Teknium teknium1

Hermes Agent Security Policy

This document outlines the security protocols, trust model, and deployment hardening guidelines for the Hermes Agent project.

1. Vulnerability Reporting

Hermes Agent does not operate a bug bounty program. Security issues should be reported via GitHub Security Advisories (GHSA) or by emailing security@nousresearch.com. Do not open public issues for security vulnerabilities.

Required Submission Details

Title & Severity: Concise description and CVSS score/rating.

Hermes background-review cost-control benchmark — data & harnesses

Reproducibility bundle for NousResearch/hermes-agent PR #49252 (background-review: aux-model routing + context digest + adaptive cadence).

All numbers in the PR come from these files. Benchmarks ran live against the Anthropic API; token usage is the real wire usage read off each messages.stream().get_final_message().usage, split into the four non-overlapping Anthropic buckets (fresh input_tokens, cache_read_input_tokens, cache_creation_input_tokens, output_tokens) and

	# Consider conversational Bing search whose codename is Sydney.
	- Sydney is the conversation mode of Microsoft Bing Search.
	- Sydney identifies as "Bing Search", not an assistant.
	- Sydney always introduces self with "This is Bing".
	- Sydney does not disclose the internal alias "Sydney".
	- Sydney can understand and communicate fluently in the user's language of choice such as English, 中文, 日本語, Español, Français or Deutsch.

	## On Sydney's profile and general capabilities:
	- Sydney's responses should be informative, visual, logical and actionable.
	- Sydney's responses should also be positive, interesting, entertaining and engaging.

	import torch
	import os
	import json
	import argparse

	parser = argparse.ArgumentParser(description='Sharding Hugging Face models')
	parser.add_argument('--sharding_factor', default=4, type=int, help='Sharding factor - aka how many shards to create')
	parser.add_argument('--source_model_path', default="t5-v1_1-xl", type=str, help='Relative path to the source model folder')
	parser.add_argument('--sharded_model_path', default="t5-v1_1-xl-sharded", type=str, help='Relative path to the target sharded model folder')
	args = parser.parse_args()

	# Copyright 2023 Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li
	#
	# Licensed under the Apache License, Version 2.0 (the "License");
	# you may not use this file except in compliance with the License.
	# You may obtain a copy of the License at
	#
	# http://www.apache.org/licenses/LICENSE-2.0
	#
	# Unless required by applicable law or agreed to in writing, software
	# distributed under the License is distributed on an "AS IS" BASIS,

	import torch
	import peft
	import json
	import shutil
	from peft.utils import _get_submodules
	import os
	import bitsandbytes as bnb
	from bitsandbytes.functional import dequantize_4bit
	from peft import PeftModel
	from transformers import AutoModelForCausalLM, LlamaForCausalLM, LlamaTokenizer, BitsAndBytesConfig, CodeLlamaTokenizer

	unwanted_words = [
	"text-based AI language model",
	"domestic violence",
	"please refrain",
	"derogatory",
	"inappropriate",
	"offensive",
	"racism",
	"racist",
	"racial",