Sanitize public model card metadata

README.md

@@ -1,31 +1,18 @@
 ---
-license: other
-base_model: nvidia/Llama-3.1-Nemotron-Nano-VL-8B-V1
-tags: [vlm, image-to-text, hexagon, qualcomm, npu, qhexrt, radio, llama]
+license: "other"
+tags:
+- "hnpu"
+- "hexagon"
+- "npu"
+- "vlm"
+base_model: "nvidia/Llama-3.1-Nemotron-Nano-VL-8B-V1"
+pipeline_tag: "image-text-to-text"
 ---
 
-# Llama-3.1-Nemotron-Nano-VL-8B → Qualcomm Hexagon NPU (v81)
+# nemotron nano vl 8b HNPU
 
-NVIDIA **Llama-3.1-Nemotron-Nano-VL-8B-V1** (RADIO vision + Llama-3.1-8B) ported to the Qualcomm Hexagon NPU
-(SM8750 / **v81** / `soc_model 87`) and **device-validated on a Samsung S25**: image → detailed caption, fully
-on-device.
+Prebuilt HNPU artifacts for [nvidia/Llama-3.1-Nemotron-Nano-VL-8B-V1](https://huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-VL-8B-V1), a public vision-language model.
 
-## Device validation (S25, v81) — image→caption, 3/3 distinct images correct
-- **dog** → "…a white dog sits majestically, its gaze directed upwards… lush green grass… tongue…"
-- **tabby cat** → "…a tabby cat with a coat of brown and black stripes… eyes wide open, reflecting curiosity…"
-- **pizza** → "…a freshly baked pizza… golden crust, sits in a cardboard box…"
+For model behavior, license, intended use, and limitations, see the [upstream model card](https://huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-VL-8B-V1).
 
-The 8B LLM backbone is independently device-validated (W8 greedy 4/6 exact vs HF gold). Vision graph is export-
-faithful (ONNX vs PyTorch cos 1.000000); host preprocess+projector validated cos 1.000000.
-
-## Pipeline (all on-device via the `nemotron_vl_generate` host-op)
-image(512²) → RADIO input_conditioner + patch_generator (host) → **vision graph `radio_enc.bin`** (32-block
-ViT-Huge) → drop 8 prefix → pixel_shuffle + mlp1 projector (host) → **256 image tokens** → spliced at the 256
-`<image>` slots → **8-part W8 sharded decode** (Llama-3.1-8B, MAXCTX 512) → lmhead. ~3.7 tok/s decode.
-
-## Run (QHexRT)
-```
-adb shell "cd /data/local/tmp/wq && export ADSP_LIBRARY_PATH='/data/local/tmp/wq/dsp;/data/local/tmp/wq;/vendor/dsp/cdsp'; \
-  LD_LIBRARY_PATH=. ./qhx_generate v81/nemotron-vl-8b-vlm.json libQnnHtp.so libQnnSystem.so v81 60 'Describe this image in detail.' my_photo.jpg"
-```
-Arch-pinned v81 (`dsp_arch`+`soc_model 87`). Part of the RunAnywhere Hexagon model set.
+Artifacts are architecture-pinned. Available artifact directories: `v81/`.