Skip to content

Wip: Longcat-Image support #1053

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
stduhpf wants to merge 13 commits into
base: master
Choose a base branch
from
+369 −106
Draft

Wip: Longcat-Image support #1053

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
80784fb
add Flux.2 VAE proj matrix for previews
stduhpf Dec 2, 2025
beef322
Enable flux.2 proj for preview with flux model
stduhpf Dec 2, 2025
77e4620
support Flux.2 patched latents for proj preview
stduhpf Dec 2, 2025
da8e95e
move latent shuffle logic to latents-preview.h
stduhpf Dec 2, 2025
c054c23
refactor preview_latent_video to support flux.2 patchified latents
stduhpf Dec 2, 2025
e47c8c4
Support LongCat Image model
stduhpf Dec 5, 2025
2c44904
temp fix cuda error on quant concat for splitlinear
stduhpf Dec 5, 2025
6e9f5ec
Merge branch 'flux2-proj' into longcat
stduhpf Dec 5, 2025
00071aa
pre-patchify
stduhpf Dec 6, 2025
7711efb
longcat rope ids
stduhpf Dec 6, 2025
535543a
Fix diffusers_style detection
stduhpf Dec 6, 2025
61b0dcf
Flux: simplify when patch_size is 1
stduhpf Dec 6, 2025
deaf939
correct rope offset for image tokens
stduhpf Dec 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions conditioner.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1825,6 +1825,17 @@ struct LLMEmbedder : public Conditioner {
prompt_attn_range.second = prompt.size();

prompt += "[/INST]";
} else if (sd_version_is_longcat(version)) {
prompt_template_encode_start_idx = 36;
// prompt_template_encode_end_idx = 5;

prompt = "<|im_start|>system\nAs an image captioning expert, generate a descriptive text prompt based on an image content, suitable for input to a text-to-image model.<|im_end|>\n<|im_start|>user\n";

prompt_attn_range.first = static_cast<int>(prompt.size());
prompt += conditioner_params.text;
prompt_attn_range.second = static_cast<int>(prompt.size());

prompt += "<|im_end|>\n<|im_start|>assistant\n";
} else {
prompt_template_encode_start_idx = 34;

Expand Down
71 changes: 53 additions & 18 deletions flux.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,14 +88,19 @@ namespace Flux {

public:
SelfAttention(int64_t dim,
int64_t num_heads = 8,
bool qkv_bias = false,
bool proj_bias = true)
int64_t num_heads = 8,
bool qkv_bias = false,
bool proj_bias = true,
bool diffusers_style = false)
: num_heads(num_heads) {
int64_t head_dim = dim / num_heads;
blocks["qkv"] = std::shared_ptr<GGMLBlock>(new Linear(dim, dim * 3, qkv_bias));
blocks["norm"] = std::shared_ptr<GGMLBlock>(new QKNorm(head_dim));
blocks["proj"] = std::shared_ptr<GGMLBlock>(new Linear(dim, dim, proj_bias));
if (diffusers_style) {
blocks["qkv"] = std::shared_ptr<GGMLBlock>(new SplitLinear(dim, {dim, dim, dim}, qkv_bias));
} else {
blocks["qkv"] = std::shared_ptr<GGMLBlock>(new Linear(dim, dim * 3, qkv_bias));
}
blocks["norm"] = std::shared_ptr<GGMLBlock>(new QKNorm(head_dim));
blocks["proj"] = std::shared_ptr<GGMLBlock>(new Linear(dim, dim, proj_bias));
}

std::vector<struct ggml_tensor*> pre_attention(GGMLRunnerContext* ctx, struct ggml_tensor* x) {
Expand Down Expand Up @@ -210,7 +215,8 @@ namespace Flux {
bool prune_mod = false,
bool share_modulation = false,
bool mlp_proj_bias = true,
bool use_mlp_silu_act = false)
bool use_mlp_silu_act = false,
bool diffusers_style = false)
: idx(idx), prune_mod(prune_mod), use_mlp_silu_act(use_mlp_silu_act) {
int64_t mlp_hidden_dim = hidden_size * mlp_ratio;
int64_t mlp_mult_factor = use_mlp_silu_act ? 2 : 1;
Expand All @@ -219,7 +225,7 @@ namespace Flux {
blocks["img_mod"] = std::shared_ptr<GGMLBlock>(new Modulation(hidden_size, true));
}
blocks["img_norm1"] = std::shared_ptr<GGMLBlock>(new LayerNorm(hidden_size, 1e-6f, false));
blocks["img_attn"] = std::shared_ptr<GGMLBlock>(new SelfAttention(hidden_size, num_heads, qkv_bias, mlp_proj_bias));
blocks["img_attn"] = std::shared_ptr<GGMLBlock>(new SelfAttention(hidden_size, num_heads, qkv_bias, mlp_proj_bias, diffusers_style));

blocks["img_norm2"] = std::shared_ptr<GGMLBlock>(new LayerNorm(hidden_size, 1e-6f, false));
blocks["img_mlp.0"] = std::shared_ptr<GGMLBlock>(new Linear(hidden_size, mlp_hidden_dim * mlp_mult_factor, mlp_proj_bias));
Expand All @@ -230,7 +236,7 @@ namespace Flux {
blocks["txt_mod"] = std::shared_ptr<GGMLBlock>(new Modulation(hidden_size, true));
}
blocks["txt_norm1"] = std::shared_ptr<GGMLBlock>(new LayerNorm(hidden_size, 1e-6f, false));
blocks["txt_attn"] = std::shared_ptr<GGMLBlock>(new SelfAttention(hidden_size, num_heads, qkv_bias, mlp_proj_bias));
blocks["txt_attn"] = std::shared_ptr<GGMLBlock>(new SelfAttention(hidden_size, num_heads, qkv_bias, mlp_proj_bias, diffusers_style));

blocks["txt_norm2"] = std::shared_ptr<GGMLBlock>(new LayerNorm(hidden_size, 1e-6f, false));
blocks["txt_mlp.0"] = std::shared_ptr<GGMLBlock>(new Linear(hidden_size, mlp_hidden_dim * mlp_mult_factor, mlp_proj_bias));
Expand Down Expand Up @@ -383,6 +389,7 @@ namespace Flux {
int idx = 0;
bool use_mlp_silu_act;
int64_t mlp_mult_factor;
bool diffusers_style = false;

public:
SingleStreamBlock(int64_t hidden_size,
Expand All @@ -393,7 +400,8 @@ namespace Flux {
bool prune_mod = false,
bool share_modulation = false,
bool mlp_proj_bias = true,
bool use_mlp_silu_act = false)
bool use_mlp_silu_act = false,
bool diffusers_style = false)
: hidden_size(hidden_size), num_heads(num_heads), idx(idx), prune_mod(prune_mod), use_mlp_silu_act(use_mlp_silu_act) {
int64_t head_dim = hidden_size / num_heads;
float scale = qk_scale;
Expand All @@ -405,8 +413,11 @@ namespace Flux {
if (use_mlp_silu_act) {
mlp_mult_factor = 2;
}

blocks["linear1"] = std::shared_ptr<GGMLBlock>(new Linear(hidden_size, hidden_size * 3 + mlp_hidden_dim * mlp_mult_factor, mlp_proj_bias));
if (diffusers_style) {
blocks["linear1"] = std::shared_ptr<GGMLBlock>(new SplitLinear(hidden_size, {hidden_size, hidden_size, hidden_size, mlp_hidden_dim * mlp_mult_factor}, mlp_proj_bias));
} else {
blocks["linear1"] = std::shared_ptr<GGMLBlock>(new Linear(hidden_size, hidden_size * 3 + mlp_hidden_dim * mlp_mult_factor, mlp_proj_bias));
}
blocks["linear2"] = std::shared_ptr<GGMLBlock>(new Linear(hidden_size + mlp_hidden_dim, hidden_size, mlp_proj_bias));
blocks["norm"] = std::shared_ptr<GGMLBlock>(new QKNorm(head_dim));
blocks["pre_norm"] = std::shared_ptr<GGMLBlock>(new LayerNorm(hidden_size, 1e-6f, false));
Expand Down Expand Up @@ -728,6 +739,7 @@ namespace Flux {
bool share_modulation = false;
bool use_mlp_silu_act = false;
float ref_index_scale = 1.f;
bool diffusers_style = false;
ChromaRadianceParams chroma_radiance_params;
};

Expand Down Expand Up @@ -770,7 +782,8 @@ namespace Flux {
params.is_chroma,
params.share_modulation,
!params.disable_bias,
params.use_mlp_silu_act);
params.use_mlp_silu_act,
params.diffusers_style);
}

for (int i = 0; i < params.depth_single_blocks; i++) {
Expand All @@ -782,7 +795,8 @@ namespace Flux {
params.is_chroma,
params.share_modulation,
!params.disable_bias,
params.use_mlp_silu_act);
params.use_mlp_silu_act,
params.diffusers_style);
}

if (params.version == VERSION_CHROMA_RADIANCE) {
Expand Down Expand Up @@ -829,6 +843,11 @@ namespace Flux {
int64_t C = x->ne[2];
int64_t H = x->ne[1];
int64_t W = x->ne[0];
if (params.patch_size == 1) {
x = ggml_reshape_3d(ctx, x, H * W, C, N); // [N, C, H*W]
x = ggml_cont(ctx, ggml_permute(ctx, x, 1, 0, 2, 3)); // [N, H*W, C]
return x;
}
int64_t p = params.patch_size;
int64_t h = H / params.patch_size;
int64_t w = W / params.patch_size;
Expand Down Expand Up @@ -863,6 +882,12 @@ namespace Flux {
int64_t W = w * params.patch_size;
int64_t p = params.patch_size;

if (params.patch_size == 1) {
x = ggml_cont(ctx, ggml_permute(ctx, x, 1, 0, 2, 3)); // [N, C, H*W]
x = ggml_reshape_4d(ctx, x, W, H, C, N); // [N, C, H, W]
return x;
}

GGML_ASSERT(C * p * p == x->ne[0]);

x = ggml_reshape_4d(ctx, x, p * p, C, w * h, N); // [N, h*w, C, p*p]
Expand Down Expand Up @@ -1222,6 +1247,10 @@ namespace Flux {
flux_params.share_modulation = true;
flux_params.ref_index_scale = 10.f;
flux_params.use_mlp_silu_act = true;
} else if (sd_version_is_longcat(version)) {
flux_params.context_in_dim = 3584;
flux_params.vec_in_dim = 0;
flux_params.patch_size = 1;
}
for (auto pair : tensor_storage_map) {
std::string tensor_name = pair.first;
Expand All @@ -1231,6 +1260,9 @@ namespace Flux {
// not schnell
flux_params.guidance_embed = true;
}
if (tensor_name.find("model.diffusion_model.single_blocks.0.linear1.weight.1") != std::string::npos) {
flux_params.diffusers_style = true;
}
if (tensor_name.find("distilled_guidance_layer.in_proj.weight") != std::string::npos) {
// Chroma
flux_params.is_chroma = true;
Expand Down Expand Up @@ -1260,6 +1292,10 @@ namespace Flux {
LOG_INFO("Flux guidance is disabled (Schnell mode)");
}

if (flux_params.diffusers_style) {
LOG_INFO("Using diffusers-style attention blocks");
}

flux = Flux(flux_params);
flux.init(params_ctx, tensor_storage_map, prefix);
}
Expand Down Expand Up @@ -1363,7 +1399,6 @@ namespace Flux {
for (int i = 0; i < ref_latents.size(); i++) {
ref_latents[i] = to_backend(ref_latents[i]);
}

pe_vec = Rope::gen_flux_pe(x->ne[1],
x->ne[0],
flux_params.patch_size,
Expand All @@ -1373,10 +1408,10 @@ namespace Flux {
sd_version_is_flux2(version) ? true : increase_ref_index,
flux_params.ref_index_scale,
flux_params.theta,
flux_params.axes_dim);
flux_params.axes_dim,
sd_version_is_longcat(version));
int pos_len = pe_vec.size() / flux_params.axes_dim_sum / 2;
// LOG_DEBUG("pos_len %d", pos_len);
auto pe = ggml_new_tensor_4d(compute_ctx, GGML_TYPE_F32, 2, 2, flux_params.axes_dim_sum / 2, pos_len);
auto pe = ggml_new_tensor_4d(compute_ctx, GGML_TYPE_F32, 2, 2, flux_params.axes_dim_sum / 2, pos_len);
// pe->data = pe_vec.data();
// print_ggml_tensor(pe);
// pe->data = nullptr;
Expand Down
77 changes: 77 additions & 0 deletions ggml_extend.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2159,6 +2159,83 @@ class Linear : public UnaryBlock {
}
};

class SplitLinear : public Linear {
protected:
int64_t in_features;
std::vector<int64_t> out_features_vec;
bool bias;
bool force_f32;
bool force_prec_f32;
float scale;
std::string prefix;

void init_params(struct ggml_context* ctx, const String2TensorStorage& tensor_storage_map = {}, const std::string prefix = "") override {
this->prefix = prefix;
enum ggml_type wtype = get_type(prefix + "weight", tensor_storage_map, GGML_TYPE_F32);
if (in_features % ggml_blck_size(wtype) != 0 || force_f32) {
wtype = GGML_TYPE_F32;
}
params["weight"] = ggml_new_tensor_2d(ctx, wtype, in_features, out_features_vec[0]);
for (int i = 1; i < out_features_vec.size(); i++) {
// most likely same type as the first weight
params["weight." + std::to_string(i)] = ggml_new_tensor_2d(ctx, wtype, in_features, out_features_vec[i]);
}
if (bias) {
enum ggml_type wtype = GGML_TYPE_F32;
params["bias"] = ggml_new_tensor_1d(ctx, wtype, out_features_vec[0]);
for (int i = 1; i < out_features_vec.size(); i++) {
params["bias." + std::to_string(i)] = ggml_new_tensor_1d(ctx, wtype, out_features_vec[i]);
}
}
}

public:
SplitLinear(int64_t in_features,
std::vector<int64_t> out_features_vec,
bool bias = true,
bool force_f32 = false,
bool force_prec_f32 = false,
float scale = 1.f)
: Linear(in_features, out_features_vec[0], bias, force_f32, force_prec_f32, scale),
in_features(in_features),
out_features_vec(out_features_vec),
bias(bias),
force_f32(force_f32),
force_prec_f32(force_prec_f32),
scale(scale) {}

struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x) {
struct ggml_tensor* w = params["weight"];
struct ggml_tensor* b = nullptr;
if (bias) {
b = params["bias"];
}
if (ctx->weight_adapter) {
// concat all weights and biases together so it runs in one linear layer
for (int i = 1; i < out_features_vec.size(); i++) {
w = ggml_concat(ctx->ggml_ctx, w, params["weight." + std::to_string(i)], 1);
if (bias) {
b = ggml_concat(ctx->ggml_ctx, b, params["bias." + std::to_string(i)], 0);
}
}
WeightAdapter::ForwardParams forward_params;
forward_params.op_type = WeightAdapter::ForwardParams::op_type_t::OP_LINEAR;
forward_params.linear.force_prec_f32 = force_prec_f32;
forward_params.linear.scale = scale;
return ctx->weight_adapter->forward_with_lora(ctx->ggml_ctx, x, w, b, prefix, forward_params);
}
auto x0 = ggml_ext_linear(ctx->ggml_ctx, x, w, b, force_prec_f32, scale);
for (int i = 1; i < out_features_vec.size(); i++) {
auto wi = params["weight." + std::to_string(i)];
auto bi = bias ? params["bias." + std::to_string(i)] : nullptr;
auto xi = ggml_ext_linear(ctx->ggml_ctx, x, wi, bi, force_prec_f32, scale);
x0 = ggml_concat(ctx->ggml_ctx, x0, xi, 0);
}

return x0;
}
};

__STATIC_INLINE__ bool support_get_rows(ggml_type wtype) {
std::set<ggml_type> allow_types = {GGML_TYPE_F16, GGML_TYPE_Q8_0, GGML_TYPE_Q5_1, GGML_TYPE_Q5_0, GGML_TYPE_Q4_1, GGML_TYPE_Q4_0};
if (allow_types.find(wtype) != allow_types.end()) {
Expand Down
Loading
Loading