Update model card with GitHub link and inference example
Browse files
README.md
CHANGED
|
@@ -14,6 +14,14 @@ This model card accompanies the released foundation checkpoint for:
|
|
| 14 |
|
| 15 |
**MultiPathFormer: A Foundation Model for Multipath Wireless Propagation**
|
| 16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
## Model
|
| 18 |
|
| 19 |
- Architecture: first-step residual corridor-concat MultiPathFormer.
|
|
@@ -32,6 +40,56 @@ The checkpoint was trained on 27 DeepMIMO ray-tracing scenarios listed in `confi
|
|
| 32 |
|
| 33 |
The model is intended for research on geometry-aware wireless propagation modeling and downstream task reuse. It should be used with the preprocessing artifacts released alongside the checkpoint.
|
| 34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
## Limitations
|
| 36 |
|
| 37 |
The checkpoint is trained on simulated ray-tracing data. It has not yet been validated as a standalone replacement for real-world measurement campaigns. Phase-accurate channel synthesis is especially sensitive and should be evaluated carefully for each use case.
|
|
|
|
| 14 |
|
| 15 |
**MultiPathFormer: A Foundation Model for Multipath Wireless Propagation**
|
| 16 |
|
| 17 |
+
Code release:
|
| 18 |
+
|
| 19 |
+
https://github.com/gblessed/MultiPathFormer
|
| 20 |
+
|
| 21 |
+
Hugging Face model repo:
|
| 22 |
+
|
| 23 |
+
https://huggingface.co/gblessed/multipathformer
|
| 24 |
+
|
| 25 |
## Model
|
| 26 |
|
| 27 |
- Architecture: first-step residual corridor-concat MultiPathFormer.
|
|
|
|
| 40 |
|
| 41 |
The model is intended for research on geometry-aware wireless propagation modeling and downstream task reuse. It should be used with the preprocessing artifacts released alongside the checkpoint.
|
| 42 |
|
| 43 |
+
## Quick Inference Example
|
| 44 |
+
|
| 45 |
+
Install the code release and dependencies:
|
| 46 |
+
|
| 47 |
+
```bash
|
| 48 |
+
git clone https://github.com/gblessed/MultiPathFormer.git
|
| 49 |
+
cd MultiPathFormer
|
| 50 |
+
python -m pip install -r requirements.txt
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
Run inference from an already constructed augmented prompt and first-step baseline:
|
| 54 |
+
|
| 55 |
+
```python
|
| 56 |
+
import json
|
| 57 |
+
from pathlib import Path
|
| 58 |
+
|
| 59 |
+
import numpy as np
|
| 60 |
+
from huggingface_hub import hf_hub_download
|
| 61 |
+
|
| 62 |
+
from multipathformer.inference import MultiPathFormerPredictor, prediction_to_rows
|
| 63 |
+
|
| 64 |
+
repo_id = "gblessed/multipathformer"
|
| 65 |
+
checkpoint = hf_hub_download(
|
| 66 |
+
repo_id=repo_id,
|
| 67 |
+
filename="first_step_residual_corridor_concat_27scenarios_44710a4a_best_model_checkpoint.pth",
|
| 68 |
+
)
|
| 69 |
+
artifacts = hf_hub_download(repo_id=repo_id, filename="preprocessing_artifacts.json")
|
| 70 |
+
|
| 71 |
+
augmented_prompt = np.asarray(
|
| 72 |
+
json.loads(Path("examples/augmented_prompt.json").read_text()),
|
| 73 |
+
dtype=np.float32,
|
| 74 |
+
)
|
| 75 |
+
first_step_baseline = np.asarray(
|
| 76 |
+
json.loads(Path("examples/first_step_baseline.json").read_text()),
|
| 77 |
+
dtype=np.float32,
|
| 78 |
+
)
|
| 79 |
+
|
| 80 |
+
predictor = MultiPathFormerPredictor(checkpoint, artifacts_path=artifacts)
|
| 81 |
+
prediction = predictor.predict_from_augmented_prompt(
|
| 82 |
+
augmented_prompt,
|
| 83 |
+
first_step_baseline,
|
| 84 |
+
max_steps=25,
|
| 85 |
+
)
|
| 86 |
+
|
| 87 |
+
for row in prediction_to_rows(prediction)[:5]:
|
| 88 |
+
print(row)
|
| 89 |
+
```
|
| 90 |
+
|
| 91 |
+
The augmented prompt follows the released corridor-concat recipe: TX/RX position, first-step delay/power baseline, first-step standard deviation, and scene/corridor descriptors. The helper script `inference_example.py` in this model repo contains the same flow.
|
| 92 |
+
|
| 93 |
## Limitations
|
| 94 |
|
| 95 |
The checkpoint is trained on simulated ray-tracing data. It has not yet been validated as a standalone replacement for real-world measurement campaigns. Phase-accurate channel synthesis is especially sensitive and should be evaluated carefully for each use case.
|