Upload README.md with huggingface_hub

README.md

@@ -1,44 +1,41 @@
----
-tags:
-- activation-avatars
-- milady
-- flux
-- adapter
-- qwen3
-license: mit
-pipeline_tag: image-generation
----
-
-# Milady Avatar Adapter
-
-A CrossAttention adapter that maps Qwen3-4B LLM activations to FLUX.2-Klein
-prompt embedding space, producing Milady NFT-style avatar expressions in real-time.
+# Milady Avatar Adapter (SDXL)
 
-## Architecture
-
-- **Input**: Qwen3-4B hidden states from layers [9, 18, 27] (7680 dims)
-- **Layer Weighting**: Learned weighted combination → 2560 dims
-- **Adapter**: CrossAttentionAdapter (rank=128, 64 output tokens, 2 decoder layers)
-- **Output**: 64 × 7680 prompt embeddings for FLUX.2-Klein
+Neural adapter that maps Qwen3-4B language model activations to SDXL prompt embedding space,
+enabling real-time emotional avatar generation in the Milady art style.
 
-## Usage
+## Architecture
 
-```python
-from adapter import load_adapter
+### SDXL Adapter (NEW - Higher Quality)
+- **Input**: Qwen3-4B hidden states from layers [9, 18, 27] → 7680 dims
+- **Layer Weighting**: Learned weighted combination → 2560 dims  
+- **Cross-Attention Decoder**: 3-layer transformer decoder with 8 heads
+- **Output**: SDXL prompt embeddings [77, 2048] + pooled embeddings [1280]
+- **Parameters**: 5.28M
+- **Training**: 500 epochs on 200 emotion-labeled samples, MSE loss
+- **Best Val Loss**: 6.762
 
-adapter = load_adapter("milady_adapter_best.pt", device="cuda:0")
-# activation: [7680] tensor from Qwen3-4B layers [9,18,27]
-expression = adapter(activation, emotion_scale=4.0)
-# expression: [64, 7680] — concatenate with base prompt_embeds for Klein
+### Pipeline
 ```
+Emotional Text → Qwen3-4B (hooks on layers 9,18,27) → Adapter → SDXL + Milady LoRA → Avatar Image
+```
+
+## Files
 
-## Training
+### SDXL Version (Recommended)
+- `sdxl/best_sdxl_adapter.pt` - Trained adapter weights
+- `sdxl/sdxl_adapter.py` - Adapter architecture
+- `sdxl/test_sdxl_pipeline.py` - End-to-end inference script
+- `sdxl/train_sdxl_adapter.py` - Training script
 
-- Trained on 200 Milady reference images
-- Transfer learning from existing activation-avatars adapter
-- 80 epochs, AdamW optimizer, cosine+MSE loss
-- Style descriptions mapped through Klein's text encoder (layers 10,20,30)
+### Klein Version (Legacy)
+- `adapters/` - Original FLUX.2-Klein adapter weights
 
-## Samples
+## Requirements
+- SDXL base model: `stabilityai/stable-diffusion-xl-base-1.0`
+- Milady LoRA: CivitAI Milady SDXL LoRA
+- Qwen3-4B: `Qwen/Qwen3-4B`
+- Python packages: `torch`, `transformers`, `diffusers`, `safetensors`
 
-See test_results/ for generated avatar samples across 10 emotions.
+## Emotions Supported
+20 emotions: happy, sad, angry, surprised, scared, disgusted, neutral, excited, calm, anxious,
+confident, shy, proud, loving, jealous, curious, bored, amused, thoughtful, determined