File size: 3,345 Bytes
cd964a7
 
 
085561b
cd964a7
 
 
 
96d26ec
9b8a00e
 
085561b
 
9b4b075
 
 
 
96d26ec
 
1299047
9b4b075
96d26ec
2cc956b
96d26ec
 
f70680d
9b8a00e
 
c1c5c0d
547ccfc
5817bbf
 
9b8a00e
 
c1c5c0d
9b8a00e
c1c5c0d
 
 
547ccfc
a0818d2
f1d9108
c77aabc
183b69b
a00aa7c
c1c5c0d
5817bbf
f70680d
5817bbf
c1c5c0d
 
5817bbf
f70680d
5817bbf
 
 
 
 
f70680d
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
---
license: apache-2.0
pipeline_tag: text-to-image
library_name: gguf
tags:
- flex
- flux
- gguf
- safetensors
base_model:
- ostris/Flex.2-preview
---

# Info

Various quantizations for [hf:ostris/Flex.2-preview](https://huggingface.co/ostris/Flex.2-preview/).

# Safetensors

| Filename | Quant Type | File Size | Description | Example Image |
| -------- | ---------- | --------- | ----------- | ------------- |
| [Flex.2-preview-fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Eviation/Flex.2-preview/blob/main/Flex.2-preview-fp8_e4m3fn_scaled.safetensors) | FP8 E4M3FN | 8.17GB | - | - |
| [Flex.2-preview-fp8_e5m2_scaled.safetensors](https://huggingface.co/Eviation/Flex.2-preview/blob/main/Flex.2-preview-fp8_e5m2_scaled.safetensors) | FP8 E5M2 | 8.17GB | - | - |


# Pure GGUF

- pure, conversion from safetensors BF16 via F32 gguf
- architecture: flex.2 (as not all tensor shapes match to flux)
- no imatrix was used to quantize
- biases and norms: F32
- img_in.weight: BF16 (due to tensor shape and block sizes)
- everything else according to file type

| Filename | Quant Type | File Size | Description / L2 Loss Step 25 | Example Image |
| -------- | ---------- | --------- | ----------------------------- | ------------- |
| [Flex.2-preview-BF16.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/Flex.2-preview-BF16.gguf) | BF16 | 16.3GB | - | - |
| [Flex.2-preview-Q8_0.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-Q8_0.gguf) | Q8_0 | 8.68GB | TBC | - |
| [Flex.2-preview-Q6_K.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-Q6_K.gguf) | Q6_K | 6.70GB | TBC | - |
| [Flex.2-preview-Q5_1.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-Q5_1.gguf) | Q5_1 | 6.13GB | TBC | - |
| [Flex.2-preview-Q5_0.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-Q5_0.gguf) | Q5_0 | 5.62GB | TBC | - |
| [Flex.2-preview-Q4_1.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-Q4_1.gguf) | Q4_1 | 5.11GB | TBC | - |
| [Flex.2-preview-IQ4_NL.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-IQ4_NL.gguf) | IQ4_NL | 4.60GB | TBC | - |
| [Flex.2-preview-Q4_0.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-Q4_0.gguf) | Q4_0 | 4.60GB | TBC | - |
| [Flex.2-preview-Q3_K_S.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/pure/Flex.2-preview-Q3_K_S.gguf) | Q3_K_S | 3.52GB | TBC | - |


# Fluxified GGUF

- conversion from safetensors BF16 via F32 gguf
- truncated img_in.weight tensor to first 16 latent channels
- lost ability to do inpainting and process control image
- should be a drop-in replacement for FLUX
- architecture: flux
- dynamic quantization?

| Filename | Quant type | File Size | Description / L2 Loss Step 25 | Example Image |
| -------- | ---------- | --------- | ----------------------------- | ------------- |
| [Flex.2-preview-fluxified-Q8_0.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/fluxified/Flex.2-preview-fluxified-Q8_0.gguf) | Q8_0 | 8.39GB | TBC | - |
| [Flex.2-preview-fluxified-Q3_K_S.gguf](https://huggingface.co/Eviation/Flex.2-preview/blob/main/fluxified/Flex.2-preview-fluxified-Q3_K_S.gguf) | Q3_K_S | 3.52GB | TBC | - |