Prune QHexRT HNPU bundle to runtime-minimum artifacts

Remove 6 non-runtime/sensitive files from the QHexRT HNPU bundle.
Update model card arch metadata to match remaining artifact directories.
Strip non-runtime _comment metadata from 2 manifest(s) without changing runtime fields.
Audit artifacts: QHexRT/hf_hnpu_audit cleanup plan, manifest reference check, and multi-agent review.

README.md

@@ -5,7 +5,6 @@ tags:
 - "hexagon"
 - "npu"
 - "asr"
-- "v75"
 - "v79"
 - "v81"
 base_model: "UsefulSensors/moonshine-tiny"
@@ -18,4 +17,4 @@ Prebuilt HNPU artifacts for [UsefulSensors/moonshine-tiny](https://huggingface.c
 
 For model behavior, license, intended use, and limitations, see the [upstream model card](https://huggingface.co/UsefulSensors/moonshine-tiny).
 
-Artifacts are architecture-pinned. Available artifact directories: `v75/`, `v79/`, `v81/`.
+Artifacts are architecture-pinned. Available artifact directories: `v79/`, `v81/`.

config.json

@@ -1,7 +0,0 @@
-{
-  "model_type": "qhexrt",
-  "name": "moonshine_tiny_HNPU",
-  "framework": "qhexrt",
-  "runtime": "qnn-htp",
-  "_note": "Root config.json present so the Hugging Face Hub counts downloads (default query file for repos without a registered library)."
-}

v75/moonshine-tiny.json

@@ -1,47 +0,0 @@
-{
-  "schema_version": 1,
-  "_comment": "Moonshine-tiny ASR (encoder-decoder, Useful Sensors). Raw-waveform conv stem -> encoder (RoPE) -> autoregressive decoder (RoPE self-attn + cross-attn, gated SwiGLU, tied lm-head). Gated on WER vs HF.",
-  "model": {
-    "name": "moonshine-tiny",
-    "family": "asr",
-    "dsp_arch": "v75",
-    "tokenizer_pre": "gemma"
-  },
-  "params": {
-    "hidden": 288,
-    "vocab": 32768,
-    "n_layers": 6,
-    "max_ctx": 200,
-    "kv_dim": 288,
-    "head_dim": 36,
-    "rope_theta": 10000.0,
-    "eos_token_id": 2
-  },
-  "artifacts": {
-    "contexts": {
-      "encoder": {
-        "bin": "moonshinetiny_enc_f16.bin"
-      },
-      "decoder": {
-        "bin": "moonshinetiny_dec_f16.bin"
-      }
-    },
-    "tokenizer": "tokenizer.json"
-  },
-  "plan": {
-    "steps": [
-      {
-        "host": "moonshine_transcribe",
-        "params": {
-          "encoder": "moonshinetiny_enc_f16",
-          "decoder": "moonshinetiny_dec_f16",
-          "sot": 1,
-          "conv_weights": "moonshine_conv_stem.bin",
-          "frontend": "raw",
-          "max_audio_samples": 160000,
-          "max_decode": 180
-        }
-      }
-    ]
-  }
-}

v75/moonshine_conv_stem.bin

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:cc0d6966eb3022193620527682b44b34035090c4a440bf5caafd8a253f50daea
-size 6787584

v75/moonshinetiny_dec_f16.bin

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:826f1c1dd47ade34f00cde6a423890356abeece76dde5f33efcec0a2ddc5826d
-size 56393728

v75/moonshinetiny_enc_f16.bin

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:0c261541ff9f7ac0d0b2b18ffc3a763019ee09b8f0ce8347adef5c8e541e4082
-size 17219584

v79/moonshine-tiny.json

@@ -1,6 +1,5 @@
 {
   "schema_version": 1,
-  "_comment": "Moonshine-tiny ASR (encoder-decoder, Useful Sensors). Raw-waveform conv stem -> encoder (RoPE) -> autoregressive decoder (RoPE self-attn + cross-attn, gated SwiGLU, tied lm-head). Gated on WER vs HF.",
   "model": {
     "name": "moonshine-tiny",
     "family": "asr",

v81/moonshine-tiny.json

@@ -1,6 +1,5 @@
 {
   "schema_version": 1,
-  "_comment": "moonshine-tiny - Qualcomm Hexagon v81 (QHexRT) deployment manifest.",
   "model": {
     "name": "moonshine-tiny",
     "family": "asr",