---
license: apache-2.0
tags: [qhexrt, hexagon, npu, qnn, llm, gated-delta-net]
base_model: Qwen/Qwen3.5-2B
pipeline_tag: text-generation
---

# Qwen3.5-2B — Hexagon NPU (QHexRT) — v81

A prebuilt **QHexRT** bundle of **Qwen/Qwen3.5-2B** (the **Qwen3-Next "GatedDeltaNet" hybrid** lineage)
running on the **Qualcomm Hexagon v81 NPU** (SM8850). text→text.

The text decoder is a **hybrid**: 3 of every 4 layers are a **recurrent gated delta-rule linear attention**
(`Qwen3_5GatedDeltaNet`: a short causal-conv FIFO + an SSM state + gated RMSNorm), and every 4th layer is
**gated softmax attention** (q-proj output gate, per-head q/k RMSNorm, **partial-rotary** M-RoPE θ=1e7). The
runtime drives it with the `qwen3_5_generate` host-op (decode-over-prompt; conv + SSM + KV state carried
host-side), the lm-head as a separate graph.

## What's here (`v81/` — the flat artifacts root)
| file | role | ~size |
|---|---|---|
| `qwen3.5-2b-1024.json` | QHexRT manifest (the declarative run plan) | 2 KB |
| `qwen352b_decode_f16.bin` | decode context binary (24 layers, hid 2048, fp16) | 2.76 GB |
| `qwen352b_lmhead_f16.bin` | lm-head context binary (tied embed, fp16) | 1.0 GB |
| `qwen352b_embed_f16.bin` | embedding table (host lookup, fp16) | 1.0 GB |
| `tokenizer.json` | Qwen2-style BPE tokenizer | 20 MB |

Arch-pinned: a v81 binary will not load on another Hexagon arch. The QNN runtime libs come from the QAIRT SDK,
not this repo.

## Run
```bash
hf download runanywhere/qwen3_5_2b_HNPU --local-dir q35_2b
adb push q35_2b/v81 /data/local/tmp/wq/qwen35-2b
# also stage the QAIRT runtime libs + a qhx_generate built with the qwen3_5_generate host-op (see caveats)
adb shell "cd /data/local/tmp/wq && LD_LIBRARY_PATH=. \
  ./qhx_generate qwen35-2b/qwen3.5-2b-1024.json libQnnHtp.so libQnnSystem.so qwen35-2b 64 'The capital of France is'"
```

## Measured (device: SM8850 / v81, QAIRT 2.47)
- **Decode: ~11.0 tok/s** (fp16).
- **Parity: greedy-EXACT** — on-device tokens match the HF `Qwen/Qwen3.5-2B` greedy continuation **41/41**
  (prompt "The capital of France is").

## Caveats
- **fp16** decode + lm-head (strict greedy parity). W8 weight-only is the next perf lever (~2×, weight-bound).
- The fp16 decode weights (2.7 GB) exceed the 2 GB model-lib relocation limit, so the build uses a
  **large-code-model** path (weight blob → `.lrodata` + `-mcmodel=large`); this is baked into the published bin.
- **Text-LLM path only** — the vision tower + MTP head are not exported.
- Needs a `qhx_generate` that includes the **`qwen3_5_generate`** host-op (QHexRT branch `smonga/qwen_fam`).
- Built by the in-repo `forge` pipeline (oracle-gated export 10/10 → QAIRT-2.47 O3 compile → device greedy gate).