Info
Various quantizations for hf:ostris/Flex.2-preview.
Safetensors
| Filename | Quant Type | File Size | Description | Example Image |
|---|---|---|---|---|
| Flex.2-preview-fp8_e4m3fn_scaled.safetensors | F8_E4M3FN | 8.17GB | Scale per weight tensor | - |
| Flex.2-preview-fp8_e5m2_scaled.safetensors | F8_E5M2 | 8.17GB | Scale per weight tensor | - |
| Flex.2-preview-fp8_e4m3fn.safetensors | F8_E4M3FN | 8.16GB | - | - |
| Flex.2-preview-fp8_e5m2.safetensors | F8_E5M2 | 8.16GB | - | - |
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 | BF16 | 16.3GB | - | - |
| Flex.2-preview-Q8_0.gguf | Q8_0 | 8.68GB | TBC | - |
| Flex.2-preview-Q6_K.gguf | Q6_K | 6.70GB | TBC | - |
| Flex.2-preview-Q5_1.gguf | Q5_1 | 6.13GB | TBC | - |
| Flex.2-preview-Q5_0.gguf | Q5_0 | 5.62GB | TBC | - |
| Flex.2-preview-Q4_1.gguf | Q4_1 | 5.11GB | TBC | - |
| Flex.2-preview-IQ4_NL.gguf | IQ4_NL | 4.60GB | TBC | - |
| Flex.2-preview-Q4_0.gguf | Q4_0 | 4.60GB | TBC | - |
| 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 | Q8_0 | 8.39GB | TBC | - |
| Flex.2-preview-fluxified-Q6_K.gguf | Q6_K | 6.74GB | TBC | - |
| Flex.2-preview-fluxified-Q5_1.gguf | Q5_1 | 6.19GB | TBC | - |
| Flex.2-preview-fluxified-Q5_0.gguf | Q5_0 | 5.70GB | TBC | - |
| Flex.2-preview-fluxified-Q5_K_S.gguf | Q5_K_S | 5.67GB | TBC | - |
| Flex.2-preview-fluxified-Q4_1.gguf | Q4_1 | 5.22GB | TBC | - |
| Flex.2-preview-fluxified-Q4_0.gguf | Q4_0 | 4.72GB | TBC | - |
| Flex.2-preview-fluxified-Q4_K_S.gguf | Q4_K_S | 4.58GB | TBC | - |
| Flex.2-preview-fluxified-Q3_K_S.gguf | Q3_K_S | 3.52GB | TBC | - |
Fluxified GGUF Imatrix
- Fluxified GGUF + Importance Matrix
| Filename | Quant type | File Size | Description / L2 Loss Step 25 | Example Image |
|---|---|---|---|---|
| Flex.2-preview-fluxified-IQ4_XS.gguf | IQ4_XS | 4.37GB | TBC | - |
| Flex.2-preview-fluxified-IQ4_NL.gguf | IQ4_NL | 4.58GB | TBC | - |
| Flex.2-preview-fluxified-Q4_K_S.gguf | Q4_K_S | 4.58GB | TBC | - |
| Flex.2-preview-fluxified-Q3_K_S.gguf | Q3_K_S | 3.52GB | TBC | - |
| Flex.2-preview-fluxified-Q2_K_S.gguf | Q2_K_S | 2.82GB | TBC | - |
- Downloads last month
- 1,992
Hardware compatibility
Log In
to view the estimation
1-bit
2-bit
3-bit
4-bit
5-bit
6-bit
8-bit
16-bit
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support
Model tree for Eviation/Flex.2-preview
Base model
ostris/Flex.2-preview