Fix README: remove fp32 references, flatten file paths

README.md

@@ -36,16 +36,17 @@ Runs locally with [ONNX Runtime](https://onnxruntime.ai/) in Python, C#, C++, Ja
 
 ## What's Included
 
-Two precision variants, both fully functional:
+INT8 dynamically quantized ONNX model (2.75 GB total):
 
-| Variant | Encoder | Decoder | Total | Use Case |
-|---------|---------|---------|-------|----------|
-| **int8/** (recommended) | 2.6 GB | 147 MB | **2.75 GB** | Best balance of size and accuracy |
-| **fp32/** | 7.3 GB | 581 MB | **7.8 GB** | Maximum accuracy, larger footprint |
+| File | Size | Description |
+|------|------|-------------|
+| `cohere-encoder.int8.onnx` | 3 MB | Encoder graph (weights in .data file) |
+| `cohere-encoder.int8.onnx.data` | 2.6 GB | Encoder weights |
+| `cohere-decoder.int8.onnx` | 147 MB | Decoder (self-contained) |
+| `tokens.txt` | 219 KB | 16,384-entry vocabulary |
 
 Plus:
-- `tokens.txt` -- 16,384-entry vocabulary for decoding
-- Export and quantization scripts (reproduce from the original model)
+- Export and quantization scripts (reproduce FP32 or re-quantize from the [original model](https://huggingface.co/CohereLabs/cohere-transcribe-03-2026))
 - C# inference example
 - Complete porting guide with all concepts explained
 
@@ -63,13 +64,13 @@ import librosa
 # Load audio (must be 16kHz mono)
 audio, sr = librosa.load("your_audio.wav", sr=16000, mono=True)
 
-# Load models (use int8 for smaller size, fp32 for max accuracy)
-enc = ort.InferenceSession("int8/cohere-encoder.int8.onnx")
-dec = ort.InferenceSession("int8/cohere-decoder.int8.onnx")
+# Load models
+enc = ort.InferenceSession("cohere-encoder.int8.onnx")
+dec = ort.InferenceSession("cohere-decoder.int8.onnx")
 
 # Load tokens
 tokens = {}
-with open("int8/tokens.txt", "r", encoding="utf-8") as f:
+with open("tokens.txt", "r", encoding="utf-8") as f:
     for line in f:
         parts = line.strip().rsplit(" ", 1)
         if len(parts) == 2: