bofenghuang

{
  "name": "Vigogne Chat V3",
  "inference_params": {
    "input_prefix": "[INST]",
    "input_suffix": "[/INST]",
    "antiprompt": [
      "[INST]"
    ],
    "pre_prompt": "[INST]<<SYS>>\\nVous êtes Vigogne, un assistant IA créé par Zaion Lab. Vous suivez extrêmement bien les instructions. Aidez autant que vous le pouvez.\\n<</SYS>>\\n\\n[/INST]"
  }

bofenghuang

#!/usr/bin/env python
# Copyright 2022  Bofeng Huang
# coding=utf-8

"""
Usage:
./scripts/convert_whisper_to_openai.py \
    --hf_model_name_or_path outputs/general/whisper-large-v2-ft-french-lr4e6-bs256-augment \
    --whisper_state_path outputs/general/whisper-large-v2-ft-french-lr4e6-bs256-augment/checkpoint_openai.pt
"""

bofenghuang

#!/usr/bin/env python

import logging
import time
from typing import List, Optional

import numpy as np
import scipy.stats
import speechbrain
import torch

bofenghuang

#!/usr/bin/env python

"""
Based on https://colab.research.google.com/drive/1mypqbHDrusZaIbqPoiEGY-WIbnpMHa2I?usp=sharing#scrollTo=bY0AwDiSxdVE

- Random Sampling
- After sorting (minimum padding but no randomness)
- Org Dynamic Batching (the one in speechbrain current version)
- **Mdf Dynamic Batching w/ fitted lognorm** (bucket boundaries set up with lognormal distribution fitted on dataset)
- **Mdf Dynamic Batching w/ fitted beta** (bucket boundaries set up with beta distribution fitted on dataset, mentioned in tuto)