Use upstream Meta Llama tokenizer default

README.md

@@ -29,7 +29,7 @@ This repository contains a 4-bit TorchAO weight-only quantization of
 - **Quantization:** TorchAO `Int4WeightOnlyConfig(group_size=128)`
 - **Runtime format:** `torch.save` checkpoint containing TorchAO quantized tensor subclasses
 - **Tested GPU:** RTX 3060 12 GB
-- **Tokenizer override used for testing:** `unsloth/Llama-3.2-1B`
+- **Tokenizer:** upstream default `meta-llama/Llama-3.2-1B`
 - **Language:** English, following the base model
 
 No private prompt voice is included. Voice continuation/cloning requires user-supplied prompt audio and transcript.
@@ -50,11 +50,7 @@ source .venv/bin/activate
 pip install -r requirements.txt
 ```
 
-If the upstream tokenizer is gated/unavailable in your environment, set:
-
-```bash
-export MISO_TTS_TOKENIZER=unsloth/Llama-3.2-1B
-```
+The loader uses upstream MisoTTS tokenizer behavior by default: `meta-llama/Llama-3.2-1B`. This requires a Hugging Face token/account that has access to Meta's Llama 3.2 tokenizer repo.
 
 ## Quick smoke test
 

config.json

@@ -9,5 +9,5 @@
     "quantized_modules": "nn.Linear.weight",
     "runtime": "pytorch+torchao"
   },
-  "tokenizer_override_tested": "unsloth/Llama-3.2-1B"
+  "tokenizer": "meta-llama/Llama-3.2-1B"
 }

scripts/export_int4.py

@@ -10,7 +10,7 @@ from pathlib import Path
 os.environ.setdefault("HF_HUB_ETAG_TIMEOUT", "60")
 os.environ.setdefault("HF_HUB_DOWNLOAD_TIMEOUT", "60")
 os.environ.setdefault("NO_TORCH_COMPILE", "1")
-os.environ.setdefault("MISO_TTS_TOKENIZER", "unsloth/Llama-3.2-1B")
+os.environ.setdefault("MISO_TTS_TOKENIZER", "meta-llama/Llama-3.2-1B")
 os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "expandable_segments:True")
 
 import torch

scripts/smoke_test.py

@@ -6,7 +6,7 @@ import os
 from pathlib import Path
 
 os.environ.setdefault("NO_TORCH_COMPILE", "1")
-os.environ.setdefault("MISO_TTS_TOKENIZER", "unsloth/Llama-3.2-1B")
+os.environ.setdefault("MISO_TTS_TOKENIZER", "meta-llama/Llama-3.2-1B")
 os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "expandable_segments:True")
 
 import torch