---
license: apache-2.0
tags:
  - qhexrt
  - internvl
  - vlm
  - image-text-to-text
  - hexagon
  - npu
  - qnn
base_model: OpenGVLab/InternVL3_5-1B
pipeline_tag: image-text-to-text
language:
  - en
---

# InternVL3.5-1B — QHexRT NPU bundle (Hexagon v79)

Precompiled **InternVL3.5-1B** vision-language model for the **QHexRT** runtime on Qualcomm
Hexagon **v79** (Snapdragon 8 Elite / SM8750, e.g. Galaxy S25). Image → caption runs **entirely on
the NPU**: the InternViT-300M vision encoder and the full Qwen3-0.6B language model are
`qnn_context_binary` graphs; the host does only light glue (image resize, patch-embed, pixel-shuffle
projector, tokenize/splice, sampling). Device-validated end-to-end on diverse photos.

The model is driven by QHexRT's declarative **manifest plan-interpreter** (`Session::generate` via the
`internvl_generate` host-op) — the same engine that runs the other `*_HNPU` bundles.

## Contents (`v79/`)
| file | what | size |
|------|------|------|
| `internvl3_5-1b.json` | QHexRT manifest (vlm family, `internvl_generate` plan) | — |
| `ivl_enc2.bin` | InternViT-300M vision encoder — 24-layer, **fp16**, 2D `[seq,dim]` I/O | 621 MB |
| `ivl_pf.bin` | Qwen3-0.6B **prefill** graph, fixed `S=310` (**fp16**) | 898 MB |
| `ivl_dec_w4.bin` | Qwen3-0.6B **decode** step (1 token), **W4A16** | 225 MB |
| `ivl_lmh_q.bin` | untied **lm-head**, **W8A16** (int8) → logits over 151936 vocab | 158 MB |
| `ivl_embed_f16.bin` | token embedding table `[151936,1024]` f16 (mmap, host lookup) | 311 MB |
| `tokenizer.json` | Qwen3 tokenizer (vocab 151936) | 11 MB |
| `vlm/` | host vision preproc: Conv2d patch-embed, pos-embed, layer-norm + pixel-shuffle MLP projector, prompt template | 27 MB |

Total ≈ **2.25 GB** on disk; ≈ **2.0 GB** peak device RSS at runtime (weights are ion-shared host↔DSP,
not duplicated).

## Run (QHexRT CLI)
```bash
hf download runanywhere/internvl3_5_1b_HNPU --local-dir internvl3_5_1b_HNPU
# qhx_generate comes from a QHexRT build; QNN libs from the QAIRT SDK (lib/aarch64-android) + v79 HTP skel.
adb push internvl3_5_1b_HNPU/v79 /data/local/tmp/wq/internvl
adb push my_photo.jpg /data/local/tmp/wq/internvl/

adb shell "cd /data/local/tmp/wq && LD_LIBRARY_PATH=. ADSP_LIBRARY_PATH=. \
  ./qhx_generate internvl/internvl3_5-1b.json libQnnHtp.so libQnnSystem.so internvl \
  60 'Describe this image in detail.' internvl/my_photo.jpg"
# -> text: "The image shows ..."
```
- **Image**: any aspect ratio / format (JPEG/PNG/…); center-cropped + resized to 448×448 host-side
  (256 image tokens / 1 tile).
- **Prompt is fixed**: the shipped prefill graph is pinned to `S=310`, which is the tokenized
  `"Describe this image in detail."` (+ 256 image tokens + chat template). Pass exactly that string
  (or `''` to use it as the default). Other prompts need a re-exported prefill at the new sequence length.

## Performance (measured, v79 / S25)
| stage | time |
|-------|------|
| vision encode (one-time) | **181 ms** |
| prefill / TTFT | **101 ms** |
| decode | **~28 ms/token** on the NPU graphs |

End-to-end wall time includes single-threaded host preproc (patch-embed + projector, ~1.5 s one-time);
the NPU compute itself is the table above.

## Notes
- Arch: **v79 only** — context binaries are dsp-arch-pinned. Other arches must be re-exported.
- No custom op-package needed — all graphs are pure-native HTP (HVX/HMX).
- **Precision**: vision fp16, prefill fp16, decode **W4A16**, lm-head **W8A16**. (Prefill/vision can be
  further quantized to shrink the footprint; this bundle ships the validated balance.)
- **HTP gotcha baked in**: all graph I/O is 2D `[seq,dim]` — HTP mis-lays-out 3D `[1,seq,dim]` tensors.
  The Conv2d patch-embed and pixel-shuffle are host-side for the same reason.
- Source model: [`OpenGVLab/InternVL3_5-1B`](https://huggingface.co/OpenGVLab/InternVL3_5-1B)
  (InternViT-300M + Qwen3-0.6B), exported and compiled with QAIRT 2.45 for
  `qualcomm-snapdragon-8-elite-for-galaxy`.