mudler commited on
Commit
f691d0b
·
verified ·
1 Parent(s): 7eaa709

Add/update model card for medium

Browse files
Files changed (1) hide show
  1. README.md +3 -3
README.md CHANGED
@@ -13,7 +13,7 @@ base_model: roboflow/rfdetr
13
 
14
  # RF-DETR Medium — GGUF for rfdetr.cpp
15
 
16
- GGUF-format weights of [Roboflow RF-DETR Medium](https://github.com/roboflow/rf-detr) (detection variant) for use with [rfdetr.cpp](https://github.com/mudler/rt-detr.cpp), a C++/ggml implementation that matches the upstream PyTorch model on CPU.
17
 
18
  This repo contains all four standard quantizations of this variant. **F16 is the recommended default** — same accuracy as F32, 1.85× smaller, and typically the fastest on modern CPUs thanks to ggml's F32×F16 matmul fast path.
19
 
@@ -48,7 +48,7 @@ All accuracy numbers are computed against the upstream PyTorch reference (`rfdet
48
 
49
  ```bash
50
  # 1. Clone + build rfdetr.cpp
51
- git clone https://github.com/mudler/rt-detr.cpp
52
  cd rt-detr.cpp
53
  cmake -B build -DRFDETR_BUILD_CLI=ON && cmake --build build -j
54
 
@@ -67,7 +67,7 @@ build/bin/rfdetr-cli detect \
67
 
68
  All accuracy metrics are computed against the upstream PyTorch reference (rfdetr 1.7.0) on 7 COCO val2017 images at threshold 0.5. Each detection match uses greedy Hungarian-style assignment by IoU (≥ 0.5 lenient, ≥ 0.95 strict) with class equality required.
69
 
70
- See [BENCHMARK.md](https://github.com/mudler/rt-detr.cpp/blob/main/BENCHMARK.md) and [`benchmarks/results/accuracy_sweep.json`](https://github.com/mudler/rt-detr.cpp/blob/main/benchmarks/results/accuracy_sweep.json) for the full sweep across all 32 (variant × quant) cells.
71
 
72
  ## License
73
 
 
13
 
14
  # RF-DETR Medium — GGUF for rfdetr.cpp
15
 
16
+ GGUF-format weights of [Roboflow RF-DETR Medium](https://github.com/roboflow/rf-detr) (detection variant) for use with [rfdetr.cpp](https://github.com/mudler/rf-detr.cpp), a C++/ggml implementation that matches the upstream PyTorch model on CPU.
17
 
18
  This repo contains all four standard quantizations of this variant. **F16 is the recommended default** — same accuracy as F32, 1.85× smaller, and typically the fastest on modern CPUs thanks to ggml's F32×F16 matmul fast path.
19
 
 
48
 
49
  ```bash
50
  # 1. Clone + build rfdetr.cpp
51
+ git clone https://github.com/mudler/rf-detr.cpp
52
  cd rt-detr.cpp
53
  cmake -B build -DRFDETR_BUILD_CLI=ON && cmake --build build -j
54
 
 
67
 
68
  All accuracy metrics are computed against the upstream PyTorch reference (rfdetr 1.7.0) on 7 COCO val2017 images at threshold 0.5. Each detection match uses greedy Hungarian-style assignment by IoU (≥ 0.5 lenient, ≥ 0.95 strict) with class equality required.
69
 
70
+ See [BENCHMARK.md](https://github.com/mudler/rf-detr.cpp/blob/main/BENCHMARK.md) and [`benchmarks/results/accuracy_sweep.json`](https://github.com/mudler/rf-detr.cpp/blob/main/benchmarks/results/accuracy_sweep.json) for the full sweep across all 32 (variant × quant) cells.
71
 
72
  ## License
73