Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
base_model:
|
| 6 |
+
- black-forest-labs/FLUX.2-klein-4B
|
| 7 |
+
tags:
|
| 8 |
+
- art
|
| 9 |
+
---
|
| 10 |
+
# Just some testing
|
| 11 |
+
|
| 12 |
+
Pin down what is needed between 50s,40s,30s cards
|
| 13 |
+
|
| 14 |
+
Translate info for use in TensorRT fp4 converting / onnx
|
| 15 |
+
|
| 16 |
+
Anyone that has made fp4 onnx before with NVIDIA’s Model Optimizer PTQ help in the Community tab is welcomed.
|
| 17 |
+
|
| 18 |
+
Currently working on correct Quantize/Dequantize nodes for onnx export to build fp4 RT engine for 50s cards.
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
## Klein 4B NVFP4 Variants — `txt_attn` dtype notes
|
| 22 |
+
|
| 23 |
+
### Klein4b-nvfp4.safetensors
|
| 24 |
+
- `txt_attn` weights: **NVFP4**
|
| 25 |
+
|
| 26 |
+
### Klein4b-nvfp4_BF16.safetensors
|
| 27 |
+
- `txt_attn` weights: **BF16**
|
| 28 |
+
|
| 29 |
+
### Klein4b-nvfp4_fp8e4m3fn_absmax.safetensors
|
| 30 |
+
- `txt_attn` weights: **FP8 (float8_e4m3fn)**
|
| 31 |
+
- `txt_attn` scaling: **absmax**
|
| 32 |
+
|
| 33 |
+
### Klein4b-nvfp4_fp8e5m2_absmax.safetensors
|
| 34 |
+
- `txt_attn` weights: **FP8 (float8_e5m2)**
|
| 35 |
+
- `txt_attn` scaling: **absmax**
|