Update README.md
Browse files
README.md
CHANGED
|
@@ -55,7 +55,7 @@ W_bf16 = W_fp8.to(bfloat16) * scale # dequantize at inference
|
|
| 55 |
## Hardware Requirements
|
| 56 |
|
| 57 |
- **GPU:** NVIDIA GPU with CUDA support
|
| 58 |
-
- **VRAM:** ~
|
| 59 |
- **Native FP8 tensor cores:** Ada Lovelace or Blackwell (RTX 4090, RTX 5090, H100, etc.) — recommended for full speed
|
| 60 |
- **Older GPUs (Ampere and below):** Will load and run correctly. Dequantization to bfloat16 happens on all hardware, so you still get the reduced VRAM footprint even without native FP8 cores.
|
| 61 |
|
|
|
|
| 55 |
## Hardware Requirements
|
| 56 |
|
| 57 |
- **GPU:** NVIDIA GPU with CUDA support
|
| 58 |
+
- **VRAM:** ~8 GB
|
| 59 |
- **Native FP8 tensor cores:** Ada Lovelace or Blackwell (RTX 4090, RTX 5090, H100, etc.) — recommended for full speed
|
| 60 |
- **Older GPUs (Ampere and below):** Will load and run correctly. Dequantization to bfloat16 happens on all hardware, so you still get the reduced VRAM footprint even without native FP8 cores.
|
| 61 |
|