See Nvidia FFMPEG install docs here. I had to modify their instructions slightly to avoid issues with A100 cards.
git clone https://git.videolan.org/git/ffmpeg/nv-codec-headers.git
cd nv-codec-headers && sudo make install && cd –
git clone --branch release/7.1 https://git.ffmpeg.org/ffmpeg.git ffmpeg/
| """ | |
| # Source: https://gist.github.com/jneuff/682d47b786329f19291d166957b3274a | |
| /// Fix a huggingface tokenizer to which tokens have been added after training. | |
| /// | |
| /// Adding tokens after training via `add_special_tokens` leads to them being added to the | |
| /// `added_tokens` section but not to the `model.vocab` section. This yields warnings like: | |
| /// ``` | |
| /// [2023-10-17T07:54:05Z WARN tokenizers::tokenizer::serialization] Warning: Token '<|empty_usable_token_space_1023|>' was expected to have ID '129023' but was given ID 'None' | |
| /// ``` |
| #!/bin/bash | |
| #SBATCH --job-name=llm-swarm | |
| #SBATCH --partition hopper-prod | |
| #SBATCH --gpus={{gpus}} | |
| #SBATCH --cpus-per-task=12 | |
| #SBATCH --mem-per-cpu=11G | |
| #SBATCH -o slurm/logs/%x_%j.out | |
| # See original source here: | |
| # https://github.com/huggingface/llm-swarm/blob/main/templates/tgi_h100.template.slurm |
| # For directory | |
| du -h ./ | sort -hr | head -n 10 | |
| # For the largest files within a given directory and its subdirectories | |
| find ./ -type f -exec du -h {} + | sort -hr | head -n 10 | |
| # Find venv/.venv folders (limit recursive lookup to 4) | |
| find ~/ -maxdepth 4 -type d \( -name ".venv" -o -name "venv" \) 2>/dev/null |
| import typing as t | |
| import jsonlines | |
| from langchain.schema import Document | |
| def save_docs_to_jsonl(documents: t.Iterable[Document], file_path: str) -> None: | |
| with jsonlines.open(file_path, mode="w") as writer: | |
| for doc in documents: | |
| writer.write(doc.dict()) |
| """ | |
| # See https://huggingface.co/intfloat/e5-mistral-7b-instruct for model inference code | |
| ## Quickstart | |
| Install requirements | |
| ```bash | |
| pip install fastapi uvicorn torch transformers | |
| ``` |
| # TODO: Add this to PR for Langchain so that it will be easy to use across all our different LLM projects | |
| import re | |
| import time | |
| import warnings | |
| from pathlib import Path | |
| from typing import List, Union, Optional | |
| import requests | |
| from tqdm import tqdm | |
| from pydantic import BaseModel, Field, field_validator, computed_field |
| import os | |
| from typing import List, Union | |
| def listdirs(path, extensions: Union[List[str], str] = None): | |
| """ List all files in directory (including walking all subdirectories). | |
| Can filter by file extension by providing either, for example: | |
| extensions='png' | |
| extensions=['png', 'jpeg'] |
| # Uses python3 | |
| import sys | |
| from typing import List | |
| def calc_fib(n): | |
| if n == 1 or n == 2: | |
| return 1 | |
| elif n == 0: | |
| return 0 |