Text-to-Image
Diffusers
PyTorch
flow-matching
rectified-flow
dit
dual-stream-dit
sprint
anime
aipracticecafe commited on
Commit
7b2cee9
·
verified ·
1 Parent(s): ad580cf

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -6
README.md CHANGED
@@ -47,6 +47,8 @@ control (panning and zoom) at inference time.
47
  - **Objective**: Conditional Flow Matching (Logit-Normal timestep sampling)
48
  - **License**: OpenRAIL++
49
 
 
 
50
  ---
51
 
52
  ## Key Architecture & Features
@@ -57,8 +59,8 @@ The base architecture is based on the results from (i1: A Simple and Fully Open
57
  per-block AdaLN modulation, reducing parameter overhead.
58
  - **Sandwich Normalization & SwiGLU**: RMSNorm placed before and after
59
  attention and FFN operations ensures training stability in bfloat16.
60
- - **SPRINT Acceleration**: Employs structured group-wise $2 \times 2$ patch
61
- dropping ($75\%$ drop ratio during early pretraining), cutting compute
62
  demands by half while retaining global semantic context.
63
 
64
  ### 2. HDM Continuous Aspect-Normalized RoPE (Camera Control)
@@ -80,8 +82,8 @@ framing at inference time:
80
  ### 3. Components
81
  - **Text Encoder**: `LiquidAI/LFM2.5-230M` (unidirectional causal LM hidden
82
  states passed via a 2-layer Transformer Text Adapter).
83
- - **Autoencoder**: `kaiyuyue/FLUX.2-dev-vae` ($C_{\text{lat}}=32$, $f=8$,
84
- patch size $p=2$, effective image token size $16 \times 16$).
85
 
86
  ---
87
 
@@ -99,8 +101,8 @@ resolution with shifted square cropping and cached in-RAM latents.
99
 
100
  - **Total Epochs**: 175
101
  - **Optimizer**: 8-bit AdamW (`bitsandbytes`), $\beta=(0.9, 0.95)$, WD=0.01
102
- - **Learning Rate**: $5 \times 10^{-4}$ with warmup and cosine annealing only for last stage
103
- - **Effective Batch Size**: 256 (Batch size 64 $\times$ 4 gradient accumulation)
104
  - **Precision**: Mixed Precision (`torch.bfloat16` with `torch.compile`)
105
 
106
  ---
 
47
  - **Objective**: Conditional Flow Matching (Logit-Normal timestep sampling)
48
  - **License**: OpenRAIL++
49
 
50
+ <Gallery />
51
+
52
  ---
53
 
54
  ## Key Architecture & Features
 
59
  per-block AdaLN modulation, reducing parameter overhead.
60
  - **Sandwich Normalization & SwiGLU**: RMSNorm placed before and after
61
  attention and FFN operations ensures training stability in bfloat16.
62
+ - **SPRINT Acceleration**: Employs structured group-wise 2x2 patch
63
+ dropping (75% drop ratio during early pretraining), cutting compute
64
  demands by half while retaining global semantic context.
65
 
66
  ### 2. HDM Continuous Aspect-Normalized RoPE (Camera Control)
 
82
  ### 3. Components
83
  - **Text Encoder**: `LiquidAI/LFM2.5-230M` (unidirectional causal LM hidden
84
  states passed via a 2-layer Transformer Text Adapter).
85
+ - **Autoencoder**: `kaiyuyue/FLUX.2-dev-vae` (C=32, f=8,
86
+ patch size p=2, effective image token size 16x16).
87
 
88
  ---
89
 
 
101
 
102
  - **Total Epochs**: 175
103
  - **Optimizer**: 8-bit AdamW (`bitsandbytes`), $\beta=(0.9, 0.95)$, WD=0.01
104
+ - **Learning Rate**: 5e-4 with warmup and cosine annealing only for last stage
105
+ - **Effective Batch Size**: 256 (Batch size 64x4 gradient accumulation)
106
  - **Precision**: Mixed Precision (`torch.bfloat16` with `torch.compile`)
107
 
108
  ---