Upload folder using huggingface_hub

README.md

@@ -23,288 +23,268 @@ tags:
   - native-apple
 ---
 
-# GigaAM v3 MLX
+# GigaAM v3 · MLX Runtime Bundle
 
-**GigaAM v3 MLX** is a native Apple MLX runtime bundle for offline Russian automatic speech recognition on iPhone, iPad, and Mac.
+> Offline Russian speech recognition for Apple devices.  
+> Native MLX inference on iPhone, iPad, and Mac — no Python, no cloud, no server required at runtime.
 
-This repository contains converted model assets for running `ai-sage/GigaAM-v3` revision `e2e_rnnt` with a native Apple runtime.
+Converted from [`ai-sage/GigaAM-v3`](https://huggingface.co/ai-sage/GigaAM-v3) (`e2e_rnnt` revision).  
+No additional training or fine-tuning. The native runtime produced identical decoded text to the Python reference on the validated test inputs.
 
-The bundle is intended for native on-device inference on Apple platforms without Python, PyTorch, Transformers, torchaudio, librosa, pyannote, or server-side inference at runtime.
+---
 
-## Model
+## Platforms
 
-* Base model: `ai-sage/GigaAM-v3`
-* Revision: `e2e_rnnt`
-* Architecture: Conformer RNN-T
-* Language: Russian
-* Runtime target: native Apple MLX
-* Precision: FP16
-* Sample rate: 16 kHz
-* Audio channels: mono
-* Tokenizer: SentencePiece
-* Vocabulary size: 1024
-* Blank ID: 1024
-* Output classes: 1025
+| Platform | Target | Runtime |
+|----------|--------|---------|
+| iOS | iPhone | Apple MLX (native) |
+| iPadOS | iPad | Apple MLX (native) |
+| macOS | Apple Silicon Mac | Apple MLX (native) |
 
-## Target platforms
+---
 
-This model bundle is intended for native Apple applications:
+## Performance · Apple M1 Max
 
-| Platform | Target            |
-| -------- | ----------------- |
-| iOS      | iPhone            |
-| iPadOS   | iPad              |
-| macOS    | Apple Silicon Mac |
+### Short audio (~6 s)
 
-## Intended use
+| Runtime | Total time | RTF | Speed |
+|---------|-----------|-----|-------|
+| **Native Apple MLX** | ~0.168 s | ~0.028 | **~35.8× realtime** |
+| Python reference | ~0.701 s | ~0.117 | ~8.6× realtime |
 
-This bundle is intended for offline speech recognition in native Apple applications:
+The native runtime is approximately **4.2× faster** than the Python reference on the same hardware.
 
-* iPhone apps
-* iPad apps
-* macOS apps
-* local transcription tools
-* privacy-preserving offline ASR
-* Russian speech-to-text without cloud inference
-* native Swift/MLX ASR runtimes
+### Long-form (911 s · 15 min 11 s)
 
-This repository contains model assets only. It is not a Python inference package.
+| Metric | Value |
+|--------|-------|
+| Chunk size | 20.0 s |
+| Chunk count | 46 |
+| Total transcription time | 24.2 s |
+| Real-time factor | 0.027 |
+| **Speed** | **~37.6× realtime** |
+| Peak resident memory | ~1.15 GB |
 
-## Repository files
+<details>
+<summary>Stage breakdown (long-form)</summary>
 
-```text
-README.md
-.gitattributes
-manifest.json
-checksums.sha256
-weights.fp16.safetensors
-tokenizer.model
-hann_window.f32.bin
-mel_filterbank_mel_freq.f32.bin
-```
+| Stage | Time | Share |
+|-------|------|-------|
+| Audio load | 0.019 s | ~0.1% |
+| Mel frontend | 5.993 s | ~24.8% |
+| Encoder | 5.463 s | ~22.6% |
+| RNNT greedy decoding | 12.736 s | ~52.6% |
+| — RNNT decoder | 1.821 s | ~7.5% |
+| — RNNT joint | 10.680 s | ~44.1% |
+| — RNNT readback | 0.169 s | ~0.7% |
+| Tokenizer | 0.003 s | ~0.0% |
 
-| File                              | Description                                                  |
-| --------------------------------- | ------------------------------------------------------------ |
-| `weights.fp16.safetensors`        | FP16 converted model weights                                 |
-| `tokenizer.model`                 | SentencePiece tokenizer                                      |
-| `manifest.json`                   | Runtime manifest with model, frontend, and decoding metadata |
-| `hann_window.f32.bin`             | Hann window used by the frontend                             |
-| `mel_filterbank_mel_freq.f32.bin` | Mel filterbank used by the frontend                          |
-| `checksums.sha256`                | SHA-256 checksums for integrity checks                       |
-| `.gitattributes`                  | Git LFS rules for binary model assets                        |
+The current main bottleneck is the **RNNT joint network** during greedy decoding.
+</details>
 
-## Runtime pipeline
+### Memory
 
-The intended native runtime pipeline is:
+| Scenario | Peak RSS |
+|----------|----------|
+| Native MLX · short audio | ~1.10 GB |
+| Native MLX · long-form audio | ~1.15 GB |
+| Python reference · short audio | ~1.76 GB |
 
-```text
-Audio file / PCM samples
-→ native audio loader
-→ 16 kHz mono Float32 PCM
-→ mel spectrogram frontend
-→ Conformer encoder
-→ RNNT greedy decoder
-→ SentencePiece tokenizer
-→ text
-```
+---
 
-The model bundle includes frontend assets so that native runtimes can reproduce the original preprocessing without relying on Python audio libraries.
-
-## Frontend configuration
-
-| Parameter          | Value |
-| ------------------ | ----: |
-| Sample rate        | 16000 |
-| Channels           |     1 |
-| Number of mel bins |    64 |
-| FFT size           |   320 |
-| Window length      |   320 |
-| Hop length         |   160 |
-| Center             | false |
-| Mel scale          |   HTK |
-| Mel normalization  |  none |
-| Power              |   2.0 |
-
-The effective feature hop is 10 ms before encoder subsampling. The encoder uses a subsampling factor of 4, so one encoder frame corresponds approximately to 40 ms of audio.
-
-## Architecture details
-
-| Component                | Value                 |
-| ------------------------ | --------------------- |
-| Encoder type             | Conformer             |
-| Number of encoder layers | 16                    |
-| Model dimension          | 768                   |
-| Attention heads          | 16                    |
-| Attention type           | Rotary self-attention |
-| Convolution kernel size  | 5                     |
-| Subsampling              | Conv1D                |
-| Subsampling factor       | 4                     |
-| Prediction network       | RNNT predictor        |
-| Joint network            | RNNT joint            |
-| Decoding                 | Greedy RNNT           |
-| Tokenizer                | SentencePiece         |
-| Vocabulary size          | 1024                  |
-| Blank ID                 | 1024                  |
-| Output classes           | 1025                  |
+## Runtime Pipeline
 
-## Validation
+```
+Audio file / PCM samples
+  → native audio loader
+  → 16 kHz mono Float32 PCM
+  → mel spectrogram frontend
+  → Conformer encoder
+  → RNNT greedy decoder
+  → SentencePiece tokenizer
+  → text
+```
 
-The conversion was validated against the original PyTorch/Hugging Face model using tensor-level golden references.
+All preprocessing assets (`hann_window`, `mel_filterbank`) are bundled so native runtimes
+can reproduce the original pipeline exactly without any Python audio libraries.
 
-Validated components include:
+---
 
-* audio frontend
-* mel spectrogram
-* pre-encoder
-* Conformer feed-forward blocks
-* rotary self-attention
-* Conformer convolution block
-* full Conformer layer
-* encoder stack
-* RNNT predictor
-* RNNT joint network
-* RNNT greedy decoding
-* SentencePiece tokenizer
-* full WAV-to-text pipeline
+## Requirements
 
-### Selected validation results
+This bundle is designed for **native Apple platform development**.  
+It is **not a Python package** — there is nothing to `pip install`.
 
-#### Mel frontend parity
+### Native runtime (target)
 
-```text
-feature_shape: [64, 99]
-max_abs_diff: 0.0004234314
-mean_abs_diff: 2.8040542e-05
-```
+| Requirement | Value |
+|-------------|-------|
+| Platform | iOS, iPadOS, and macOS — any version supported by MLX Swift |
+| Architecture | arm64 Apple devices  |
+| Framework | [Apple MLX](https://github.com/ml-explore/mlx) |
+| Language | Swift |
+| Runtime deps | MLX Swift runtime and bundled model assets |
 
-#### Encoder stack parity
+To use this bundle in a native app, add the **MLX Swift** package to your Xcode project:
 
-```text
-stack_max_abs_diff: 2.5629997e-06
-stack_mean_abs_diff: 3.8420205e-07
 ```
-
-#### Full encoder parity
-
-```text
-output_shape: [1, 768, 25]
-max_abs_diff: 2.682209e-06
-mean_abs_diff: 4.0401252e-07
+https://github.com/ml-explore/mlx-swift
 ```
 
-#### End-to-end smoke test
-
-A short Russian WAV sample was used to verify end-to-end decoding against the Python reference implementation.
-
-The native runtime and the Python reference produced identical text for the same input audio.
-
-The audio fixture is not included in this model repository. It is used only for runtime validation.
-
-## Performance
-
-Benchmarks were measured on Apple M1 Max with a native Apple MLX runtime in release mode.
+This repository provides model assets. A native runtime should load `manifest.json`, `weights.fp16.safetensors`, tokenizer files, and bundled frontend assets, then execute the pipeline described below.
 
-### Short audio benchmark
+---
 
-| Runtime                  | Audio duration | Total time |    RTF |           Speed |
-| ------------------------ | -------------: | ---------: | -----: | --------------: |
-| Native Apple MLX runtime |           ~6 s |   ~0.168 s | ~0.028 | ~35.8× realtime |
-| Python reference         |           ~6 s |   ~0.701 s | ~0.117 |  ~8.6× realtime |
+## Repository Files
 
-The native runtime was approximately 4.2× faster than the Python reference in this short-audio warm benchmark.
+```
+README.md
+.gitattributes
+manifest.json
+checksums.sha256
+weights.fp16.safetensors
+tokenizer.model
+tokenizer_vocab.json
+hann_window.f32.bin
+mel_filterbank_mel_freq.f32.bin
+```
 
-### Long-form benchmark
+| File | Description |
+|------|-------------|
+| `weights.fp16.safetensors` | FP16 model weights (MLX-compatible) |
+| `tokenizer.model` | SentencePiece tokenizer model |
+| `tokenizer_vocab.json` | Vocabulary export for native tokenizer implementations |
+| `manifest.json` | Runtime manifest — model, frontend, and decoding metadata |
+| `hann_window.f32.bin` | Hann window for mel frontend |
+| `mel_filterbank_mel_freq.f32.bin` | Mel filterbank for mel frontend |
+| `checksums.sha256` | SHA-256 checksums for integrity verification |
 
-Long-form audio is processed in chunks to keep memory usage predictable and enable efficient transcription on Apple devices.
+---
 
-| Metric                   |           Value |
-| ------------------------ | --------------: |
-| Audio duration           |       911.252 s |
-| Audio duration           |     15 min 11 s |
-| Chunk size               |          20.0 s |
-| Chunk count              |              46 |
-| Total transcription time |       24.2145 s |
-| Real-time factor         |         0.02657 |
-| Speed                    | ~37.6× realtime |
-| Peak resident memory     |        ~1.15 GB |
+## Architecture
+
+| Component | Value |
+|-----------|-------|
+| Encoder type | Conformer |
+| Encoder layers | 16 |
+| Model dimension | 768 |
+| Attention heads | 16 |
+| Attention type | Rotary self-attention |
+| Convolution kernel size | 5 |
+| Subsampling | Conv1D · factor 4 |
+| Prediction network | RNNT predictor |
+| Joint network | RNNT joint |
+| Decoding | Greedy RNNT |
+| Tokenizer | SentencePiece |
+| Vocabulary size | 1024 |
+| Blank ID | 1024 |
+| Output classes | 1025 |
+| Precision | FP16 |
 
-### Long-form stage breakdown
+---
 
-| Stage                |     Time |  Share |
-| -------------------- | -------: | -----: |
-| Audio load           |  0.019 s |  ~0.1% |
-| Mel frontend         |  5.993 s | ~24.8% |
-| Model total          | 18.203 s | ~75.2% |
-| Encoder              |  5.463 s | ~22.6% |
-| RNNT greedy decoding | 12.736 s | ~52.6% |
-| RNNT decoder         |  1.821 s |  ~7.5% |
-| RNNT joint           | 10.680 s | ~44.1% |
-| RNNT readback        |  0.169 s |  ~0.7% |
-| Tokenizer            |  0.003 s |  ~0.0% |
+## Frontend Configuration
 
-The current main runtime bottleneck is the RNNT joint network during greedy decoding.
+| Parameter | Value |
+|-----------|-------|
+| Sample rate | 16 000 Hz |
+| Channels | 1 (mono) |
+| Mel bins | 64 |
+| FFT size | 320 |
+| Window length | 320 |
+| Hop length | 160 |
+| Center | false |
+| Mel scale | HTK |
+| Mel normalization | none |
+| Power | 2.0 |
 
-### Memory
+Effective feature hop: **10 ms** before encoder subsampling.  
+Encoder subsampling factor: **4** → one encoder frame ≈ **40 ms** of audio.
 
-| Scenario                                  | Peak RSS |
-| ----------------------------------------- | -------: |
-| Native Apple MLX runtime, short audio     | ~1.10 GB |
-| Native Apple MLX runtime, long-form audio | ~1.15 GB |
-| Python reference, short audio             | ~1.76 GB |
+---
 
-## Long-form transcription
+## Long-Form Transcription
 
-Long-form audio is intended to be processed in chunks.
+Long audio is intended to be processed in chunks to keep memory usage bounded and inference latency predictable.
 
-Recommended initial long-form settings:
+**Recommended settings:**
 
-```text
+```yaml
 chunk_seconds: 20
-overlap_seconds: 0-2
+overlap_seconds: 0–2
 sample_rate: 16000
 channels: mono
 ```
 
-Future runtimes may use VAD, overlap merging, and timestamp-aware segmentation for improved long-form quality.
+> Future runtimes may add VAD segmentation, overlap-aware merging, and timestamp-aware chunking for improved accuracy on long-form content.
 
-## Limitations
+---
 
-* This bundle is optimized for native Apple MLX runtimes.
-* Long audio should be processed in chunks.
-* Current validation focuses on numerical parity and runtime behavior.
-* Word-level timestamps are not included in the model bundle itself.
-* Diarization is not included.
-* This repository contains model assets only, not application code or SDK source code.
+## Validation
 
-## Relation to the original model
+The conversion was validated against the original PyTorch/Hugging Face model using tensor-level golden references at each stage of the pipeline.
 
-This bundle is a native Apple MLX runtime conversion of:
+**Validated components:**
 
-```text
-ai-sage/GigaAM-v3
-revision: e2e_rnnt
+- Audio frontend · mel spectrogram
+- Pre-encoder
+- Conformer feed-forward blocks
+- Rotary self-attention
+- Conformer convolution block
+- Full Conformer layer · encoder stack
+- RNNT predictor · RNNT joint network
+- RNNT greedy decoding
+- SentencePiece tokenizer
+- Full WAV-to-text pipeline (end-to-end)
+
+**Selected numerical results:**
+
+```
+Mel frontend
+  feature_shape      [64, 99]
+  max_abs_diff       0.0004234314
+  mean_abs_diff      2.8040542e-05
+
+Encoder stack
+  stack_max_abs_diff 2.5629997e-06
+  stack_mean_abs_diff 3.8420205e-07
+
+Full encoder
+  output_shape       [1, 768, 25]
+  max_abs_diff       2.682209e-06
+  mean_abs_diff      4.0401252e-07
 ```
 
-No additional training or fine-tuning was performed.
+End-to-end: the native runtime and the Python reference produce **identical decoded text** for the same input audio.
 
-The conversion preserves the original Conformer RNN-T architecture, SentencePiece tokenizer layout, and preprocessing configuration, while packaging the model assets for native offline inference on iOS, iPadOS, and macOS.
+---
 
-## License
+## Limitations
 
-This model bundle follows the license terms of the original `ai-sage/GigaAM-v3` model.
+- Optimized for native Apple MLX runtimes; not intended for server or Python-based inference.
+- Long audio should be processed in chunks by the host runtime.
+- Word-level timestamps are not included in the bundle.
+- Speaker diarization is not supported.
+- This repository contains **model assets only** — no application code, no Swift SDK source.
 
-License: MIT
+---
 
-## Attribution
+## License
 
-If you use this bundle, please also reference the original GigaAM model:
+MIT — follows the license of the original [`ai-sage/GigaAM-v3`](https://huggingface.co/ai-sage/GigaAM-v3) model.
 
-```text
-ai-sage/GigaAM-v3
-```
+---
 
-## Summary
+## Citation & Attribution
 
-GigaAM v3 MLX provides a native Apple MLX model bundle for offline Russian ASR.
+If you use this bundle, please also cite the original GigaAM model:
 
-It is intended for local, private, on-device speech recognition on Apple platforms without requiring Python or server-side inference at runtime.
+```bibtex
+@misc{gigaam-v3,
+  author       = {ai-sage},
+  title        = {GigaAM-v3},
+  year         = {2024},
+  publisher    = {Hugging Face},
+  howpublished = {\url{https://huggingface.co/ai-sage/GigaAM-v3}}
+}
+```

checksums.sha256

@@ -1,6 +1,7 @@
-f2791bfaa8c2fdb030797cbcdf9129d15d0ddef338c0155fd6702f4f79d86db5  README.md
-d78fae2160b53303e6200a6b7c9ae3df574ab083b243b80e7b5c8bee37a51d6b  manifest.json
+1d42658ae5dcc0424dade66b07b9b067d329084d53a8e549b830feae8b58c54f  README.md
+993c538b6dceffb7b018c633a7cc8e4034e7ca975b5c27ecc8ecd679fa574d16  manifest.json
 2ff6d955a61003e0d3cb2155be36c14aa3df69c879713f30880ddd07d71fde21  weights.fp16.safetensors
 828c12c991019eef952a960661f25a92d6ad279591e2ea466b4aeddf1d20a18a  tokenizer.model
+84e15355333a8eb460b2f899947e9976b8c036f4dd086002e43b987ddf4cc606  tokenizer_vocab.json
 834a6744851a775dbb1831e9e17f0f702599b4d52c1d97d7638b3f170824419f  hann_window.f32.bin
 e9425a25940359fb85efeaf6651f81679971a1c46bec1af3aba943ec8a5fd793  mel_filterbank_mel_freq.f32.bin

manifest.json

@@ -13,6 +13,7 @@
   "total_params": 222519937,
   "weights_file": "weights.fp16.safetensors",
   "tokenizer_file": "tokenizer.model",
+  "tokenizer_vocab_file": "tokenizer_vocab.json",
   "blank_id": 1024,
   "vocab_size": 1024,
   "num_classes": 1025,

tokenizer_vocab.json

@@ -0,0 +1,1035 @@
+{
+  "source": "Models/GigaAMKitPro/tokenizer.model",
+  "type": "sentencepiece_id_to_piece",
+  "size": 1024,
+  "unk_id": 0,
+  "bos_id": -1,
+  "eos_id": -1,
+  "pad_id": -1,
+  "pieces": [
+    "<unk>",
+    ".",
+    ",",
+    "▁",
+    "е",
+    "а",
+    "с",
+    "▁с",
+    "и",
+    "▁в",
+    "н",
+    "у",
+    "т",
+    "м",
+    "▁на",
+    "?",
+    "▁по",
+    "й",
+    "к",
+    "д",
+    "ли",
+    "л",
+    "р",
+    "я",
+    "▁и",
+    "-",
+    "в",
+    "▁не",
+    "ка",
+    "▁у",
+    "ла",
+    "ра",
+    "на",
+    "ль",
+    "з",
+    "ю",
+    "х",
+    "то",
+    "но",
+    "о",
+    "ж",
+    "ни",
+    "ш",
+    "ть",
+    "▁за",
+    "п",
+    "ри",
+    "▁«",
+    "ро",
+    "ва",
+    "ло",
+    "ле",
+    "ч",
+    "▁А",
+    "го",
+    "та",
+    "▁С",
+    "ре",
+    "г",
+    "▁что",
+    "да",
+    "те",
+    "ко",
+    "▁И",
+    "ки",
+    "▁о",
+    "ы",
+    "ти",
+    "▁я",
+    "б",
+    "▁к",
+    "ст",
+    "не",
+    "во",
+    "▁это",
+    "▁В",
+    "▁Включи",
+    "▁от",
+    "▁про",
+    "e",
+    "▁вы",
+    "ку",
+    "ма",
+    "ру",
+    "ми",
+    "a",
+    "▁По",
+    "▁—",
+    "че",
+    "▁при",
+    "▁как",
+    "ем",
+    "ь",
+    "ом",
+    "ов",
+    "ет",
+    "ди",
+    "▁О",
+    "».",
+    "o",
+    "▁мне",
+    "▁У",
+    "▁со",
+    "сь",
+    "▁Я",
+    "ё",
+    "э",
+    "▁а",
+    "ви",
+    "ля",
+    "чи",
+    "▁до",
+    "га",
+    "по",
+    "ме",
+    "му",
+    "i",
+    "ся",
+    "▁К",
+    "жи",
+    "ста",
+    "▁есть",
+    "▁так",
+    "ить",
+    "ой",
+    "▁да",
+    "до",
+    "▁Ну",
+    "ту",
+    "▁об",
+    "▁мо",
+    "ше",
+    "s",
+    "ча",
+    "t",
+    "ер",
+    "мо",
+    "▁ко",
+    "ей",
+    "▁вот",
+    "▁то",
+    "де",
+    "па",
+    "ан",
+    "▁б",
+    "ну",
+    "бо",
+    "▁На",
+    "же",
+    "ты",
+    "сти",
+    "ве",
+    "▁П",
+    "▁Да",
+    "▁Б",
+    "▁Т",
+    "вер",
+    "ны",
+    "ный",
+    "тор",
+    "сто",
+    "▁раз",
+    "▁э",
+    "ной",
+    "▁из",
+    "▁Сбер",
+    "▁под",
+    "0",
+    "ные",
+    "ши",
+    "▁ка",
+    "▁Д",
+    "ца",
+    "▁тебя",
+    "▁п",
+    "ги",
+    "ения",
+    "ба",
+    "ф",
+    "▁бы",
+    "▁меня",
+    "ня",
+    "ду",
+    "за",
+    "▁ли",
+    "▁он",
+    "▁ты",
+    "ке",
+    "▁всё",
+    "▁во",
+    "r",
+    "▁но",
+    "ник",
+    "им",
+    "об",
+    "би",
+    "ци",
+    "▁т",
+    "мен",
+    "вет",
+    "▁фильм",
+    "ую",
+    "▁пере",
+    "си",
+    "жа",
+    "u",
+    "ц",
+    "▁Салют",
+    "щи",
+    "▁Как",
+    "▁же",
+    "▁ре",
+    "ар",
+    "ев",
+    "▁мы",
+    "ная",
+    "▁там",
+    "ры",
+    "ать",
+    "▁М",
+    "ил",
+    "ние",
+    "▁Афина",
+    "▁его",
+    "▁д",
+    "d",
+    "ите",
+    "ал",
+    "ого",
+    "ят",
+    "▁будет",
+    "▁За",
+    "▁1",
+    "▁все",
+    "▁сезон",
+    "ин",
+    "▁Г",
+    "дел",
+    "▁ме",
+    "l",
+    "он",
+    "▁Джой",
+    "ск",
+    "y",
+    "сть",
+    "5",
+    "»",
+    "ор",
+    "прав",
+    "▁Не",
+    "ров",
+    "вы",
+    "m",
+    "▁де",
+    "9",
+    "▁Ко",
+    "▁те",
+    "ря",
+    "це",
+    "ёт",
+    "▁г",
+    "ая",
+    "▁То",
+    "▁Это",
+    "▁Но",
+    "лу",
+    "гра",
+    "▁Э",
+    "С",
+    "ый",
+    "8",
+    "из",
+    "рт",
+    "са",
+    "▁ТВ",
+    "со",
+    "лю",
+    "▁Вот",
+    "ков",
+    "6",
+    "▁ну",
+    "7",
+    "нов",
+    "пи",
+    "ще",
+    "▁Ма",
+    "▁Ф",
+    "▁для",
+    "ия",
+    "их",
+    "h",
+    "ных",
+    "▁ф",
+    "каз",
+    "▁Он",
+    "смотр",
+    "лы",
+    "▁Что",
+    ":",
+    "1",
+    "пу",
+    "лов",
+    "фи",
+    "тер",
+    "ок",
+    "ём",
+    "зи",
+    "▁2",
+    "▁S",
+    "▁ещё",
+    "ного",
+    "!",
+    "ное",
+    "хо",
+    "▁з",
+    "▁сейчас",
+    "ова",
+    "ит",
+    "ход",
+    "▁Л",
+    "пол",
+    "вать",
+    "Н",
+    "бе",
+    "▁До",
+    "▁вас",
+    "▁уже",
+    "ение",
+    "ется",
+    "луч",
+    "ство",
+    "▁ш",
+    "гу",
+    "зна",
+    "▁Поставь",
+    "ха",
+    "n",
+    "лен",
+    "щ",
+    "▁Ка",
+    "▁пред",
+    "ком",
+    "▁M",
+    "пис",
+    "чно",
+    "ами",
+    "чу",
+    "про",
+    "▁Е",
+    "c",
+    "g",
+    "2",
+    "ды",
+    "ович",
+    "▁вам",
+    "след",
+    "ним",
+    "дь",
+    "▁если",
+    "зы",
+    "ша",
+    "ным",
+    "ский",
+    "▁был",
+    "▁Угу",
+    "ешь",
+    "▁Так",
+    "ально",
+    "▁сериал",
+    "вой",
+    "нь",
+    "▁Покажи",
+    "k",
+    "p",
+    "▁нас",
+    "▁друг",
+    "▁Ты",
+    "▁кон",
+    "▁она",
+    "▁па",
+    "ции",
+    "▁Про",
+    "П",
+    "от",
+    "тра",
+    "»,",
+    "▁было",
+    "▁Ш",
+    "▁Во",
+    "О",
+    "род",
+    "▁ком",
+    "▁включи",
+    "▁канал",
+    "су",
+    "▁просто",
+    "▁З",
+    "▁Вы",
+    "ания",
+    "сы",
+    "▁Ро",
+    "▁Со",
+    "бы",
+    "▁Ра",
+    "▁хо",
+    "an",
+    "▁B",
+    "мер",
+    "4",
+    "В",
+    "Т",
+    "К",
+    "ень",
+    "▁C",
+    "аться",
+    "▁бо",
+    "in",
+    "▁когда",
+    "▁Бо",
+    "▁рас",
+    "▁они",
+    "er",
+    "мы",
+    "М",
+    "3",
+    "вод",
+    "▁...",
+    "▁пожалуйста",
+    "пер",
+    "ран",
+    "▁чтобы",
+    "b",
+    "▁эти",
+    "▁3",
+    "▁Мо",
+    "бу",
+    "▁Ку",
+    "▁может",
+    "плат",
+    "▁где",
+    "ную",
+    "▁очень",
+    "▁люб",
+    "кой",
+    "▁Мы",
+    "ская",
+    "дер",
+    "ключ",
+    "олог",
+    "рь",
+    "▁вид",
+    "тов",
+    "▁Та",
+    "ний",
+    "▁кино",
+    "▁Ев",
+    "▁ис",
+    "▁или",
+    "звон",
+    "▁ро",
+    "сси",
+    "ам",
+    "▁пре",
+    "▁ж",
+    "стро",
+    "▁дела",
+    "▁A",
+    "▁Ле",
+    "▁только",
+    "ён",
+    "▁можно",
+    "Д",
+    "▁Хочу",
+    "%",
+    "ном",
+    "▁её",
+    "v",
+    "▁Го",
+    "▁смотрешке",
+    "▁Ни",
+    "жу",
+    "▁нет",
+    "форм",
+    "лось",
+    "▁телефон",
+    "ое",
+    "▁сам",
+    "»?",
+    "▁могу",
+    "ают",
+    "зов",
+    "ar",
+    "▁P",
+    "▁Открой",
+    "ически",
+    "▁кран",
+    "овой",
+    "ах",
+    "▁При",
+    "R",
+    "▁Ю",
+    "нул",
+    "▁человек",
+    "on",
+    "▁ос",
+    "▁серия",
+    "▁бу",
+    "▁Найди",
+    "Б",
+    "стра",
+    "спе",
+    "дин",
+    "▁тре",
+    "▁пер",
+    "▁быть",
+    "гру",
+    "▁воз",
+    "Ч",
+    "▁Па",
+    "▁4",
+    "фе",
+    "▁D",
+    "▁Р",
+    "лё",
+    "дума",
+    "▁один",
+    "▁сказал",
+    "▁Медси",
+    "▁без",
+    "\"",
+    "▁чё",
+    "И",
+    "▁нужно",
+    "ление",
+    "ых",
+    "K",
+    "▁надо",
+    "▁Александр",
+    "ец",
+    "▁часть",
+    "ща",
+    "▁номер",
+    "▁Че",
+    "ского",
+    "▁F",
+    "ность",
+    "тив",
+    "ступ",
+    "вед",
+    "▁какие",
+    "ъ",
+    "енно",
+    "▁Ми",
+    "ской",
+    "вая",
+    "▁Сколько",
+    "▁10",
+    "▁пе",
+    "▁кто",
+    "▁Мне",
+    "▁год",
+    "ёр",
+    "▁такое",
+    "▁N",
+    "▁какой",
+    "▁ваш",
+    "ью",
+    "итель",
+    "▁банк",
+    "▁пу",
+    "I",
+    "▁вопрос",
+    "▁день",
+    "кры",
+    "ственно",
+    "ился",
+    "▁пят",
+    "йте",
+    "пре",
+    "f",
+    "ался",
+    "▁Де",
+    "▁три",
+    "▁этот",
+    "▁Нет",
+    "▁пи",
+    "▁два",
+    "▁этого",
+    "▁здесь",
+    "▁Ба",
+    "▁Фильм",
+    "ент",
+    "цен",
+    "▁тоже",
+    "▁20",
+    "▁дом",
+    "ман",
+    "нный",
+    "▁расс",
+    "▁Х",
+    "▁посмотреть",
+    "▁буду",
+    "ишь",
+    "инг",
+    "▁само",
+    "ности",
+    "▁тебе",
+    "▁Всё",
+    "▁сегодня",
+    "▁Ж",
+    "▁вообще",
+    "▁дев",
+    "▁Подключи",
+    "▁G",
+    "▁От",
+    "А",
+    "стве",
+    "иться",
+    "w",
+    "чь",
+    "▁Мар",
+    "▁T",
+    "ится",
+    "▁Ли",
+    "вяз",
+    "▁Сер",
+    "ских",
+    "ства",
+    "цио",
+    "▁Ре",
+    "▁Если",
+    "ское",
+    "▁оста",
+    "O",
+    "▁против",
+    "▁время",
+    "ются",
+    "бер",
+    "▁Можешь",
+    "E",
+    "Р",
+    "▁такой",
+    "▁сколько",
+    "▁Пере",
+    "нова",
+    "▁хочу",
+    "▁Давай",
+    "▁себя",
+    "цы",
+    "L",
+    "▁между",
+    "▁говорит",
+    "en",
+    "ген",
+    "le",
+    "▁Га",
+    "▁знаю",
+    "▁Владимир",
+    "st",
+    "чита",
+    "ация",
+    "▁которые",
+    "▁работа",
+    "нять",
+    "тельно",
+    "работ",
+    "▁час",
+    "▁Манчестер",
+    "▁боль",
+    "евич",
+    "▁даже",
+    "▁Хо",
+    "▁потому",
+    "вид",
+    "▁телеканал",
+    "▁эпизод",
+    "▁Отключи",
+    "▁мир",
+    "50",
+    "▁авто",
+    "▁Найти",
+    "шёл",
+    "▁Сбербанк",
+    "▁задачу",
+    "ic",
+    "▁свои",
+    "▁₽",
+    "▁счёт",
+    "▁рук",
+    "З",
+    "▁через",
+    "▁Открыть",
+    "▁одно",
+    "станов",
+    "▁деньги",
+    "▁Есть",
+    "▁отеле",
+    "альный",
+    "▁Андр",
+    "▁включить",
+    "▁тогда",
+    "▁ему",
+    "or",
+    "▁ничего",
+    "▁больше",
+    "▁Закажи",
+    "▁чем",
+    "▁Ла",
+    "ывает",
+    "▁который",
+    "H",
+    "▁передачу",
+    "▁хотел",
+    "▁жив",
+    "▁того",
+    "W",
+    "al",
+    "▁стран",
+    "жение",
+    "▁тут",
+    "луша",
+    "▁телевизор",
+    "▁HD",
+    "Л",
+    "пуск",
+    "▁Ха",
+    "V",
+    "▁Ру",
+    "▁YouTube",
+    "▁хорошо",
+    "ству",
+    "▁поставь",
+    "атель",
+    "il",
+    "▁какая",
+    "▁Кино",
+    "▁знаешь",
+    "ается",
+    "ll",
+    "ские",
+    "▁реш",
+    "ирован",
+    "▁потом",
+    "ению",
+    "ении",
+    "▁Все",
+    "образ",
+    "вести",
+    "▁матч",
+    "▁мал",
+    "▁стоит",
+    "▁гор",
+    "ходит",
+    "x",
+    "▁город",
+    "▁минут",
+    "▁много",
+    "ительно",
+    "▁конечно",
+    "▁можешь",
+    "▁интерес",
+    "▁Где",
+    "фон",
+    "▁пар",
+    "алась",
+    "Ц",
+    "▁две",
+    "▁себе",
+    "▁Включай",
+    "▁блядь",
+    "▁TV",
+    "▁Запусти",
+    "▁Алиса",
+    "▁ссылке",
+    "▁услуг",
+    "▁шторы",
+    "очка",
+    "30",
+    "▁Запись",
+    "▁Бел",
+    "▁кредит",
+    "▁общ",
+    "нибудь",
+    "▁почему",
+    "▁стар",
+    "▁Николаев",
+    "щение",
+    "z",
+    "▁действ",
+    "▁показать",
+    "У",
+    "▁добр",
+    "▁давай",
+    "▁Сейчас",
+    "▁понял",
+    "▁Когда",
+    "▁после",
+    "▁гол",
+    "брать",
+    "▁Виктор",
+    "езд",
+    "▁документ",
+    "▁Сити",
+    "▁экран",
+    "▁крас",
+    "▁подготовить",
+    "▁Закрыть",
+    "▁сделать",
+    "овым",
+    "▁сказать",
+    "▁валют",
+    "▁руб",
+    "▁Какие",
+    "▁стал",
+    "▁жизни",
+    "▁бес",
+    "▁смотрёшках",
+    "▁сторон",
+    "▁оператор",
+    "▁Ча",
+    "ировать",
+    "▁цел",
+    "▁штору",
+    "▁Сезон",
+    "▁начал",
+    "▁хорош",
+    "▁Сергее",
+    "▁прямо",
+    "говори",
+    "▁заяв",
+    "▁000",
+    "▁жанре",
+    "Г",
+    "40",
+    "▁Спасибо",
+    "▁чего",
+    "▁значит",
+    "десят",
+    "▁теперь",
+    "очки",
+    "сервис",
+    "▁проблем",
+    "▁знает",
+    "▁первый",
+    "олод",
+    "▁подключи",
+    "▁онлайн",
+    "▁люди",
+    "▁Выключи",
+    "▁отключи",
+    "▁готов",
+    "▁голос",
+    "▁найдётся",
+    "▁доллар",
+    "S",
+    "▁Хорошо",
+    "▁видео",
+    "▁статус",
+    "помощник",
+    "▁всегда",
+    "'",
+    "▁работы",
+    "▁баланс",
+    "▁более",
+    "ивает",
+    "▁семь",
+    "▁спасибо",
+    "очку",
+    "▁душ",
+    "▁Гарри",
+    "▁смотреть",
+    "▁Почему",
+    "ставить",
+    "нимает",
+    "▁стол",
+    "J",
+    "▁Юнайтед",
+    "▁Первый",
+    "U",
+    "▁футбол",
+    "▁купить",
+    "▁возможно",
+    "▁находится",
+    "▁машин",
+    "▁Поэтому",
+    "▁момент",
+    "▁жизнь",
+    "▁должны",
+    "▁лучше",
+    "▁узнать",
+    "▁людей",
+    "A",
+    "▁Алло",
+    "▁Здравствуйте",
+    "M",
+    "Э",
+    "гром",
+    "▁Алекс",
+    "▁новый",
+    "▁никак",
+    "▁страхование",
+    "▁посылку",
+    "▁именно",
+    "▁говорил",
+    "▁Валер",
+    "▁говорю",
+    "▁получается",
+    "▁Петр",
+    "Ф",
+    "▁Сериал",
+    "▁говоря",
+    "▁встреч",
+    "▁запись",
+    "▁которая",
+    "▁Диагноз",
+    "▁орган",
+    "▁АПЛ",
+    "▁глаза",
+    "▁команд",
+    "друг",
+    "C",
+    "▁идёт",
+    "▁необходим",
+    "▁Поищи",
+    "▁Закрой",
+    "▁понимаю",
+    "▁завтра",
+    "▁договор",
+    "▁Отзыв",
+    "▁вечер",
+    "▁программ",
+    "▁Может",
+    "надцать",
+    "▁женщин",
+    "T",
+    "Арсенал",
+    "▁наверное",
+    "▁Li",
+    "▁сразу",
+    "▁несколько",
+    "▁каталоге",
+    "▁свидания",
+    "▁глав",
+    "▁Васильев",
+    "▁Михайлов",
+    "▁четыре",
+    "▁Можно",
+    "▁четвёр",
+    "▁опять",
+    "▁открыт",
+    "▁Джо",
+    "▁должен",
+    "▁Сергей",
+    "Х",
+    "ложен",
+    "здоров",
+    "▁страхов",
+    "▁музыку",
+    "▁поэтому",
+    "▁новых",
+    "▁Челси",
+    "▁денег",
+    "▁добавить",
+    "▁заказать",
+    "▁сделал",
+    "▁будем",
+    "▁месяц",
+    "▁La",
+    "▁эфир",
+    "мобиль",
+    "▁случае",
+    "▁Beauty",
+    "▁государств",
+    "▁правильно",
+    "▁принцип",
+    "▁является",
+    "▁Смотреть",
+    "опыт",
+    "▁последний",
+    "▁соглас",
+    "▁Премьер",
+    "▁кабинет",
+    "D",
+    "▁стоимость",
+    "▁существ",
+    "▁Лайкни",
+    "▁никогда",
+    "▁торгов",
+    "▁Услугу",
+    "▁объект",
+    "▁Значит",
+    "▁Joy",
+    "слышал",
+    "▁опцию",
+    "Я",
+    "B",
+    "P",
+    "Ж",
+    "Е",
+    "&",
+    "Y",
+    "Z",
+    "X",
+    "+",
+    "Ш",
+    "/",
+    "Щ",
+    "F",
+    "N",
+    ";",
+    "G",
+    "—",
+    "j",
+    "Й",
+    "q",
+    "Q",
+    "°",
+    "Ё",
+    "Ю",
+    "Ы",
+    "₽",
+    "€",
+    "$",
+    "«"
+  ]
+}