Upload folder using huggingface_hub

README.md

@@ -29,7 +29,7 @@ tags:
 > Native MLX inference on iPhone, iPad, and Mac — no Python, no cloud, no server required at 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.
+No additional training or fine-tuning. The native runtime produces identical decoded text to the Python reference on validated test inputs.
 
 ---
 
@@ -43,6 +43,20 @@ No additional training or fine-tuning. The native runtime produced identical dec
 
 ---
 
+## Quality
+
+Evaluated on **Apple M1 Max**, release build.  
+Normalization: lowercase, punctuation removed.
+
+| Dataset | Type | Clips | WER | CER |
+|---------|------|------:|----:|----:|
+| [Common Voice Scripted Speech 25.0 ru](https://mozilladatacollective.com/datasets/cmn2h1dg201gro107lpynbbd6) | read speech | 995 | **6.76%** | **2.55%** |
+| [Common Voice Spontaneous Speech 3.0 ru](https://mozilladatacollective.com/datasets/cmn1pnb4n00vfmm07eqydzilq) | spontaneous speech | 276 | **9.86%** | **5.83%** |
+
+Spontaneous Speech is a challenging scenario: conversational style, filler words, non-standard vocabulary. WER on read speech is expected to be lower — and the 6.76% result on Scripted Speech confirms this.
+
+---
+
 ## Performance · Apple M1 Max
 
 ### Short audio (~6 s)
@@ -82,12 +96,23 @@ The native runtime is approximately **4.2× faster** than the Python reference o
 The current main bottleneck is the **RNNT joint network** during greedy decoding.
 </details>
 
+### Benchmark · 276 clips (CV Spontaneous Speech 3.0)
+
+| Mode | Time / clip | RTF avg | RTF median | Peak RSS |
+|------|------------:|--------:|-----------:|---------:|
+| Cold (per-process) | 1.87 s | 0.16 | 0.12 | 992 MB |
+| Warm (batch, model in memory) | 0.49 s | 0.05 | 0.03 | 981 MB |
+
+**Cold** — each clip starts a new process including model load (~1.4 s overhead).  
+**Warm** — model loaded once, all clips processed sequentially in a single process.
+
 ### Memory
 
 | Scenario | Peak RSS |
 |----------|----------|
 | Native MLX · short audio | ~1.10 GB |
 | Native MLX · long-form audio | ~1.15 GB |
+| Native MLX · benchmark (276 clips) | ~992 MB |
 | Python reference · short audio | ~1.76 GB |
 
 ---
@@ -120,19 +145,19 @@ It is **not a Python package** — there is nothing to `pip install`.
 | Requirement | Value |
 |-------------|-------|
 | Platform | iOS, iPadOS, and macOS — any version supported by MLX Swift |
-| Architecture | arm64 Apple devices  |
+| Architecture | arm64 Apple devices |
 | Framework | [Apple MLX](https://github.com/ml-explore/mlx) |
 | Language | Swift |
 | Runtime deps | MLX Swift runtime and bundled model assets |
 
+Swift SDK and CLI: [kruatech/gigaam-v3-mlx](https://github.com/kruatech/gigaam-v3-mlx)
+
 To use this bundle in a native app, add the **MLX Swift** package to your Xcode project:
 
 ```
 https://github.com/ml-explore/mlx-swift
 ```
 
-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.
-
 ---
 
 ## Repository Files
@@ -205,7 +230,7 @@ Encoder subsampling factor: **4** → one encoder frame ≈ **40 ms** of audio.
 
 ## Long-Form Transcription
 
-Long audio is intended to be processed in chunks to keep memory usage bounded and inference latency predictable.
+Long audio is processed in chunks to keep memory usage bounded and inference latency predictable.
 
 **Recommended settings:**
 
@@ -264,7 +289,7 @@ End-to-end: the native runtime and the Python reference produce **identical deco
 - 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.
+- iOS/iPadOS memory pressure should be verified on target devices.
 - This repository contains **model assets only** — no application code, no Swift SDK source.
 
 ---

checksums.sha256

@@ -1,4 +1,4 @@
-1d42658ae5dcc0424dade66b07b9b067d329084d53a8e549b830feae8b58c54f  README.md
+f321086edd313c75a1d287f922ab6e851a9f3dabdc78a5820605230919543e87  README.md
 993c538b6dceffb7b018c633a7cc8e4034e7ca975b5c27ecc8ecd679fa574d16  manifest.json
 2ff6d955a61003e0d3cb2155be36c14aa3df69c879713f30880ddd07d71fde21  weights.fp16.safetensors
 828c12c991019eef952a960661f25a92d6ad279591e2ea466b4aeddf1d20a18a  tokenizer.model