Model card: original fp32 / fp16 accuracy columns (stereo), table formatting
Browse files
README.md
CHANGED
|
@@ -48,7 +48,7 @@ Tested up to (portable, fp16, RTX 5060 Ti 16 GB): `efficientvit_b2.r288_in1k` 28
|
|
| 48 |
|
| 49 |
| model | time (ms): original fp32 | original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
|
| 50 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
| 51 |
-
| `efficientvit_b2.r288_in1k` | 10.
|
| 52 |
|
| 53 |
### Startup and first call
|
| 54 |
|
|
@@ -60,7 +60,7 @@ Tested up to (portable, fp16, RTX 5060 Ti 16 GB): `efficientvit_b2.r288_in1k` 28
|
|
| 60 |
|
| 61 |
| model | input | portable ms | compile ms | compile s (cold) | TensorRT fp16 ms | ONNX + build s | kernels ms | peak MiB (alloc / reserved): portable | compile | TensorRT (device) |
|
| 62 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
| 63 |
-
| `efficientvit_b2.r288_in1k` | 224x224 | 11.9 | 6.9 | 11.0 | 0.7 | 39 | 1.
|
| 64 |
|
| 65 |
### Fixed-size speed and peak memory, all models
|
| 66 |
|
|
@@ -77,7 +77,7 @@ Hardware, protocol and caveats: [benchmarks](https://github.com/kleinicke/fastfo
|
|
| 77 |
|
| 78 |
### Parity with the original code
|
| 79 |
|
| 80 |
-
Every variant is checked against
|
| 81 |
|
| 82 |
## Usage
|
| 83 |
|
|
@@ -96,4 +96,4 @@ Install with `pip install fastforks` (see <https://github.com/kleinicke/fastfork
|
|
| 96 |
|
| 97 |
`classification.load("efficientvit_b2.r288_in1k", mode="kernels")` runs our CUDA kernels (`cudnn_hacc`, `fvm_conv`, `fvm_extra`, `fvm_mb`, `fvm_r3`). It needs a prebuilt kernel for your GPU architecture (downloaded from <https://huggingface.co/fastforks/kernels> on first use) and a recent NVIDIA driver; if there is none it raises `BackendUnavailableError` with the reason and never falls back silently. `mode="auto"` picks `kernels` when it can run and logs the choice.
|
| 98 |
|
| 99 |
-
<!-- staged from fastforks
|
|
|
|
| 48 |
|
| 49 |
| model | time (ms): original fp32 | original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
|
| 50 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
| 51 |
+
| `efficientvit_b2.r288_in1k` | 10.7 | 14.7 | 11.9 | 1.10 | 9.7x | 10.8x | 173 / 194 | 85 / 124 | 120 / 212 |
|
| 52 |
|
| 53 |
### Startup and first call
|
| 54 |
|
|
|
|
| 60 |
|
| 61 |
| model | input | portable ms | compile ms | compile s (cold) | TensorRT fp16 ms | ONNX + build s | kernels ms | peak MiB (alloc / reserved): portable | compile | TensorRT (device) |
|
| 62 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
| 63 |
+
| `efficientvit_b2.r288_in1k` | 224x224 | 11.9 | 6.9 | 11.0 | 0.7 | 39 | 1.10 | 85 / 124 | 120 / 164 | 72 |
|
| 64 |
|
| 65 |
### Fixed-size speed and peak memory, all models
|
| 66 |
|
|
|
|
| 77 |
|
| 78 |
### Parity with the original code
|
| 79 |
|
| 80 |
+
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.05 and p99 <= 0.2 (absolute error of the output `logits`).
|
| 81 |
|
| 82 |
## Usage
|
| 83 |
|
|
|
|
| 96 |
|
| 97 |
`classification.load("efficientvit_b2.r288_in1k", mode="kernels")` runs our CUDA kernels (`cudnn_hacc`, `fvm_conv`, `fvm_extra`, `fvm_mb`, `fvm_r3`). It needs a prebuilt kernel for your GPU architecture (downloaded from <https://huggingface.co/fastforks/kernels> on first use) and a recent NVIDIA driver; if there is none it raises `BackendUnavailableError` with the reason and never falls back silently. `mode="auto"` picks `kernels` when it can run and logs the choice.
|
| 98 |
|
| 99 |
+
<!-- staged from fastforks badda9aa (2026-11-06) -->
|