feat: --stream-layers for streaming weights from CPU during generation

examples/common/common.cpp

@@ -418,6 +418,10 @@ ArgOptions SDContextParams::get_options() {
     };
 
     options.bool_options = {
+        {"",
+         "--stream-layers",
+         "enable residency+prefetch streaming on top of --max-vram (no effect without --max-vram; defaults to false)",
+         true, &stream_layers},
         {"",
          "--force-sdxl-vae-conv-scale",
          "force use of conv scale on sdxl vae",
@@ -694,6 +698,7 @@ std::string SDContextParams::to_string() const {
         << "  sampler_rng_type: " << sd_rng_type_name(sampler_rng_type) << ",\n"
         << "  offload_params_to_cpu: " << (offload_params_to_cpu ? "true" : "false") << ",\n"
         << "  max_vram: " << max_vram << ",\n"
+        << "  stream_layers: " << (stream_layers ? "true" : "false") << ",\n"
         << "  backend: \"" << backend << "\",\n"
         << "  params_backend: \"" << params_backend << "\",\n"
         << "  enable_mmap: " << (enable_mmap ? "true" : "false") << ",\n"
@@ -773,6 +778,7 @@ sd_ctx_params_t SDContextParams::to_sd_ctx_params_t(bool vae_decode_only, bool f
         chroma_t5_mask_pad,
         qwen_image_zero_cond_t,
         max_vram,
+        stream_layers,
         backend.c_str(),
         params_backend.c_str(),
     };

examples/common/common.h

@@ -112,6 +112,7 @@ struct SDContextParams {
     rng_type_t sampler_rng_type = RNG_TYPE_COUNT;
     bool offload_params_to_cpu  = false;
     float max_vram              = 0.f;
+    bool stream_layers          = false;
     std::string backend;
     std::string params_backend;
     bool enable_mmap           = false;

include/stable-diffusion.h

@@ -213,6 +213,7 @@ typedef struct {
     int chroma_t5_mask_pad;
     bool qwen_image_zero_cond_t;
     float max_vram;  // GiB budget for graph-cut segmented param offload (0 = disabled, -1 = auto free VRAM minus 1 GiB)
+    bool stream_layers;  // Enable residency+prefetch streaming on top of --max-vram (no effect without --max-vram)
     const char* backend;
     const char* params_backend;
 } sd_ctx_params_t;

src/conditioner.hpp

@@ -118,6 +118,7 @@ struct Conditioner {
     virtual void get_param_tensors(std::map<std::string, ggml_tensor*>& tensors)           = 0;
     virtual size_t get_params_buffer_size()                                                = 0;
     virtual void set_max_graph_vram_bytes(size_t max_vram_bytes) {}
+    virtual void set_stream_layers_enabled(bool enabled) {}
     virtual void set_flash_attention_enabled(bool enabled) = 0;
     virtual void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) {}
     virtual std::tuple<SDCondition, std::vector<bool>> get_learned_condition_with_trigger(int n_threads,
@@ -210,6 +211,13 @@ struct FrozenCLIPEmbedderWithCustomWords : public Conditioner {
         }
     }
 
+    void set_stream_layers_enabled(bool enabled) override {
+        text_model->set_stream_layers_enabled(enabled);
+        if (sd_version_is_sdxl(version)) {
+            text_model2->set_stream_layers_enabled(enabled);
+        }
+    }
+
     void set_flash_attention_enabled(bool enabled) override {
         text_model->set_flash_attention_enabled(enabled);
         if (sd_version_is_sdxl(version)) {
@@ -843,6 +851,18 @@ struct SD3CLIPEmbedder : public Conditioner {
         }
     }
 
+    void set_stream_layers_enabled(bool enabled) override {
+        if (clip_l) {
+            clip_l->set_stream_layers_enabled(enabled);
+        }
+        if (clip_g) {
+            clip_g->set_stream_layers_enabled(enabled);
+        }
+        if (t5) {
+            t5->set_stream_layers_enabled(enabled);
+        }
+    }
+
     void set_flash_attention_enabled(bool enabled) override {
         if (clip_l) {
             clip_l->set_flash_attention_enabled(enabled);
@@ -1201,6 +1221,15 @@ struct FluxCLIPEmbedder : public Conditioner {
         }
     }
 
+    void set_stream_layers_enabled(bool enabled) override {
+        if (clip_l) {
+            clip_l->set_stream_layers_enabled(enabled);
+        }
+        if (t5) {
+            t5->set_stream_layers_enabled(enabled);
+        }
+    }
+
     void set_flash_attention_enabled(bool enabled) override {
         if (clip_l) {
             clip_l->set_flash_attention_enabled(enabled);
@@ -1435,6 +1464,12 @@ struct T5CLIPEmbedder : public Conditioner {
         }
     }
 
+    void set_stream_layers_enabled(bool enabled) override {
+        if (t5) {
+            t5->set_stream_layers_enabled(enabled);
+        }
+    }
+
     void set_flash_attention_enabled(bool enabled) override {
         if (t5) {
             t5->set_flash_attention_enabled(enabled);
@@ -1618,6 +1653,10 @@ struct AnimaConditioner : public Conditioner {
         llm->set_max_graph_vram_bytes(max_vram_bytes);
     }
 
+    void set_stream_layers_enabled(bool enabled) override {
+        llm->set_stream_layers_enabled(enabled);
+    }
+
     void set_flash_attention_enabled(bool enabled) override {
         llm->set_flash_attention_enabled(enabled);
     }
@@ -1762,6 +1801,10 @@ struct LLMEmbedder : public Conditioner {
         llm->set_max_graph_vram_bytes(max_vram_bytes);
     }
 
+    void set_stream_layers_enabled(bool enabled) override {
+        llm->set_stream_layers_enabled(enabled);
+    }
+
     void set_flash_attention_enabled(bool enabled) override {
         llm->set_flash_attention_enabled(enabled);
     }