diff --git a/comfy/ldm/lens/model.py b/comfy/ldm/lens/model.py new file mode 100644 index 000000000..d609507c4 --- /dev/null +++ b/comfy/ldm/lens/model.py @@ -0,0 +1,578 @@ +"""Lens denoising transformer (DiT)""" + +from __future__ import annotations + +from typing import Any, Dict, List, Optional, Tuple, Union + +import torch +import torch.nn as nn +import torch.nn.functional as F + +import comfy.ldm.flux.layers +import comfy.patcher_extension +from comfy.ldm.modules.attention import optimized_attention + + +def _lens_time_proj(t: torch.Tensor, dim: int = 256) -> torch.Tensor: + return comfy.ldm.flux.layers.timestep_embedding(t, dim) + + +def apply_rotary_emb_lens(x: torch.Tensor, freqs_cis: torch.Tensor) -> torch.Tensor: + x_complex = torch.view_as_complex(x.float().reshape(*x.shape[:-1], -1, 2)) + freqs_cis = freqs_cis.unsqueeze(1) + x_out = torch.view_as_real(x_complex * freqs_cis).flatten(3) + return x_out.type_as(x) + + +class LensEmbedRope(nn.Module): + """Frame/H/W axial RoPE shared between image and text streams.""" + + def __init__(self, theta: int = 10000, axes_dim=(8, 28, 28), scale_rope: bool = True) -> None: + super().__init__() + self.theta = theta + self.axes_dim = list(axes_dim) + self.scale_rope = scale_rope + pos_index = torch.arange(4096) + neg_index = torch.arange(4096).flip(0) * -1 - 1 + # Plain attributes (not buffers): register_buffer strips complex imag. + self.pos_freqs = torch.cat( + [self._rope_params(pos_index, d, theta) for d in self.axes_dim], dim=1 + ) + self.neg_freqs = torch.cat( + [self._rope_params(neg_index, d, theta) for d in self.axes_dim], dim=1 + ) + self.rope_cache: Dict[str, torch.Tensor] = {} + + @staticmethod + def _rope_params(index: torch.Tensor, dim: int, theta: int = 10000) -> torch.Tensor: + assert dim % 2 == 0 + freqs = torch.outer( + index, 1.0 / torch.pow(theta, torch.arange(0, dim, 2).float().div(dim)) + ) + return torch.polar(torch.ones_like(freqs), freqs) + + def forward( + self, + video_fhw: Union[List[Tuple[int, int, int]], Tuple[int, int, int]], + txt_seq_lens: Union[List[int], int], + device: torch.device, + ) -> Tuple[torch.Tensor, torch.Tensor]: + if self.pos_freqs.device != device: + self.pos_freqs = self.pos_freqs.to(device) + self.neg_freqs = self.neg_freqs.to(device) + + if isinstance(video_fhw, list): + video_fhw = video_fhw[0] + if not isinstance(video_fhw, list): + video_fhw = [video_fhw] + if not isinstance(txt_seq_lens, list): + txt_seq_lens = [txt_seq_lens] + assert len(video_fhw) == 1, "video_fhw must have length 1" + + vid_freqs = [] + max_vid_index = 0 + for idx, fhw in enumerate(video_fhw): + frame, height, width = fhw + rope_key = f"{idx}_{height}_{width}_{device}" + video_freq = self.rope_cache.get(rope_key) + if video_freq is None: + video_freq = self._compute_video_freqs(frame, height, width, idx=0).to(device) + self.rope_cache[rope_key] = video_freq + if self.scale_rope: + max_vid_index = max(height // 2, width // 2, max_vid_index) + else: + max_vid_index = max(height, width, max_vid_index) + vid_freqs.append(video_freq) + + max_len = max(txt_seq_lens) + txt_freqs = self.pos_freqs[max_vid_index : max_vid_index + max_len, ...] + return torch.cat(vid_freqs, dim=0), txt_freqs + + def _compute_video_freqs(self, frame: int, height: int, width: int, idx: int = 0) -> torch.Tensor: + seq_lens = frame * height * width + freqs_pos = self.pos_freqs.split([d // 2 for d in self.axes_dim], dim=1) + freqs_neg = self.neg_freqs.split([d // 2 for d in self.axes_dim], dim=1) + + freqs_frame = freqs_pos[0][idx : idx + frame].view(frame, 1, 1, -1).expand(frame, height, width, -1) + if self.scale_rope: + freqs_height = torch.cat( + [freqs_neg[1][-(height - height // 2) :], freqs_pos[1][: height // 2]], dim=0 + ).view(1, height, 1, -1).expand(frame, height, width, -1) + freqs_width = torch.cat( + [freqs_neg[2][-(width - width // 2) :], freqs_pos[2][: width // 2]], dim=0 + ).view(1, 1, width, -1).expand(frame, height, width, -1) + else: + freqs_height = freqs_pos[1][:height].view(1, height, 1, -1).expand(frame, height, width, -1) + freqs_width = freqs_pos[2][:width].view(1, 1, width, -1).expand(frame, height, width, -1) + + freqs = torch.cat([freqs_frame, freqs_height, freqs_width], dim=-1).reshape(seq_lens, -1) + return freqs.clone().contiguous() + + +class _TimestepEmbedder(nn.Module): + def __init__(self, in_channels: int, time_embed_dim: int, dtype=None, device=None, operations=None) -> None: + super().__init__() + self.linear_1 = operations.Linear(in_channels, time_embed_dim, dtype=dtype, device=device) + self.linear_2 = operations.Linear(time_embed_dim, time_embed_dim, dtype=dtype, device=device) + + def forward(self, x: torch.Tensor) -> torch.Tensor: + x = self.linear_1(x) + x = F.silu(x) + return self.linear_2(x) + + +class LensTimestepProjEmbeddings(nn.Module): + def __init__(self, embedding_dim: int, dtype=None, device=None, operations=None) -> None: + super().__init__() + self.timestep_embedder = _TimestepEmbedder(256, embedding_dim, dtype=dtype, device=device, operations=operations) + + def forward(self, timestep: torch.Tensor, hidden_states: torch.Tensor) -> torch.Tensor: + proj = _lens_time_proj(timestep, 256) + return self.timestep_embedder(proj.to(dtype=hidden_states.dtype)) + + +class GateMLP(nn.Module): + """SwiGLU MLP.""" + + def __init__(self, dim: int, hidden_dim: int, dtype=None, device=None, operations=None) -> None: + super().__init__() + self.w1 = operations.Linear(dim, hidden_dim, bias=False, dtype=dtype, device=device) + self.w2 = operations.Linear(hidden_dim, dim, bias=False, dtype=dtype, device=device) + self.w3 = operations.Linear(dim, hidden_dim, bias=False, dtype=dtype, device=device) + + def forward(self, x): + return self.w2(F.silu(self.w1(x), inplace=True).mul_(self.w3(x))) + + +class LensJointAttention(nn.Module): + """Joint image+text attention with fused QKV per stream.""" + + def __init__( + self, + query_dim: int, + added_kv_proj_dim: int, + dim_head: int = 64, + heads: int = 8, + out_dim: Optional[int] = None, + eps: float = 1e-5, + dtype=None, + device=None, + operations=None, + ) -> None: + super().__init__() + self.inner_dim = out_dim if out_dim is not None else dim_head * heads + self.heads = self.inner_dim // dim_head + self.dim_head = dim_head + self.out_dim = out_dim if out_dim is not None else query_dim + + self.norm_q = operations.RMSNorm(dim_head, eps=eps, dtype=dtype, device=device) + self.norm_k = operations.RMSNorm(dim_head, eps=eps, dtype=dtype, device=device) + self.norm_added_q = operations.RMSNorm(dim_head, eps=eps, dtype=dtype, device=device) + self.norm_added_k = operations.RMSNorm(dim_head, eps=eps, dtype=dtype, device=device) + + self.img_qkv = operations.Linear(query_dim, 3 * self.inner_dim, bias=True, dtype=dtype, device=device) + self.txt_qkv = operations.Linear(added_kv_proj_dim, 3 * self.inner_dim, bias=True, dtype=dtype, device=device) + + # ModuleList([Linear, Identity]) for state-dict key compatibility. + self.to_out = nn.ModuleList([ + operations.Linear(self.inner_dim, self.out_dim, bias=True, dtype=dtype, device=device), + nn.Identity(), + ]) + self.to_add_out = operations.Linear(self.inner_dim, query_dim, bias=True, dtype=dtype, device=device) + + def forward( + self, + hidden_states: torch.Tensor, + encoder_hidden_states: torch.Tensor, + image_rotary_emb: Tuple[torch.Tensor, torch.Tensor], + attention_mask: Optional[torch.Tensor] = None, + transformer_options: Optional[Dict[str, Any]] = None, + ) -> Tuple[torch.Tensor, torch.Tensor]: + bsz, seq_img, _ = hidden_states.shape + seq_txt = encoder_hidden_states.shape[1] + + # image stream + img_qkv = self.img_qkv(hidden_states).view(bsz, seq_img, 3, self.heads, self.dim_head) + img_q, img_k, img_v = img_qkv.unbind(dim=2) + img_q = self.norm_q(img_q) + img_k = self.norm_k(img_k) + img_v = img_v.contiguous() + del img_qkv + + # text stream + txt_qkv = self.txt_qkv(encoder_hidden_states).view(bsz, seq_txt, 3, self.heads, self.dim_head) + txt_q, txt_k, txt_v = txt_qkv.unbind(dim=2) + txt_q = self.norm_added_q(txt_q) + txt_k = self.norm_added_k(txt_k) + txt_v = txt_v.contiguous() + del txt_qkv + + img_freqs, txt_freqs = image_rotary_emb + if img_freqs.shape[0] < seq_img: + raise ValueError(f"Image RoPE length {img_freqs.shape[0]} < {seq_img}") + img_freqs = img_freqs[:seq_img] + img_q = apply_rotary_emb_lens(img_q, img_freqs) + img_k = apply_rotary_emb_lens(img_k, img_freqs) + if seq_txt > 0: + if txt_freqs.shape[0] < seq_txt: + raise ValueError(f"Text RoPE length {txt_freqs.shape[0]} < {seq_txt}") + txt_freqs = txt_freqs[:seq_txt] + txt_q = apply_rotary_emb_lens(txt_q, txt_freqs) + txt_k = apply_rotary_emb_lens(txt_k, txt_freqs) + + # [B, S, H, D] → [B, H, S, D] for attention, dels to avoid VRAM peaks + q = torch.cat([img_q, txt_q], dim=1).transpose(1, 2) + del img_q, txt_q + k = torch.cat([img_k, txt_k], dim=1).transpose(1, 2) + del img_k, txt_k + v = torch.cat([img_v, txt_v], dim=1).transpose(1, 2) + del img_v, txt_v + + if attention_mask is not None: + expected = (bsz, 1, 1, seq_img + seq_txt) + if attention_mask.shape != expected: + raise ValueError( + f"attention_mask must be {expected}, got {tuple(attention_mask.shape)}" + ) + attention_mask = attention_mask.to(q.dtype) + + out = optimized_attention( + q, k, v, self.heads, mask=attention_mask, skip_reshape=True, + transformer_options=transformer_options, + ) + + img_out = self.to_out[1](self.to_out[0](out[:, :seq_img, :])) + txt_out = self.to_add_out(out[:, seq_img:, :]) + return img_out, txt_out + + +class LensTransformerBlock(nn.Module): + def __init__( + self, + dim: int, + num_attention_heads: int, + attention_head_dim: int, + eps: float = 1e-6, + rms_norm: bool = True, + dtype=None, + device=None, + operations=None, + ) -> None: + super().__init__() + + self.attn = LensJointAttention( + query_dim=dim, + added_kv_proj_dim=dim, + dim_head=attention_head_dim, + heads=num_attention_heads, + out_dim=dim, + eps=1e-5, + dtype=dtype, + device=device, + operations=operations, + ) + + if rms_norm: + NormCls = operations.RMSNorm + norm_kwargs = {} + else: + NormCls = operations.LayerNorm + norm_kwargs = {"elementwise_affine": False} + + mlp_hidden = int(dim / 3 * 8) + + # Sequential(SiLU, Linear) so state-dict lands at img_mod.1.{weight,bias}. + self.img_mod = nn.Sequential( + nn.SiLU(), + operations.Linear(dim, 6 * dim, bias=True, dtype=dtype, device=device), + ) + self.img_norm1 = NormCls(dim, eps=eps, dtype=dtype, device=device, **norm_kwargs) + self.img_norm2 = NormCls(dim, eps=eps, dtype=dtype, device=device, **norm_kwargs) + self.img_mlp = GateMLP(dim, mlp_hidden, dtype=dtype, device=device, operations=operations) + + self.txt_mod = nn.Sequential( + nn.SiLU(), + operations.Linear(dim, 6 * dim, bias=True, dtype=dtype, device=device), + ) + self.txt_norm1 = NormCls(dim, eps=eps, dtype=dtype, device=device, **norm_kwargs) + self.txt_norm2 = NormCls(dim, eps=eps, dtype=dtype, device=device, **norm_kwargs) + self.txt_mlp = GateMLP(dim, mlp_hidden, dtype=dtype, device=device, operations=operations) + + @staticmethod + def _modulate(x: torch.Tensor, mod_params: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]: + shift, scale, gate = mod_params.chunk(3, dim=-1) + return x * (1 + scale.unsqueeze(1)) + shift.unsqueeze(1), gate.unsqueeze(1) + + def forward( + self, + hidden_states: torch.Tensor, + encoder_hidden_states: torch.Tensor, + temb: torch.Tensor, + image_rotary_emb: Tuple[torch.Tensor, torch.Tensor], + attention_mask: Optional[torch.Tensor] = None, + transformer_options: Optional[Dict[str, Any]] = None, + ) -> Tuple[torch.Tensor, torch.Tensor]: + img_mod1, img_mod2 = self.img_mod(temb).chunk(2, dim=-1) + txt_mod1, txt_mod2 = self.txt_mod(temb).chunk(2, dim=-1) + + img_modulated, img_gate1 = self._modulate(self.img_norm1(hidden_states), img_mod1) + txt_modulated, txt_gate1 = self._modulate(self.txt_norm1(encoder_hidden_states), txt_mod1) + + img_attn, txt_attn = self.attn( + hidden_states=img_modulated, + encoder_hidden_states=txt_modulated, + image_rotary_emb=image_rotary_emb, + attention_mask=attention_mask, + transformer_options=transformer_options, + ) + + hidden_states = hidden_states + img_gate1 * img_attn + encoder_hidden_states = encoder_hidden_states + txt_gate1 * txt_attn + + img_modulated2, img_gate2 = self._modulate(self.img_norm2(hidden_states), img_mod2) + hidden_states = hidden_states + img_gate2 * self.img_mlp(img_modulated2) + + txt_modulated2, txt_gate2 = self._modulate(self.txt_norm2(encoder_hidden_states), txt_mod2) + encoder_hidden_states = encoder_hidden_states + txt_gate2 * self.txt_mlp(txt_modulated2) + + return encoder_hidden_states, hidden_states + + +class _AdaLayerNormContinuousNoAffine(nn.Module): + """AdaLayerNormContinuous(elementwise_affine=False). + + The reference uses ``scale, shift = chunk(2)`` (scale first) — opposite + to Flux's ``LastLayer``. + """ + + def __init__(self, embedding_dim: int, conditioning_embedding_dim: int, eps: float = 1e-6, + dtype=None, device=None, operations=None) -> None: + super().__init__() + self.linear = operations.Linear( + conditioning_embedding_dim, embedding_dim * 2, bias=True, dtype=dtype, device=device + ) + self.eps = eps + self.embedding_dim = embedding_dim + + def forward(self, x: torch.Tensor, conditioning: torch.Tensor) -> torch.Tensor: + emb = self.linear(F.silu(conditioning)) + scale, shift = torch.chunk(emb, 2, dim=-1) + x = F.layer_norm(x, (self.embedding_dim,), None, None, self.eps) + return x * (1 + scale.unsqueeze(1)) + shift.unsqueeze(1) + + +class LensTransformer2DModel(nn.Module): + """Lens dual-stream MMDiT (48 blocks, inner_dim=1536, multi-layer text).""" + + def __init__( + self, + patch_size: int = 2, + in_channels: int = 128, + out_channels: Optional[int] = 32, + num_layers: int = 48, + attention_head_dim: int = 64, + num_attention_heads: int = 24, + enc_hidden_dim: int = 2880, + axes_dims_rope: Tuple[int, int, int] = (8, 28, 28), + rms_norm: bool = True, + multi_layer_encoder_feature: bool = True, + selected_layer_index: Tuple[int, ...] = (5, 11, 17, 23), + image_model=None, # unused; accepted for detection-side configs. + dtype=None, + device=None, + operations=None, + ) -> None: + super().__init__() + self.patch_size = patch_size + self.in_channels = in_channels + self.out_channels = out_channels if out_channels is not None else in_channels + self.inner_dim = num_attention_heads * attention_head_dim + self.multi_layer_encoder_feature = multi_layer_encoder_feature + self.selected_layer_index = list(selected_layer_index) + self.dtype = dtype + + self.pos_embed = LensEmbedRope(theta=10000, axes_dim=list(axes_dims_rope), scale_rope=True) + self.time_text_embed = LensTimestepProjEmbeddings( + embedding_dim=self.inner_dim, dtype=dtype, device=device, operations=operations + ) + + if self.multi_layer_encoder_feature: + self.txt_norm = nn.ModuleList( + [operations.RMSNorm(enc_hidden_dim, eps=1e-5, dtype=dtype, device=device) + for _ in self.selected_layer_index] + ) + self.txt_in = operations.Linear( + enc_hidden_dim * len(self.selected_layer_index), + self.inner_dim, bias=True, dtype=dtype, device=device, + ) + else: + self.txt_norm = operations.RMSNorm(enc_hidden_dim, eps=1e-5, dtype=dtype, device=device) + self.txt_in = operations.Linear(enc_hidden_dim, self.inner_dim, bias=True, dtype=dtype, device=device) + + self.img_in = operations.Linear(in_channels, self.inner_dim, bias=True, dtype=dtype, device=device) + + self.transformer_blocks = nn.ModuleList([ + LensTransformerBlock( + dim=self.inner_dim, + num_attention_heads=num_attention_heads, + attention_head_dim=attention_head_dim, + eps=1e-6, + rms_norm=rms_norm, + dtype=dtype, device=device, operations=operations, + ) + for _ in range(num_layers) + ]) + + self.norm_out = _AdaLayerNormContinuousNoAffine( + self.inner_dim, self.inner_dim, eps=1e-6, + dtype=dtype, device=device, operations=operations, + ) + self.proj_out = operations.Linear( + self.inner_dim, patch_size * patch_size * self.out_channels, bias=True, + dtype=dtype, device=device, + ) + + def forward(self, x: torch.Tensor, timestep: torch.Tensor, context: torch.Tensor, attention_mask: Optional[torch.Tensor] = None, + transformer_options: Optional[Dict[str, Any]] = None, **kwargs) -> torch.Tensor: + if transformer_options is None: + transformer_options = {} + return comfy.patcher_extension.WrapperExecutor.new_class_executor( + self._forward, self, + comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options), + ).execute(x, timestep, context, attention_mask, transformer_options, **kwargs) + + def _forward( + self, + x: torch.Tensor, + timestep: torch.Tensor, + context: torch.Tensor, + attention_mask: Optional[torch.Tensor] = None, + transformer_options: Optional[Dict[str, Any]] = None, + control: Optional[Dict[str, Any]] = None, + **kwargs, + ) -> torch.Tensor: + """ComfyUI bridge: ``(x[B,128,h,w], t[B], context[B,S,L*H], mask[B,S])``.""" + if transformer_options is None: + transformer_options = {} + transformer_options = transformer_options.copy() + patches = transformer_options.get("patches", {}) + patches_replace = transformer_options.get("patches_replace", {}) + blocks_replace = patches_replace.get("dit", {}) + + B, C, h, w = x.shape + hidden_states = x.permute(0, 2, 3, 1).reshape(B, h * w, C) + + if self.multi_layer_encoder_feature: + L = len(self.selected_layer_index) + enc_dim = context.shape[-1] // L + encoder_hidden_states = list( + context.reshape(B, -1, L, enc_dim).unbind(dim=2) + ) + text_seq_len = encoder_hidden_states[0].shape[1] + else: + encoder_hidden_states = context + text_seq_len = context.shape[1] + + if attention_mask is None: + attention_mask = torch.ones( + (B, text_seq_len), dtype=torch.bool, device=x.device + ) + + img_len = h * w + joint_mask = self._build_joint_attention_mask(attention_mask, img_len) + + hidden_states = self.img_in(hidden_states) + timestep = timestep.to(hidden_states.dtype) + + if self.multi_layer_encoder_feature: + normed = [self.txt_norm[i](encoder_hidden_states[i]) for i in range(L)] + encoder_hidden_states = torch.cat(normed, dim=-1) + else: + encoder_hidden_states = self.txt_norm(encoder_hidden_states) + encoder_hidden_states = self.txt_in(encoder_hidden_states) + + if "post_input" in patches: + for p in patches["post_input"]: + out = p({ + "img": hidden_states, + "txt": encoder_hidden_states, + "transformer_options": transformer_options, + }) + hidden_states = out["img"] + encoder_hidden_states = out["txt"] + + temb = self.time_text_embed(timestep, hidden_states) + image_rotary_emb = self.pos_embed([(1, h, w)], [text_seq_len], device=hidden_states.device) + + transformer_options["total_blocks"] = len(self.transformer_blocks) + transformer_options["block_type"] = "double" + for i, block in enumerate(self.transformer_blocks): + transformer_options["block_index"] = i + if ("double_block", i) in blocks_replace: + def block_wrap(args): + out = {} + out["txt"], out["img"] = block( + hidden_states=args["img"], + encoder_hidden_states=args["txt"], + temb=args["vec"], + image_rotary_emb=args["pe"], + attention_mask=args.get("attn_mask"), + transformer_options=args.get("transformer_options"), + ) + return out + out = blocks_replace[("double_block", i)]( + { + "img": hidden_states, + "txt": encoder_hidden_states, + "vec": temb, + "pe": image_rotary_emb, + "attn_mask": joint_mask, + "transformer_options": transformer_options, + }, + {"original_block": block_wrap}, + ) + encoder_hidden_states = out["txt"] + hidden_states = out["img"] + else: + encoder_hidden_states, hidden_states = block( + hidden_states=hidden_states, + encoder_hidden_states=encoder_hidden_states, + temb=temb, + image_rotary_emb=image_rotary_emb, + attention_mask=joint_mask, + transformer_options=transformer_options, + ) + + if "double_block" in patches: + for p in patches["double_block"]: + out = p({ + "img": hidden_states, + "txt": encoder_hidden_states, + "x": x, + "block_index": i, + "transformer_options": transformer_options, + }) + hidden_states = out["img"] + encoder_hidden_states = out["txt"] + + if control is not None: + control_i = control.get("input") + if control_i is not None and i < len(control_i): + add = control_i[i] + if add is not None: + hidden_states[:, :add.shape[1]] += add + + hidden_states = self.norm_out(hidden_states, temb) + out = self.proj_out(hidden_states) + return out.reshape(B, h, w, C).permute(0, 3, 1, 2).contiguous() + + @staticmethod + def _build_joint_attention_mask(text_mask: torch.Tensor, img_len: int) -> torch.Tensor: + if text_mask.dtype != torch.bool: + text_mask = text_mask.bool() + bsz = text_mask.shape[0] + img_ones = torch.ones((bsz, img_len), dtype=torch.bool, device=text_mask.device) + joint = torch.cat([img_ones, text_mask], dim=1) + additive = torch.zeros_like(joint, dtype=torch.float32) + additive.masked_fill_(~joint, torch.finfo(torch.float32).min) + return additive[:, None, None, :] diff --git a/comfy/model_base.py b/comfy/model_base.py index d81f13c69..d4ab1499e 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -35,6 +35,7 @@ import comfy.ldm.hydit.models import comfy.ldm.audio.dit import comfy.ldm.audio.embedders import comfy.ldm.flux.model +import comfy.ldm.lens.model import comfy.ldm.lightricks.model import comfy.ldm.hunyuan_video.model import comfy.ldm.cosmos.model @@ -1058,6 +1059,27 @@ class Flux2(Flux): out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn) return out + +class Lens(BaseModel): + def __init__(self, model_config, model_type=ModelType.FLUX, device=None): + super().__init__( + model_config, model_type, device=device, + unet_model=comfy.ldm.lens.model.LensTransformer2DModel, + ) + + def encode_adm(self, **kwargs): + return None # Lens has no pooled/ADM conditioning. + + def extra_conds(self, **kwargs): + out = super().extra_conds(**kwargs) + cross_attn = kwargs.get("cross_attn", None) + if cross_attn is not None: + out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn) + attention_mask = kwargs.get("attention_mask", None) + if attention_mask is not None: + out['attention_mask'] = comfy.conds.CONDRegular(attention_mask) + return out + class GenmoMochi(BaseModel): def __init__(self, model_config, model_type=ModelType.FLOW, device=None): super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.genmo.joint_model.asymm_models_joint.AsymmDiTJoint) diff --git a/comfy/model_detection.py b/comfy/model_detection.py index 70b4df8b3..2b0b98cd8 100644 --- a/comfy/model_detection.py +++ b/comfy/model_detection.py @@ -755,6 +755,30 @@ def detect_unet_config(state_dict, key_prefix, metadata=None): dit_config["timestep_scale"] = 1000.0 return dit_config + if '{}transformer_blocks.0.attn.norm_added_q.weight'.format(key_prefix) in state_dict_keys \ + and '{}transformer_blocks.0.img_mlp.w1.weight'.format(key_prefix) in state_dict_keys: # Lens + img_in_w = state_dict['{}img_in.weight'.format(key_prefix)] + proj_out_w = state_dict['{}proj_out.weight'.format(key_prefix)] + multi_layer = '{}txt_norm.0.weight'.format(key_prefix) in state_dict_keys + if multi_layer: + enc_hidden_dim = state_dict['{}txt_norm.0.weight'.format(key_prefix)].shape[0] + # Indices are TE-side; the DiT just consumes L layers in order. + selected_layer_index = tuple(range(count_blocks(state_dict_keys, '{}txt_norm.'.format(key_prefix) + '{}.'))) + else: + enc_hidden_dim = state_dict['{}txt_norm.weight'.format(key_prefix)].shape[0] + selected_layer_index = (0,) + + return { + "image_model": "lens", + "in_channels": img_in_w.shape[1], + "out_channels": proj_out_w.shape[0] // 4, # patch_size ** 2 (=2² default) + "num_layers": count_blocks(state_dict_keys, '{}transformer_blocks.'.format(key_prefix) + '{}.'), + "num_attention_heads": img_in_w.shape[0] // 64, # // attention_head_dim default + "enc_hidden_dim": enc_hidden_dim, + "multi_layer_encoder_feature": multi_layer, + "selected_layer_index": selected_layer_index, + } + if '{}txt_norm.weight'.format(key_prefix) in state_dict_keys: # Qwen Image dit_config = {} dit_config["image_model"] = "qwen_image" diff --git a/comfy/sd.py b/comfy/sd.py index 7bd07ed3a..b18f32dec 100644 --- a/comfy/sd.py +++ b/comfy/sd.py @@ -69,6 +69,7 @@ import comfy.text_encoders.ernie import comfy.text_encoders.gemma4 import comfy.text_encoders.cogvideo import comfy.text_encoders.sa3 +import comfy.text_encoders.gpt_oss import comfy.model_patcher import comfy.lora @@ -1269,6 +1270,7 @@ class CLIPType(Enum): FLUX2 = 25 LONGCAT_IMAGE = 26 COGVIDEOX = 27 + LENS = 28 @@ -1321,6 +1323,7 @@ class TEModel(Enum): GEMMA_4_E2B = 30 GEMMA_4_31B = 31 T5_GEMMA = 32 + GPT_OSS_20B = 33 def detect_te_model(sd): @@ -1362,6 +1365,12 @@ def detect_te_model(sd): else: return TEModel.GEMMA_3_4B return TEModel.GEMMA_2_2B + # Must precede the Qwen2.5-7B k_proj.bias=512 check (GPT-OSS also has 8*64=512). + if "model.layers.0.self_attn.sinks" in sd and ( + "model.layers.0.mlp.experts.gate_up_proj" in sd + or "model.layers.0.mlp.experts.gate_up_proj_blocks" in sd + ): + return TEModel.GPT_OSS_20B if 'model.layers.0.self_attn.k_proj.bias' in sd: weight = sd['model.layers.0.self_attn.k_proj.bias'] if weight.shape[0] == 256: @@ -1544,6 +1553,11 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip clip_target.clip = comfy.text_encoders.flux.flux2_te(**llama_detect(clip_data), pruned=te_model == TEModel.MISTRAL3_24B_PRUNED_FLUX2) clip_target.tokenizer = comfy.text_encoders.flux.Flux2Tokenizer tokenizer_data["tekken_model"] = clip_data[0].get("tekken_model", None) + elif te_model == TEModel.GPT_OSS_20B: + mxfp4 = any("model.layers.0.mlp.experts.gate_up_proj_blocks" in sd for sd in clip_data) + clip_target.clip = comfy.text_encoders.gpt_oss.lens_te(**llama_detect(clip_data), mxfp4_runtime=mxfp4) + clip_target.tokenizer = comfy.text_encoders.gpt_oss.LensTokenizer + tokenizer_data["tokenizer_json"] = clip_data[0].get("tokenizer_json", None) elif te_model == TEModel.QWEN3_4B: if clip_type == CLIPType.FLUX or clip_type == CLIPType.FLUX2: clip_target.clip = comfy.text_encoders.flux.klein_te(**llama_detect(clip_data), model_type="qwen3_4b") diff --git a/comfy/supported_models.py b/comfy/supported_models.py index 617db4f28..c5fb5e5cd 100644 --- a/comfy/supported_models.py +++ b/comfy/supported_models.py @@ -829,6 +829,50 @@ class Flux2(Flux): return None + +class Lens(supported_models_base.BASE): + """Microsoft Lens (3.8B dual-stream MMDiT, GPT-OSS-20B text features, Flux2 VAE).""" + + unet_config = { + "image_model": "lens", + } + + sampling_settings = { + "shift": 1.829, # Default mu for 1440x1440 (and any seq_len > 4300 + } + + unet_extra_config = {} + latent_format = latent_formats.Flux2 + + supported_inference_dtypes = [torch.bfloat16, torch.float32] # fp16 causes NaNs + + vae_key_prefix = ["vae."] + text_encoder_key_prefix = ["text_encoders."] + + def __init__(self, unet_config): + super().__init__(unet_config) + + def get_model(self, state_dict, prefix="", device=None): + return model_base.Lens(self, model_type=model_base.ModelType.FLUX, device=device) + + def clip_target(self, state_dict={}): + pref = self.text_encoder_key_prefix[0] + for hint in ("gpt_oss.transformer.", ""): + full_prefix = "{}{}".format(pref, hint) + if "{}model.layers.0.self_attn.sinks".format(full_prefix) in state_dict: + detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, full_prefix) + if "{}model.layers.0.mlp.experts.gate_up_proj_blocks".format(full_prefix) in state_dict: + detect["mxfp4_runtime"] = True + return supported_models_base.ClipTarget( + comfy.text_encoders.gpt_oss.LensTokenizer, + comfy.text_encoders.gpt_oss.lens_te(**detect), + ) + return supported_models_base.ClipTarget( + comfy.text_encoders.gpt_oss.LensTokenizer, + comfy.text_encoders.gpt_oss.lens_te(), + ) + + class GenmoMochi(supported_models_base.BASE): unet_config = { "image_model": "mochi_preview", @@ -2096,6 +2140,7 @@ models = [ Omnigen2, QwenImage, Flux2, + Lens, Kandinsky5Image, Kandinsky5, Anima, diff --git a/comfy/text_encoders/gpt_oss.py b/comfy/text_encoders/gpt_oss.py new file mode 100644 index 000000000..7c390dd3e --- /dev/null +++ b/comfy/text_encoders/gpt_oss.py @@ -0,0 +1,789 @@ +"""GPT-OSS text encoder for Lens.""" + +from __future__ import annotations + +import gc +import logging +import math +from dataclasses import dataclass +from typing import Any, Dict, List, Optional, Sequence, Tuple + +import torch +import torch.nn as nn +import torch.nn.functional as F + +import comfy.ops +from comfy import sd1_clip +from comfy.ldm.modules.attention import TORCH_HAS_GQA, optimized_attention_for_device +from comfy.text_encoders.llama import RMSNorm, apply_rope + + +@dataclass +class GptOss20BConfig: + vocab_size: int = 201088 + hidden_size: int = 2880 + intermediate_size: int = 2880 + num_hidden_layers: int = 24 + num_attention_heads: int = 64 + num_key_value_heads: int = 8 + head_dim: int = 64 + num_local_experts: int = 32 + num_experts_per_tok: int = 4 + sliding_window: int = 128 + original_max_position_embeddings: int = 4096 + rope_theta: float = 150000.0 + rope_factor: float = 32.0 + rope_beta_fast: float = 32.0 + rope_beta_slow: float = 1.0 + rope_truncate: bool = False + rms_norm_eps: float = 1e-5 + attention_bias: bool = True + layer_types: Optional[List[str]] = None + moe_alpha: float = 1.702 + moe_limit: float = 7.0 + + def __post_init__(self): + if self.layer_types is None: + self.layer_types = [ + "sliding_attention" if (i + 1) % 2 else "full_attention" + for i in range(self.num_hidden_layers) + ] + + +def _yarn_inv_freq( + head_dim: int, + base: float, + factor: float, + beta_fast: float, + beta_slow: float, + original_max_position_embeddings: int, + truncate: bool, + device=None, +) -> tuple[torch.Tensor, float]: + """YARN inv_freq + attention scaling (matches transformers).""" + dim = head_dim + + def find_correction_dim(num_rotations: float) -> float: + return (dim * math.log(original_max_position_embeddings / (num_rotations * 2 * math.pi))) / ( + 2 * math.log(base) + ) + + def find_correction_range() -> tuple[float, float]: + low = find_correction_dim(beta_fast) + high = find_correction_dim(beta_slow) + if truncate: + low = math.floor(low) + high = math.ceil(high) + return max(low, 0), min(high, dim - 1) + + def linear_ramp_factor(min_: float, max_: float, n: int) -> torch.Tensor: + if min_ == max_: + max_ += 0.001 + linear = (torch.arange(n, dtype=torch.float32, device=device) - min_) / (max_ - min_) + return torch.clamp(linear, 0, 1) + + def get_mscale(scale: float) -> float: + if scale <= 1: + return 1.0 + return 0.1 * math.log(scale) + 1.0 + + attention_scaling = get_mscale(factor) + + pos_freqs = base ** (torch.arange(0, dim, 2, dtype=torch.float32, device=device) / dim) + inv_freq_extrapolation = 1.0 / pos_freqs + inv_freq_interpolation = 1.0 / (factor * pos_freqs) + + low, high = find_correction_range() + extrap_factor = 1 - linear_ramp_factor(low, high, dim // 2) + inv_freq = inv_freq_interpolation * (1 - extrap_factor) + inv_freq_extrapolation * extrap_factor + return inv_freq, attention_scaling + + +def _build_freqs_cis(inv_freq: torch.Tensor, attention_scaling: float, position_ids: torch.Tensor, dtype: torch.dtype, +) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]: + inv_freq_e = inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1) + pos_e = position_ids[:, None, :].float() + freqs = (inv_freq_e @ pos_e).transpose(1, 2) + emb = torch.cat((freqs, freqs), dim=-1) + cos = (emb.cos() * attention_scaling).to(dtype).unsqueeze(1) + sin = (emb.sin() * attention_scaling).to(dtype).unsqueeze(1) + sin_split = sin.shape[-1] // 2 + return cos, sin[..., :sin_split], -sin[..., sin_split:] + + +def _attention_with_sinks( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + sinks: torch.Tensor, + attention_mask: Optional[torch.Tensor], + num_heads: int, + num_kv_groups: int, +) -> torch.Tensor: + """Attention with per-head sinks. + + Sinks add a learned term to each row's softmax denominator but contribute + nothing to the output. We fake this by appending one zero k/v position and + putting the sink logit in the mask at that column. + """ + + if num_kv_groups > 1 and not TORCH_HAS_GQA: + k = k.repeat_interleave(num_kv_groups, dim=1) + v = v.repeat_interleave(num_kv_groups, dim=1) + + B, _, S_q, D = q.shape + H_kv = k.shape[1] + S_kv = k.shape[-2] + + k = torch.cat([k, k.new_zeros(B, H_kv, 1, D)], dim=-2) + v = torch.cat([v, v.new_zeros(B, H_kv, 1, D)], dim=-2) + + sinks_col = sinks.to(q.dtype).view(1, num_heads, 1, 1).expand(B, num_heads, S_q, 1) + if attention_mask is not None: + mask_left = attention_mask[..., :S_kv].expand(B, num_heads, S_q, S_kv) + else: + mask_left = q.new_zeros(B, num_heads, S_q, S_kv) + mask = torch.cat([mask_left, sinks_col], dim=-1) + + op = optimized_attention_for_device(q.device, mask=True, small_input=True) + return op(q, k, v, num_heads, mask=mask, skip_reshape=True, enable_gqa=True) + + +class GptOssAttention(nn.Module): + def __init__(self, config: GptOss20BConfig, layer_idx: int, device=None, dtype=None, ops: Any = None): + super().__init__() + self.layer_idx = layer_idx + self.layer_type = config.layer_types[layer_idx] + self.num_heads = config.num_attention_heads + self.num_kv_heads = config.num_key_value_heads + self.num_kv_groups = self.num_heads // self.num_kv_heads + self.head_dim = config.head_dim + self.hidden_size = config.hidden_size + self.sliding_window = config.sliding_window if self.layer_type == "sliding_attention" else None + + bias = config.attention_bias + self.q_proj = ops.Linear(config.hidden_size, self.num_heads * self.head_dim, bias=bias, device=device, dtype=dtype) + self.k_proj = ops.Linear(config.hidden_size, self.num_kv_heads * self.head_dim, bias=bias, device=device, dtype=dtype) + self.v_proj = ops.Linear(config.hidden_size, self.num_kv_heads * self.head_dim, bias=bias, device=device, dtype=dtype) + self.o_proj = ops.Linear(self.num_heads * self.head_dim, config.hidden_size, bias=bias, device=device, dtype=dtype) + self.sinks = nn.Parameter(torch.empty(self.num_heads, device=device, dtype=dtype)) + + def forward(self, hidden_states: torch.Tensor, attention_mask: Optional[torch.Tensor], freqs_cis) -> torch.Tensor: + B, S, _ = hidden_states.shape + + q = self.q_proj(hidden_states).view(B, S, self.num_heads, self.head_dim).transpose(1, 2) + k = self.k_proj(hidden_states).view(B, S, self.num_kv_heads, self.head_dim).transpose(1, 2) + v = self.v_proj(hidden_states).view(B, S, self.num_kv_heads, self.head_dim).transpose(1, 2) + + q, k = apply_rope(q, k, freqs_cis) + + out = _attention_with_sinks(q, k, v, self.sinks, attention_mask, self.num_heads, self.num_kv_groups) + return self.o_proj(out) + + +# Mixture of Experts + +class GptOssTopKRouter(nn.Module): + def __init__(self, config: GptOss20BConfig, device=None, dtype=None): + super().__init__() + self.top_k = config.num_experts_per_tok + self.num_experts = config.num_local_experts + # Raw Parameters (not Linear) to match HF state-dict keys. + self.weight = nn.Parameter(torch.empty(config.num_local_experts, config.hidden_size, device=device, dtype=dtype)) + self.bias = nn.Parameter(torch.empty(config.num_local_experts, device=device, dtype=dtype)) + + def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]: + logits = F.linear(hidden_states, self.weight, self.bias) + top_vals, top_idx = torch.topk(logits, self.top_k, dim=-1) + # Softmax over top-k slice only (matches transformers), not all experts. + scores = F.softmax(top_vals, dim=-1, dtype=top_vals.dtype) + return scores, top_idx + + +class GptOssExperts(nn.Module): + def __init__(self, config: GptOss20BConfig, device=None, dtype=None): + super().__init__() + self.num_experts = config.num_local_experts + self.hidden_size = config.hidden_size + self.intermediate_size = config.intermediate_size + self.alpha = config.moe_alpha + self.limit = config.moe_limit + + E = self.num_experts + H = self.hidden_size + I = self.intermediate_size + + self.gate_up_proj_bias = nn.Parameter(torch.empty(E, 2 * I, device=device, dtype=dtype)) + self.down_proj_bias = nn.Parameter(torch.empty(E, H, device=device, dtype=dtype)) + self.gate_up_proj = nn.Parameter(torch.empty(E, H, 2 * I, device=device, dtype=dtype)) + self.down_proj = nn.Parameter(torch.empty(E, I, H, device=device, dtype=dtype)) + + def switch_to_mxfp4(self, device=None): + """Swap bf16 weight Parameters for uint8 MXFP4 packed buffers. + + On-disk MXFP4 layout: ``[E, 2*I, G_up, 16]`` uint8 + ``[E, 2*I, G_up]`` + uint8 (E8M0) for ``gate_up``; ``[E, H, G_down, 16]`` + ``[E, H, G_down]`` + for ``down``. ``G_up * 32 = H``, ``G_down * 32 = I``. + """ + E, H, I = self.num_experts, self.hidden_size, self.intermediate_size + if H % 32 != 0 or I % 32 != 0: + raise ValueError(f"MXFP4 requires H, I divisible by 32; got H={H}, I={I}") + del self.gate_up_proj + del self.down_proj + G_up = H // 32 + G_down = I // 32 + self.register_buffer("gate_up_proj_blocks", torch.empty(E, 2 * I, G_up, 16, dtype=torch.uint8, device=device)) + self.register_buffer("gate_up_proj_scales", torch.empty(E, 2 * I, G_up, dtype=torch.uint8, device=device)) + self.register_buffer("down_proj_blocks", torch.empty(E, H, G_down, 16, dtype=torch.uint8, device=device)) + self.register_buffer("down_proj_scales", torch.empty(E, H, G_down, dtype=torch.uint8, device=device)) + + def _apply_gate(self, gate_up: torch.Tensor) -> torch.Tensor: + gate = gate_up[..., ::2] + up = gate_up[..., 1::2] + gate = gate.clamp(max=self.limit) + up = up.clamp(min=-self.limit, max=self.limit) + glu = gate * torch.sigmoid(gate * self.alpha) + return (up + 1) * glu + + @staticmethod + def _dequant_one_expert( + blocks_e: torch.Tensor, + scales_e: torch.Tensor, + dtype: torch.dtype, + ) -> torch.Tensor: + """Dequant one expert's MXFP4 ``[D, G, 16]`` + ``[D, G]`` to ``[G*32, D]``.""" + D, G, B = blocks_e.shape + val_per_row = G * 32 + lut = _fp4_lut(dtype, blocks_e.device) + blocks_flat = blocks_e.reshape(D * G, B) + scales_flat = (scales_e.to(torch.int32) - 127).reshape(D * G, 1) + dec = torch.empty(D * G, B * 2, dtype=dtype, device=blocks_e.device) + dec[:, 0::2] = lut[(blocks_flat & 0x0F).to(torch.long)] + dec[:, 1::2] = lut[(blocks_flat >> 4).to(torch.long)] + torch.ldexp(dec, scales_flat, out=dec) + return dec.view(D, val_per_row).transpose(0, 1) + + def forward(self, hidden_states: torch.Tensor, router_indices: torch.Tensor, routing_weights: torch.Tensor) -> torch.Tensor: + N = hidden_states.shape[0] + top_k = router_indices.shape[-1] + H = hidden_states.shape[-1] + + per_pair = torch.zeros((N * top_k, H), dtype=hidden_states.dtype, device=hidden_states.device) + + expert_mask = F.one_hot(router_indices, num_classes=self.num_experts).permute(2, 1, 0) + expert_hit = torch.greater(expert_mask.sum(dim=(-1, -2)), 0).nonzero() + + is_mxfp4 = hasattr(self, "gate_up_proj_blocks") + + for ei in expert_hit: + expert_idx = int(ei.item()) + top_k_pos, token_idx = torch.where(expert_mask[expert_idx]) + current = hidden_states[token_idx] + + if is_mxfp4: + gate_up_w = self._dequant_one_expert( + self.gate_up_proj_blocks[expert_idx], + self.gate_up_proj_scales[expert_idx], + current.dtype, + ) + down_w = self._dequant_one_expert( + self.down_proj_blocks[expert_idx], + self.down_proj_scales[expert_idx], + current.dtype, + ) + else: + gate_up_w = comfy.ops.cast_to_input(self.gate_up_proj[expert_idx], current, copy=False) + down_w = comfy.ops.cast_to_input(self.down_proj[expert_idx], current, copy=False) + + gate_up_b = comfy.ops.cast_to_input(self.gate_up_proj_bias[expert_idx], current, copy=False) + down_b = comfy.ops.cast_to_input(self.down_proj_bias[expert_idx], current, copy=False) + + gate_up = current @ gate_up_w + gate_up_b + gated = self._apply_gate(gate_up) + expert_out = gated @ down_w + down_b + weighted = expert_out * routing_weights[token_idx, top_k_pos, None] + + flat_idx = token_idx * top_k + top_k_pos + per_pair[flat_idx] = weighted.to(per_pair.dtype) + + return per_pair.view(N, top_k, H).sum(dim=1) + + +class GptOssMLP(nn.Module): + def __init__(self, config: GptOss20BConfig, device=None, dtype=None): + super().__init__() + self.router = GptOssTopKRouter(config, device=device, dtype=dtype) + self.experts = GptOssExperts(config, device=device, dtype=dtype) + + def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: + B, S, H = hidden_states.shape + flat = hidden_states.reshape(-1, H) + scores, idx = self.router(flat) + out = self.experts(flat, idx, scores) + return out.reshape(B, S, H) + + +# Decoder layer + model + +class GptOssDecoderLayer(nn.Module): + def __init__(self, config: GptOss20BConfig, layer_idx: int, device=None, dtype=None, ops: Any = None): + super().__init__() + self.self_attn = GptOssAttention(config, layer_idx, device=device, dtype=dtype, ops=ops) + self.mlp = GptOssMLP(config, device=device, dtype=dtype) + self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + self.layer_type = config.layer_types[layer_idx] + + def forward(self, x: torch.Tensor, attention_masks: dict[str, Optional[torch.Tensor]], freqs_cis) -> torch.Tensor: + residual = x + x = self.input_layernorm(x) + x = self.self_attn(x, attention_masks[self.layer_type], freqs_cis) + x = residual + x + + residual = x + x = self.post_attention_layernorm(x) + x = self.mlp(x) + x = residual + x + return x + + +def _make_full_causal_mask(B: int, S: int, key_padding_mask: Optional[torch.Tensor], dtype, device): + neg = torch.finfo(dtype).min + mask = torch.full((S, S), neg, dtype=dtype, device=device).triu_(1) + mask = mask.unsqueeze(0).unsqueeze(0).expand(B, 1, S, S).contiguous() + if key_padding_mask is not None: + kp = key_padding_mask.to(dtype=dtype) + kp = (1.0 - kp).reshape(B, 1, 1, S) * neg + mask = mask + kp + return mask + + +def _make_sliding_causal_mask(B: int, S: int, window: int, key_padding_mask: Optional[torch.Tensor], dtype, device): + neg = torch.finfo(dtype).min + i = torch.arange(S, device=device).view(-1, 1) + j = torch.arange(S, device=device).view(1, -1) + keep = (j <= i) & (j > i - window) + mask = torch.where(keep, torch.zeros((), dtype=dtype, device=device), torch.full((), neg, dtype=dtype, device=device)) + mask = mask.unsqueeze(0).unsqueeze(0).expand(B, 1, S, S).contiguous() + if key_padding_mask is not None: + kp = key_padding_mask.to(dtype=dtype) + kp = (1.0 - kp).reshape(B, 1, 1, S) * neg + mask = mask + kp + return mask + + +class GptOssModel(nn.Module): + """GPT-OSS decoder with multi-layer hidden-state capture + early exit.""" + + def __init__(self, config: GptOss20BConfig, device=None, dtype=None, ops: Any = None): + super().__init__() + self.config = config + self.dtype = dtype + self.embed_tokens = ops.Embedding(config.vocab_size, config.hidden_size, device=device, dtype=dtype) + self.layers = nn.ModuleList( + [ + GptOssDecoderLayer(config, i, device=device, dtype=dtype, ops=ops) + for i in range(config.num_hidden_layers) + ] + ) + self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + + # Always build on CPU so the buffer survives meta-device construction. + inv_freq, attn_scaling = _yarn_inv_freq( + head_dim=config.head_dim, + base=config.rope_theta, + factor=config.rope_factor, + beta_fast=config.rope_beta_fast, + beta_slow=config.rope_beta_slow, + original_max_position_embeddings=config.original_max_position_embeddings, + truncate=config.rope_truncate, + device=torch.device("cpu"), + ) + self.register_buffer("rope_inv_freq", inv_freq, persistent=False) + self.rope_attention_scaling = float(attn_scaling) + + @property + def num_layers(self) -> int: + return self.config.num_hidden_layers + + def get_input_embeddings(self): + return self.embed_tokens + + def _build_attention_masks(self, B: int, S: int, attention_mask: Optional[torch.Tensor], dtype: torch.dtype, device, + ) -> dict[str, torch.Tensor]: + full = _make_full_causal_mask(B, S, attention_mask, dtype, device) + masks = {"full_attention": full} + if any(t == "sliding_attention" for t in self.config.layer_types): + masks["sliding_attention"] = _make_sliding_causal_mask( + B, S, self.config.sliding_window, attention_mask, dtype, device + ) + return masks + + def forward(self, input_ids: torch.LongTensor, attention_mask: Optional[torch.Tensor] = None, + capture_layers: Optional[Sequence[int]] = None) -> dict[str, Any]: + B, S = input_ids.shape + device = input_ids.device + dtype = self.dtype + + hidden_states = self.embed_tokens(input_ids, out_dtype=dtype) + + position_ids = torch.arange(S, device=device).unsqueeze(0).expand(B, -1) + freqs_cis = _build_freqs_cis(self.rope_inv_freq.to(device=device), self.rope_attention_scaling, position_ids, dtype) + + attn_masks = self._build_attention_masks(B, S, attention_mask, dtype, device) + + capture_layers = list(capture_layers) if capture_layers else None + if capture_layers: + max_layer = max(capture_layers) + wanted = {idx: pos for pos, idx in enumerate(capture_layers)} + captured: List[Optional[torch.Tensor]] = [None] * len(capture_layers) + else: + max_layer = self.config.num_hidden_layers - 1 + wanted = None + captured = None + + for i, layer in enumerate(self.layers): + hidden_states = layer(hidden_states, attn_masks, freqs_cis) + if wanted is not None and i in wanted: + captured[wanted[i]] = hidden_states + if i >= max_layer: + break + + if captured is not None: + return {"hidden_states": captured} + return {"last_hidden_state": self.norm(hidden_states)} + + +# Lens chat-template constants (verbatim from the reference pipeline). +_LENS_CHAT_SYSTEM = ( + "Describe the image by detailing the color, shape, size, texture, " + "quantity, text, spatial relationships of the objects and background." +) +_LENS_CHAT_ASSISTANT_THINKING = "Need to generate one image according to the description." +LENS_TXT_OFFSET = 97 +LENS_SELECTED_LAYERS = (5, 11, 17, 23) +LENS_MAX_TOKENS = 512 + + +# The reference GPT-OSS Harmony template injects today's date here +_LENS_CHAT_DATE = "2026-05-23" + + +def _lens_render_chat(prompt: str) -> str: + """Render the Lens prompt in GPT-OSS Harmony format.""" + return ( + f"<|start|>system<|message|>" + f"You are ChatGPT, a large language model trained by OpenAI.\n" + f"Knowledge cutoff: 2024-06\n" + f"Current date: {_LENS_CHAT_DATE}\n\n" + f"Reasoning: medium\n\n" + f"# Valid channels: analysis, commentary, final. " + f"Channel must be included for every message.<|end|>" + f"<|start|>developer<|message|># Instructions\n\n" + f"{_LENS_CHAT_SYSTEM}\n\n<|end|>" + f"<|start|>user<|message|>{prompt}<|end|>" + f"<|start|>assistant<|channel|>analysis<|message|>" + f"{_LENS_CHAT_ASSISTANT_THINKING}<|end|>" + f"<|start|>assistant<|channel|>final<|message|>" + ) + + +# GPT-OSS-20B fixed token IDs (from the tokenizer's added-tokens table). +_LENS_PAD_TOKEN_ID = 199999 # <|endoftext|> + + +class _GptOssRawTokenizer: + """Raw ``tokenizers.Tokenizer`` wrapper. + + The tokenizer JSON ships as a byte tensor inside the encoder checkpoint + (``tokenizer_json`` key) rather than as a committed file. Extracted + it in ``sd.py`` and passes it here via ``tokenizer_data``. + """ + + def __init__(self, tokenizer_json_bytes=None, **kwargs): + from tokenizers import Tokenizer + if isinstance(tokenizer_json_bytes, torch.Tensor): + tokenizer_json_bytes = bytes(tokenizer_json_bytes.tolist()) + if tokenizer_json_bytes is None: + raise ValueError( + "Lens tokenizer requires the ``tokenizer_json`` byte tensor in the " + "encoder state dict. Re-bundle the encoder via bundle_te.py so it " + "embeds the tokenizer." + ) + self.tokenizer = Tokenizer.from_str(tokenizer_json_bytes.decode("utf-8")) + + @classmethod + def from_pretrained(cls, tokenizer_data, **kwargs): + return cls(tokenizer_json_bytes=tokenizer_data, **kwargs) + + def __call__(self, text): + return {"input_ids": self.tokenizer.encode(text, add_special_tokens=False).ids} + + def get_vocab(self): + return self.tokenizer.get_vocab() + + def convert_tokens_to_ids(self, tokens): + return [self.tokenizer.token_to_id(t) for t in tokens] + + def decode(self, ids, **kwargs): + return self.tokenizer.decode(ids, skip_special_tokens=kwargs.get("skip_special_tokens", False)) + + +class LensGptOssTokenizer(sd1_clip.SDTokenizer): + tokenizer_json_data = None + + def __init__(self, embedding_directory=None, tokenizer_data={}): + tokenizer_json = tokenizer_data.get("tokenizer_json", None) + self.tokenizer_json_data = tokenizer_json + super().__init__( + tokenizer_json, + embedding_directory=embedding_directory, + pad_with_end=False, + embedding_size=2880, + embedding_key="gpt_oss", + tokenizer_class=_GptOssRawTokenizer, + has_start_token=False, + has_end_token=False, + pad_to_max_length=False, + max_length=99999999, + min_length=1, + pad_left=False, + disable_weights=True, + tokenizer_data=tokenizer_data, + ) + self.pad_token_id = _LENS_PAD_TOKEN_ID + + def tokenize_with_weights(self, text: str, return_word_ids=False, **kwargs): + # Empty prompt -> empty list; encode_token_weights returns zeros (uncond). + if not text or not text.strip(): + return [[]] + rendered = _lens_render_chat(text) + ids = self.tokenizer(rendered)["input_ids"] + if len(ids) > LENS_MAX_TOKENS: + ids = ids[:LENS_MAX_TOKENS] + return [[(int(t), 1.0) for t in ids]] + + def state_dict(self): + if self.tokenizer_json_data is not None: + return {"tokenizer_json": self.tokenizer_json_data} + return {} + + +class LensTokenizer(sd1_clip.SD1Tokenizer): + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__( + embedding_directory=embedding_directory, + tokenizer_data=tokenizer_data, + name="gpt_oss", + tokenizer=LensGptOssTokenizer, + ) + + +# MXFP4 E2M1 LUT (1 sign + 2 exp + 1 mantissa). +_FP4_VALUES: Tuple[float, ...] = ( + 0.0, 0.5, 1.0, 1.5, 2.0, 3.0, 4.0, 6.0, + -0.0, -0.5, -1.0, -1.5, -2.0, -3.0, -4.0, -6.0, +) + + +_FP4_LUT_CACHE: Dict[Tuple[torch.dtype, str], torch.Tensor] = {} + + +def _fp4_lut(dtype: torch.dtype, device: torch.device) -> torch.Tensor: + """Cached per (dtype, device) FP4 lookup table — avoids per-call allocation.""" + key = (dtype, str(device)) + lut = _FP4_LUT_CACHE.get(key) + if lut is None: + lut = torch.tensor(_FP4_VALUES, dtype=dtype, device=device) + _FP4_LUT_CACHE[key] = lut + return lut + + +def _safe_dequant_moe_tensor( + blocks: torch.Tensor, + scales: torch.Tensor, + *, + dtype: torch.dtype = torch.bfloat16, + rows_per_chunk: int = 4096, +) -> torch.Tensor: + """Eager full-tensor MXFP4 dequant -> ``[E, H, 2*I]`` ``dtype``. + + Allocates the output in its final transposed layout and writes in chunks + """ + blocks = blocks.to(torch.uint8) + scales = scales.to(torch.int32) - 127 + + assert blocks.shape[:-1] == scales.shape, ( + f"{blocks.shape[:-1]=} does not match {scales.shape=}" + ) + *prefix_shape, G, B = blocks.shape + if len(prefix_shape) != 2: + raise ValueError(f"expected 2-D prefix (E, 2*I); got {prefix_shape}") + + E, D = prefix_shape + val_per_row = G * B * 2 # this is H after dequant + + rows_total = E * D * G + blocks = blocks.reshape(rows_total, B) + scales = scales.reshape(rows_total, 1) + + lut = _fp4_lut(dtype, blocks.device) + out = torch.empty(E, val_per_row, D, dtype=dtype, device=blocks.device) + + for e in range(E): + for d0 in range(0, D, rows_per_chunk): + d1 = min(d0 + rows_per_chunk, D) + r0 = e * D * G + d0 * G + r1 = e * D * G + d1 * G + blk = blocks[r0:r1] + exp = scales[r0:r1] + dec = torch.empty((d1 - d0) * G, B * 2, dtype=dtype, device=blocks.device) + dec[:, 0::2] = lut[(blk & 0x0F).to(torch.long)] + dec[:, 1::2] = lut[(blk >> 4).to(torch.long)] + torch.ldexp(dec, exp, out=dec) + out[e, :, d0:d1] = dec.view(d1 - d0, val_per_row).transpose(0, 1) + del blk, exp, dec + return out + + +def _dequant_mxfp4_state_dict(sd: Dict[str, torch.Tensor], target_dtype: torch.dtype) -> Dict[str, torch.Tensor]: + """Eager-dequant every ``*_blocks``/``*_scales`` pair in ``sd`` in place.""" + pairs: List[Tuple[str, str, str]] = [] + for k in list(sd.keys()): + if k.endswith("_blocks"): + stem = k[: -len("_blocks")] + sk = stem + "_scales" + if sk in sd: + pairs.append((stem, k, sk)) + + if not pairs: + return sd + + logging.info("Lens: dequantizing %d MXFP4 expert tensors -> %s", len(pairs), target_dtype) + for stem, bk, sk in pairs: + blocks = sd.pop(bk) + scales = sd.pop(sk) + sd[stem] = _safe_dequant_moe_tensor(blocks, scales, dtype=target_dtype) + del blocks, scales + + gc.collect() + return sd + + +class LensGptOssClipModel(nn.Module): + """SDClipModel-shaped Lens GPT-OSS encoder (multi-layer feature extractor).""" + + def __init__(self, device="cpu", dtype=None, model_options=None, **_): + super().__init__() + model_options = dict(model_options or {}) + + operations = model_options.get("custom_operations") + quant_config = model_options.get("quantization_metadata") + if operations is None: + if quant_config is not None: + operations = comfy.ops.mixed_precision_ops( + quant_config, dtype, full_precision_mm=True + ) + else: + operations = comfy.ops.manual_cast + self.operations = operations + + cfg_overrides = model_options.get("gpt_oss_config", {}) + self.config = GptOss20BConfig(**cfg_overrides) + self.selected_layers = tuple( + model_options.get("selected_layers", LENS_SELECTED_LAYERS) + ) + self.txt_offset = int(model_options.get("txt_offset", LENS_TXT_OFFSET)) + + # mxfp4_runtime=True keeps experts packed and dequants per hit at forward. + self.mxfp4_runtime = bool(model_options.get("mxfp4_runtime", False)) + + self.transformer = GptOssModel(self.config, device=device, dtype=dtype, ops=operations) + self.num_layers = self.config.num_hidden_layers + self.dtype = dtype + self.execution_device = None + self._pad_token_id = _LENS_PAD_TOKEN_ID + + for p in self.parameters(): + p.requires_grad = False + + def set_clip_options(self, options): + self.execution_device = options.get("execution_device", self.execution_device) + + def reset_clip_options(self): + self.execution_device = None + + def _gather_tokens(self, token_weight_pairs): + ids_list = [[int(t[0]) for t in batch] for batch in token_weight_pairs] + pad_id = self._pad_token_id + max_len = max(len(x) for x in ids_list) + device = self.execution_device + ids = torch.full((len(ids_list), max_len), pad_id, dtype=torch.long, device=device) + mask = torch.zeros((len(ids_list), max_len), dtype=torch.long, device=device) + for i, x in enumerate(ids_list): + ids[i, : len(x)] = torch.tensor(x, dtype=torch.long, device=device) + mask[i, : len(x)] = 1 + return ids, mask + + def encode_token_weights(self, token_weight_pairs): + # Empty negative: emit zero-length features + zero mask + if all(len(batch) == 0 for batch in token_weight_pairs): + device = self.execution_device + B = len(token_weight_pairs) + L = len(self.selected_layers) + H = self.config.hidden_size + flat = torch.zeros(B, 0, L * H, dtype=self.dtype, device=device) + mask = torch.zeros(B, 0, dtype=torch.long, device=device) + return flat, None, {"attention_mask": mask, "num_layers_stacked": L} + + input_ids, attn_mask = self._gather_tokens(token_weight_pairs) + out = self.transformer(input_ids, attention_mask=attn_mask, capture_layers=self.selected_layers) + layers = out["hidden_states"] # list of L × [B, S, H] + stacked = torch.stack(layers, dim=2) # [B, S, L, H] + + offset = self.txt_offset + if stacked.shape[1] > offset: + stacked = stacked[:, offset:].contiguous() + mask_trim = attn_mask[:, offset:] + else: + stacked = stacked[:, :0] + mask_trim = attn_mask[:, :0] + + B, S, L, H = stacked.shape + flat = stacked.reshape(B, S, L * H) + extra = {"attention_mask": mask_trim, "num_layers_stacked": L} + return flat, None, extra + + def load_sd(self, sd): + if any(k.startswith("model.") for k in sd): + sd = {(k[len("model."):] if k.startswith("model.") else k): v for k, v in sd.items()} + sd.pop("lm_head.weight", None) + + if self.mxfp4_runtime: + device = next(self.transformer.parameters()).device + for layer in self.transformer.layers: + layer.mlp.experts.switch_to_mxfp4(device=device) + else: + sd = _dequant_mxfp4_state_dict(sd, self.dtype) + + return self.transformer.load_state_dict(sd, strict=False, assign=True) + + +class LensTEModel(sd1_clip.SD1ClipModel): + def __init__(self, device="cpu", dtype=None, model_options=None): + super().__init__(device=device, dtype=dtype, name="gpt_oss", clip_model=LensGptOssClipModel, model_options=model_options or {}) + + +def lens_te(dtype_llama=None, llama_quantization_metadata=None, mxfp4_runtime=False): + class LensTEModel_(LensTEModel): + def __init__(self, device="cpu", dtype=None, model_options=None): + mo = dict(model_options or {}) + if llama_quantization_metadata is not None: + mo["quantization_metadata"] = llama_quantization_metadata + if dtype_llama is not None: + dtype = dtype_llama + if mxfp4_runtime: + mo["mxfp4_runtime"] = True + super().__init__(device=device, dtype=dtype, model_options=mo) + + return LensTEModel_ diff --git a/comfy_extras/nodes_cfg.py b/comfy_extras/nodes_cfg.py index 4ebb4b51e..b585c560f 100644 --- a/comfy_extras/nodes_cfg.py +++ b/comfy_extras/nodes_cfg.py @@ -57,24 +57,55 @@ class CFGNorm(io.ComfyNode): inputs=[ io.Model.Input("model"), io.Float.Input("strength", default=1.0, min=0.0, max=100.0, step=0.01), + io.Boolean.Input( + "pre_cfg", + default=False, + optional=True, + tooltip=( + "If true, rescale the combined noise BEFORE the sampler's CFG combine, " + "without clamping (can amplify). Matches the norm-scaled CFG used by " + "models like Lens. Default false keeps the original post-CFG x0-space " + "attenuate-only behavior." + ), + ), ], outputs=[io.Model.Output(display_name="patched_model")], is_experimental=True, ) @classmethod - def execute(cls, model, strength) -> io.NodeOutput: + def execute(cls, model, strength, pre_cfg=False) -> io.NodeOutput: m = model.clone() - def cfg_norm(args): - cond_p = args['cond_denoised'] - pred_text_ = args["denoised"] + if pre_cfg: + def cfg_norm_pre(args): + cond = args["cond"] + uncond = args["uncond"] + cond_scale = args["cond_scale"] + comb = uncond + cond_scale * (cond - uncond) + cond_norm = torch.linalg.vector_norm(cond, dim=1, keepdim=True) + comb_norm = torch.linalg.vector_norm(comb, dim=1, keepdim=True) + rescale = torch.where( + comb_norm > 0, + cond_norm / comb_norm.clamp_min(1e-12), + torch.ones_like(comb_norm), + ) + rescaled = comb * rescale + # strength blends back toward standard linear CFG (1.0 = full rescale). + if strength != 1.0: + rescaled = strength * rescaled + (1.0 - strength) * comb + return rescaled + m.set_model_sampler_cfg_function(cfg_norm_pre) + else: + def cfg_norm(args): + cond_p = args['cond_denoised'] + pred_text_ = args["denoised"] - norm_full_cond = torch.norm(cond_p, dim=1, keepdim=True) - norm_pred_text = torch.norm(pred_text_, dim=1, keepdim=True) - scale = (norm_full_cond / (norm_pred_text + 1e-8)).clamp(min=0.0, max=1.0) - return pred_text_ * scale * strength + norm_full_cond = torch.norm(cond_p, dim=1, keepdim=True) + norm_pred_text = torch.norm(pred_text_, dim=1, keepdim=True) + scale = (norm_full_cond / (norm_pred_text + 1e-8)).clamp(min=0.0, max=1.0) + return pred_text_ * scale * strength - m.set_model_sampler_post_cfg_function(cfg_norm) + m.set_model_sampler_post_cfg_function(cfg_norm) return io.NodeOutput(m) diff --git a/nodes.py b/nodes.py index 13e46ac8a..f390fd937 100644 --- a/nodes.py +++ b/nodes.py @@ -961,7 +961,7 @@ class CLIPLoader: @classmethod def INPUT_TYPES(s): return {"required": { "clip_name": (folder_paths.get_filename_list("text_encoders"), ), - "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2", "ovis", "longcat_image", "cogvideox"], ), + "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2", "ovis", "longcat_image", "cogvideox", "lens"], ), }, "optional": { "device": (["default", "cpu"], {"advanced": True}), @@ -971,7 +971,7 @@ class CLIPLoader: CATEGORY = "advanced/loaders" - DESCRIPTION = "[Recipes]\n\nstable_diffusion: clip-l\nstable_cascade: clip-g\nsd3: t5 xxl/ clip-g / clip-l\nstable_audio: t5 base\nmochi: t5 xxl\ncogvideox: t5 xxl (226-token padding)\ncosmos: old t5 xxl\nlumina2: gemma 2 2B\nwan: umt5 xxl\n hidream: llama-3.1 (Recommend) or t5\nomnigen2: qwen vl 2.5 3B" + DESCRIPTION = "[Recipes]\n\nstable_diffusion: clip-l\nstable_cascade: clip-g\nsd3: t5 xxl/ clip-g / clip-l\nstable_audio: t5 base\nmochi: t5 xxl\ncogvideox: t5 xxl (226-token padding)\ncosmos: old t5 xxl\nlumina2: gemma 2 2B\nwan: umt5 xxl\n hidream: llama-3.1 (Recommend) or t5\nomnigen2: qwen vl 2.5 3B\nlens: gpt-oss-20b (MXFP4 dequant on load)" def load_clip(self, clip_name, type="stable_diffusion", device="default"): clip_type = getattr(comfy.sd.CLIPType, type.upper(), comfy.sd.CLIPType.STABLE_DIFFUSION)