attention_basic if device == torch.device("cpu"): return attention_sub_quad if mask: return optimized_attention_masked return optimized_attention class CrossAttention(nn.Module): def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0., dtype=None, device=None, operations=ops): super().__init__() inner_dim = dim_head * heads context_dim = default(context_dim, query_dim) self.heads = heads self.dim_head = dim_head self.to_q = operations.Linear(query_dim, inner_dim, bias=False, dtype=dtype, device=device) self.to_k = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device) self.to_v = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device) self.to_out = nn.Sequential(operations.Linear(inner_dim, query_dim, dtype=dtype, device=device), nn.Dropout(dropout)) def forward(self, x, context=None, value=None, mask=None): q = self.to_q(x) context = default(context, x) k = self.to_k(context) if value is not None: v = self.to_v(value) del value else: v = self.to_v(context) if mask is None: out = optimized_attention(q, k, v, self.heads) else: out = optimized_attention_masked(q, k, v, self.heads, mask) return self.to_out(out) class BasicTransformerBlock(nn.Module): def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True, ff_in=False, inner_dim=None, disable_self_attn=False, disable_temporal_crossattention=False, switch_temporal_ca_to_sa=False, dtype=None, device=None, operations=ops): super().__init__() self.ff_in = ff_in or inner_dim is not None if inner_dim is None: inner_dim = dim self.is_res = inner_dim == dim if self.ff_in: self.norm_in = operations.LayerNorm(dim, dtype=dtype, device=device) self.ff_in = FeedForward(dim, dim_out=inner_dim, dropout=dropout, glu=gated_ff, dtype=dtype, device=device, operations=operations) self.disable_self_attn = disable_self_attn self.attn1 = CrossAttention(query_dim=inner_dim, heads=n_heads, dim_head=d_head, dropout=dropout, context_dim=context_dim if self.disable_self_attn else None, dtype=dtype, device=device, operations=operations) # is a self-attention if not self.disable_self_attn self.ff = FeedForward(inner_dim, dim_out=dim, dropout=dropout, glu=gated_ff, dtype=dtype, device=device, operations=operations) if disable_temporal_crossattention: if switch_temporal_ca_to_sa: raise ValueError else: self.attn2 = None else: context_dim_attn2 = None if not switch_temporal_ca_to_sa: context_dim_attn2 = context_dim self.attn2 = CrossAttention(query_dim=inner_dim, context_dim=context_dim_attn2, heads=n_heads, dim_head=d_head, dropout=dropout, dtype=dtype, device=device, operations=operations) # is self-attn if context is none self.norm2 = operations.LayerNorm(inner_dim, dtype=dtype, device=device) self.norm1 = operations.LayerNorm(inner_dim, dtype=dtype, device=device) self.norm3 = operations.LayerNorm(inner_dim, dtype=dtype, device=device) self.checkpoint = checkpoint self.n_heads = n_heads self.d_head = d_head self.switch_temporal_ca_to_sa = switch_temporal_ca_to_sa def forward(self, x, context=None, transformer_options={}): return checkpoint(self._forward, (x, context, transformer_options), self.parameters(), self.checkpoint) def _forward(self, x, context=None, transformer_options={}): extra_options = {} block = transformer_options.get("block", None) block_index = transformer_options.get("block_index", 0) transformer_patches = {} transformer_patches_replace = {} for k in transformer_options: if k == "patches": transformer_patches = transformer_options[k] elif k == "patches_replace": transformer_patches_replace = transformer_options[k] else: extra_options[k] = transformer_options[k] extra_options["n_heads"] = self.n_heads extra_options["dim_head"] = self.d_head if self.ff_in: x_skip = x x = self.ff_in(self.norm_in(x)) if self.is_res: x += x_skip n = self.norm1(x) if self.disable_self_attn: context_attn1 = context else: context_attn1 = None value_attn1 = None if "attn1_patch" in transformer_patches: patch = transformer_patches["attn1_patch"] if context_attn1 is None: context_attn1 = n value_attn1 = context_attn1 for p in patch: n, context_attn1, value_attn1 = p(n, context_attn1, value_attn1, extra_options) if block is not None: transformer_block = (block[0], block[1], block_index) else: transformer_block = None attn1_replace_patch = transformer_patches_replace.get("attn1", {}) block_attn1 = transformer_block if block_attn1 not in attn1_replace_patch: block_attn1 = block if block_attn1 in attn1_replace_patch: if context_attn1 is None: context_attn1 = n value_attn1 = n n = self.attn1.to_q(n) context_attn1 = self.attn1.to_k(context_attn1) value_attn1 = self.attn1.to_v(value_attn1) n = attn1_replace_patch[block_attn1](n, context_attn1, value_attn1, extra_options) n = self.attn1.to_out(n) else: n = self.attn1(n, context=context_attn1, value=value_attn1) if "attn1_output_patch" in transformer_patches: patch = transformer_patches["attn1_output_patch"] for p in patch: n = p(n, extra_options) x += n if "middle_patch" in transformer_patches: patch = transformer_patches["middle_patch"] for p in patch: x = p(x, extra_options) if self.attn2 is not None: n = self.norm2(x) if self.switch_temporal_ca_to_sa: context_attn2 = n else: context_attn2 = context value_attn2 = None if "attn2_patch" in transformer_patches: patch = transformer_patches["attn2_patch"] value_attn2 = context_attn2 for p in patch: n, context_attn2, value_attn2 = p(n, context_attn2, value_attn2, extra_options) attn2_replace_patch = transformer_patches_replace.get("attn2", {}) block_attn2 = transformer_block if block_attn2 not in attn2_replace_patch: block_attn2 = block if block_attn2 in attn2_replace_patch: if value_attn2 is None: value_attn2 = context_attn2 n = self.attn2.to_q(n) context_attn2 = self.attn2.to_k(context_attn2) value_attn2 = self.attn2.to_v(value_attn2) n = attn2_replace_patch[block_attn2](n, context_attn2, value_attn2, extra_options) n = self.attn2.to_out(n) else: n = self.attn2(n, context=context_attn2, value=value_attn2) if "attn2_output_patch" in transformer_patches: patch = transformer_patches["attn2_output_patch"] for p in patch: n = p(n, extra_options) x += n if self.is_res: x_skip = x x = self.ff(self.norm3(x)) if self.is_res: x += x_skip return x class SpatialTransformer(nn.Module): """ Transformer block for image-like data. First, project the input (aka embedding) and reshape to b, t, d. Then apply standard transformer action. Finally, reshape to image NEW: use_linear for more efficiency instead of the 1x1 convs """ def __init__(self, in_channels, n_heads, d_head, depth=1, dropout=0., context_dim=None, disable_self_attn=False, use_linear=False, use_checkpoint=True, dtype=None, device=None, operations=ops): super().__init__() if exists(context_dim) and not isinstance(context_dim, list): context_dim = [context_dim] * depth self.in_channels = in_channels inner_dim = n_heads * d_head self.norm = operations.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True, dtype=dtype, device=device) if not use_linear: self.proj_in = operations.Conv2d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0, dtype=dtype, device=device) else: self.proj_in = operations.Linear(in_channels, inner_dim, dtype=dtype, device=device) self.transformer_blocks = nn.ModuleList( [BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d], disable_self_attn=disable_self_attn, checkpoint=use_checkpoint, dtype=dtype, device=device, operations=operations) for d in range(depth)] ) if not use_linear: self.proj_out = operations.Conv2d(inner_dim,in_channels, kernel_size=1, stride=1, padding=0, dtype=dtype, device=device) else: self.proj_out = operations.Linear(in_channels, inner_dim, dtype=dtype, device=device) self.use_linear = use_linear def forward(self, x, context=None, transformer_options={}): # note: if no context is given, cross-attention defaults to self-attention if not isinstance(context, list): context = [context] * len(self.transformer_blocks) b, c, h, w = x.shape x_in = x x = self.norm(x) if not self.use_linear: x = self.proj_in(x) x = rearrange(x, 'b c h w -> b (h w) c').contiguous() if self.use_linear: x = self.proj_in(x) for i, block in enumerate(self.transformer_blocks): transformer_options["block_index"] = i x = block(x, context=context[i], transformer_options=transformer_options) if self.use_linear: x = self.proj_out(x) x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w).contiguous() if not self.use_linear: x = self.proj_out(x) return x + x_in class SpatialVideoTransformer(SpatialTransformer): def __init__( self, in_channels, n_heads, d_head, depth=1, dropout=0.0, use_linear=False, context_dim=None, use_spatial_context=False, timesteps=None, merge_strategy: str = "fixed", merge_factor: float = 0.5, time_context_dim=None, ff_in=False, checkpoint=False, time_depth=1, disable_self_attn=False, disable_temporal_crossattention=False, max_time_embed_period: int = 10000, dtype=None, device=None, operations=ops ): super().__init__( in_channels, n_heads, d_head, depth=depth, dropout=dropout, use_checkpoint=checkpoint, context_dim=context_dim, use_linear=use_linear, disable_self_attn=disable_self_attn, dtype=dtype, device=device, operations=operations ) self.time_depth = time_depth self.depth = depth self.max_time_embed_period = max_time_embed_period time_mix_d_head = d_head n_time_mix_heads = n_heads time_mix_inner_dim = int(time_mix_d_head * n_time_mix_heads) inner_dim = n_heads * d_head if use_spatial_context: time_context_dim = context_dim self.time_stack = nn.ModuleList( [ BasicTransformerBlock( inner_dim, n_time_mix_heads, time_mix_d_head, dropout=dropout, context_dim=time_context_dim, # timesteps=timesteps, checkpoint=checkpoint, ff_in=ff_in, inner_dim=time_mix_inner_dim, disable_self_attn=disable_self_attn, disable_temporal_crossattention=disable_temporal_crossattention, dtype=dtype, device=device, operations=operations ) for _ in range(self.depth) ] ) assert len(self.time_stack) == len(self.transformer_blocks) self.use_spatial_context = use_spatial_context self.in_channels = in_channels time_embed_dim = self.in_channels * 4 self.time_pos_embed = nn.Sequential( operations.Linear(self.in_channels, time_embed_dim, dtype=dtype, device=device), nn.SiLU(), operations.Linear(time_embed_dim, self.in_channels, dtype=dtype, device=device), ) self.time_mixer = AlphaBlender( alpha=merge_factor, merge_strategy=merge_strategy ) def forward( self, x: torch.Tensor, context: Optional[torch.Tensor] = None, time_context: Optional[torch.Tensor] = None, timesteps: Optional[int] = None, image_only_indicator: Optional[torch.Tensor] = None, transformer_options={} ) -> torch.Tensor: _, _, h, w = x.shape x_in = x spatial_context = None if exists(context): spatial_context = context if self.use_spatial_context: assert ( context.ndim == 3 ), f"n dims of spatial context should be 3 but are {context.ndim}" if time_context is None: time_context = context time_context_first_timestep = time_context[::timesteps] time_context = repeat( time_context_first_timestep, "b ... -> (b n) ...", n=h * w ) elif time_context is not None and not self.use_spatial_context: time_context = repeat(time_context, "b ... -> (b n) ...", n=h * w) if time_context.ndim == 2: time_context = rearrange(time_context, "b c -> b 1 c") x = self.norm(x) if not self.use_linear: x = self.proj_in(x) x = rearrange(x, "b c h w -> b (h w) c") if self.use_linear: x = self.proj_in(x) num_frames = torch.arange(timesteps, device=x.device) num_frames = repeat(num_frames, "t -> b t", b=x.shape[0] // timesteps) num_frames = rearrange(num_frames, "b t -> (b t)") t_emb = timestep_embedding(num_frames, self.in_channels, repeat_only=False, max_period=self.max_time_embed_period).to(x.dtype) emb = self.time_pos_embed(t_emb) emb = emb[:, None, :] for it_, (block, mix_block) in enumerate( zip(self.transformer_blocks, self.time_stack) ): transformer_options["block_index"] = it_ x = block( x, context=spatial_context, transformer_options=transformer_options, ) x_mix = x x_mix = x_mix + emb B, S, C = x_mix.shape x_mix = rearrange(x_mix, "(b t) s c -> (b s) t c", t=timesteps) x_mix = mix_block(x_mix, context=time_context) #TODO: transformer_options x_mix = rearrange( x_mix, "(b s) t c -> (b t) s c", s=S, b=B // timesteps, c=C, t=timesteps ) x = self.time_mixer(x_spatial=x, x_temporal=x_mix, image_only_indicator=image_only_indicator) if self.use_linear: x = self.proj_out(x) x = rearrange(x, "b (h w) c -> b c h w", h=h, w=w) if not self.use_linear: x = self.proj_out(x) out = x + x_in return out