Nous Research logo

Hermes 3 405B Instruct

Hermesv3 405B InstructCurrent
byNous ResearchNous Research(ai lab)
Released August 16, 2024
Run this model yourselfAPI DocsPlayground
Context131.1K tokens
CategoryLarge Language Model

About Hermes 3 405B Instruct

Hermes 3 405B Instruct is a large language model from Nous Research. Supports up to 131,072 token context window. Achieves 87.3% on MMLU. Available free with rate limits on OpenRouter.

Capabilities

textcode

Input Modalities

text

Output Modalities

text

Technical Details

API Identifier
nousresearch/hermes-3-llama-3.1-405b:free
Category
Large Language Model
Context Window
131,072 tokens

Tags

instruction-tunedlong-context

Run this model yourself

Local and cloud setup for Hermes 3 405B Instruct, with exact configurations, sources and clear evidence limits.

Find a model for your hardware
Local hardwareDocumented setupSources checked Oct 17, 2026

Run complete native Hermes 3 405B ChatML conversation and inert tools on eight owned H200 GPUs

Use all 191 native BF16 shards with the exact default and tool_use ChatML templates, full inert tool history and a separate conversation check. Native EOS is 128039. The bounded private reader uses 8,192 total positions and a 4,096-output cap. All-model fit, startup time, speed and quality remain unknown; this preserves the frozen free/provider catalog identity without asserting provider serving-byte equivalence.

Hardware
Existing owned eight-H200 host;2 TiB RAM/disk planning
Runtime
Hermes3 native BF16 TP8 vLLM0.27.1 vLLM 0.27.1 pinned amd64 image plus exact Transformers 5.8.1 wheel; source-qualified, unbuilt · CUDA
Model format
Native HF sharded safetensors · Complete native BF16 dense model; no quantization or omitted tensors · Native unquantized BF16; all dense model tensors retained
Workload
8,192 token context · 1 concurrent

Based on source documentation; this setup has not been executed by OpenTools. Memory: unknown. No compatible measurement available.

Read the complete setup guide

Before you start

  • Use an existing owned dedicated Ubuntu 24.04 x86_64 server with eight H200 GPUs, a 64-host-thread allocation, 2 TiB RAM and 2 TiB working disk planning allocation. Exact CPU SKU and actual GPU usable memory/topology are reader-inventoried. Keep port8994 loopback-only. No cloud capacity, memory or interconnect fit is transferred to this host.
  • Review the publisher license/base discrepancy and both complete official agreements/AUPs. Apply all applicable conditions, notices, attribution, commercial thresholds and prohibited-use limits. No terms acceptance, legal clearance or base-training revision proof is claimed.
  • Use the exact dense Hermes 3 Llama 3.1 405B checkpoint, not a 70B, FP8 or GGUF derivative. Retain all 191 BF16 shards, tokenizer assets and both native ChatML templates. Nominal device capacities do not prove fit.
  • The 8,192-position total and 4,096-output cap are bounded native conversation/tool fixtures, not full 131,072-position validation. Preserve complete history and refuse truncation. Additional useful tasks/backends/family scopes remain open.

Exact model and requirements

Artifact:
Nous Research / Hermes 3 Llama 3.1 405B full native BF16 dense checkpoint · checkpoint NousResearch/Hermes-3-Llama-3.1-405B
Publisher revision:
b9fe8ede1f5075f171b2c0a235e20ae4aba33767
Artifact digest:
Complete pinned all-asset manifest verifies every native repository file by published SHA256 or Git blob identity; no invented whole-bundle digest.
Download size:
755.96 GiB
Memory requirement (unknown):
Unknown — Full native BF16 target, TP8 partitions, KV, collective/kernel workspaces, runtime and peak per-GPU memory have not been measured.
System RAM / disk:
Unknown / Unknown
Placement:
supported sharding. Native vLLM Llama tensor parallel eight; 128 query heads and eight KV heads are divisible by eight. Per-device weight balance and peak usage are not measured. Full target on eight GPUs; no CPU/disk weight offload. Host staging and download buffers remain additional.
KV cache:
BF16 attention K/V via explicit model dtype and auto cache dtype
Workload:
8,192 token context; Unknown; Unknown; concurrency 1; batch 1; cache cold; repetitions Unknown.
Tokenizer:
Exact native Hermes 3 tokenizer with distinct embedded default and tool_use ChatML templates · b9fe8ede1f5075f171b2c0a235e20ae4aba33767
Sampling / reasoning:
Reader fixed deterministic protocol prescription temperature 0.0; not a publisher recommended sampler or quality proof. No Hermes4 thinking flags or parser. Native ChatML starts with one BOS and ends assistant turns with im_end128039. Complete raw tool-call assistant content and the fixed inert tool result remain in history. Exact native template selection, complete token IDs, decode identity and integer usage are checked; no external tool executes.
Runtime prerequisites:
Ubuntu 24.04 x86_64 host; pinned Ubuntu 22.04 / Python 3.12 defaults container; NVIDIA Linux driver >=580.126.20, Working Docker GPU support with NVIDIA Container Toolkit; actual visibility preflight required.

Official artifact · License

Publisher card metadata declares llama3 while naming the Llama 3.1 405B base. The native LICENSE probe returned scoped404; the preserved USE_POLICY evidence remains separately bound; this does not erase its explicit metadata-linked grant. Full official Meta Llama 3 and stated-base Llama 3.1 agreements/AUP are archived separately, with no equivalence or publisher-intent claim. Retain both applicable agreement copies, notices and attribution (Built with Meta Llama 3 / Built with Llama); comply with redistribution and derivative naming requirements and both AUPs. Each agreement has its own release-date 700-million-MAU prior-month commercial license trigger. Preserve Llama 3 restrictions on using materials/outputs to improve other large language models, subject to its stated derivative exception. Patent/IP litigation, termination, warranty and liability conditions apply. This inert own-use walkthrough provides no legal clearance and accepts no terms on the reader behalf.

  • 811,706,916,800 serialized weight bytes are a full-model storage lower bound, not total memory or peak. Tensor-index metadata and file sizes are separate accounting.
  • Eight advertised 141 GB H200 devices are not silently converted into GiB. Usable per-device bytes and runtime partition balance are unknown; no aggregate-fit claim.
  • Native Llama has 126 dense layers, width 16,384, 128 query/eight KV heads and no MoE active-parameter shortcut. Host staging, Docker image and tokenizer add overhead.

One-time setup and hardware

hardware
Unknown — Existing owned host acquisition cost unknown; no cloud rental assumed.
Existing owned host acquisition cost unknown; no cloud rental assumed. Tax: unknown.
setup
Unknown — Provisioning/imagebuild/labor cost unknown.
Provisioning/imagebuild/labor cost unknown. Tax: unknown.

Recurring costs

Owned-host electricity
Unknown — Owned host acquisition/electricity/network/administration and storage costs are reader-specific and unknown.
Existing owned eight-H200 host; no cloud rental or measured usage cost assumed. Tax: excluded.
storage
Unknown — Owned host acquisition/electricity/network/administration and storage costs are reader-specific and unknown.
Existing owned eight-H200 host; no cloud rental or measured usage cost assumed. Tax: excluded.
egress
Unknown — Owned host acquisition/electricity/network/administration and storage costs are reader-specific and unknown.
Existing owned eight-H200 host; no cloud rental or measured usage cost assumed. Tax: unknown.
administration
Unknown — Owned host acquisition/electricity/network/administration and storage costs are reader-specific and unknown.
Existing owned eight-H200 host; no cloud rental or measured usage cost assumed. Tax: unknown.

Install the runtime

  1. 1. Prepare an already owned eight-H200 instance

    Install host utilities on the existing dedicated eight-H200 host with driver >=580.126.20. Inventory actual device memory, compute capability and peer topology before native loading; no acquisition/provisioning is performed. The nominal2TiB host may report lower usable Linux memory;1900GiB MemTotal and1800GiB free disk are conservative reader allocation guards, not measured minima or fit. Inspect the actual eight-GPU topology output for intended peer/collective connectivity before loading; stop on unsupported connectivity. Compute capability9.0 and eight uniqueGPU UUIDs are explicitly checked.

    Command
    set -eu
sudo apt-get update
sudo apt-get install -y docker.io python3 curl ca-certificates openssl
sudo systemctl enable --now docker
mkdir -p "$HOME/hermes3-405b"
cd "$HOME/hermes3-405b"
nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv,noheader
python3 - <<'PYPREFLIGHT'
import csv,subprocess,shutil,os
assert os.cpu_count()>=64
mem={k:int(v.strip().split()[0])*1024 for k,v in (line.split(":",1)for line in open("/proc/meminfo"))if k in ("MemTotal","MemAvailable")}
print("Actual Linux memory bytes",mem)
assert mem["MemTotal"]>=1900*1024**3
r=list(csv.reader(subprocess.check_output(["nvidia-smi","--query-gpu=name,memory.total,driver_version","--format=csv,noheader,nounits"],text=True).splitlines()))
assert len(r)==8 and all("H200" in x[0] for x in r)
assert all(tuple(map(int,x[2].strip().split("."))) >= (580,126,20) for x in r)
assert shutil.disk_usage(".").free >=1800*1024**3
print("Eight H200 devices and prescribed disk/driver checks passed; no model fit measured.")
PYPREFLIGHT
sudo docker version
python3 - <<'PYCAP'
import csv,subprocess
r=list(csv.reader(subprocess.check_output(["nvidia-smi","--query-gpu=uuid,compute_cap","--format=csv,noheader,nounits"],text=True).splitlines()))
assert len(r)==8 and len({x[0]for x in r})==8 and all(x[1].strip()=="9.0"for x in r)
PYCAP
nvidia-smi topo -m

    Expected result: Record eight actual GPU identities, usable memory and drivers. Stop on missing devices, driver/GPU Docker failure or inadequate prescribed disk. No native fit claim.

  2. 2. Configure NVIDIA GPU containers on this dedicated host

    Use the official production repository and exact documented Container Toolkit 1.20.1-1 packages. This modifies Docker daemon configuration and restarts Docker on this dedicated host; preserve its original configuration and do not apply on a shared production daemon without its owner review. Stop on repository/signature/package errors.

    Command
    set -euo pipefail
sudo apt-get install -y gnupg2
curl --fail --silent --show-error --location https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --batch --yes --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
curl --fail --silent --show-error --location https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list >/dev/null
sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit=1.20.1-1 nvidia-container-toolkit-base=1.20.1-1 libnvidia-container-tools=1.20.1-1 libnvidia-container1=1.20.1-1
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

    Expected result: The exact GPU container packages install and Docker is configured. The subsequent container nvidia-smi check must pass; this is not model inference or fit proof.

Download the model

  1. 1. Write native reader source bundle segment 1

    This is reversible compressed UTF-8 source, not model weights. Decode and inspect all exact reader files before use.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat > reader-bundle.b64 <<'B64'
eNq9fWtzXMeR5ff9FS3OhwYssFXvBygowmNr1t6dsRWSZmM3SC4iqyqLbBPoxnY3KHIU/O978nY3CFCQrCtFXCtMAPdZWZmVeU5WVd4fn6xot3zLT2m75d128dYs/rFdr56c//hk+5qMD0/On2jne4yh66Irk48lBuUTuWJwiFzTqlb2HMjo0kvztWmyJRcuOeOoenL2ZMfvdnjSjy9Ws9mLJxu+WW+Xu/Xm/Ysn5/j7b+vb7be8ZdrU11/8hTfXvH1qn/77FV3TU7vQT53y//riydnx5rfL7XK92t9acufEjXX3Kvquoy6mKjLWs1HEjgpZG0O8u70vr3gr9z6Xv2ezH/c/5NSKrnn/1MWr5Y52u82y3O7k6rOPF22X/zVcpL3O9w7T1av1Zrl7fb1/AO5/Wq7W5Sk6UT+4vy1f8Xa3v4qCdY6opqqy8bbYEqzqBX1NnJ2LiUiZ6MuLJ/v7P5z9Qpu//fqPf/6PrxfX7fHmGu/d721vrSYpCiqwcrUzJ6+rMyrHbnrr6P8IUazRv6q9db3qy1eDtT3a4pTS720vx5pVKQWNs6klm2xu3VHMJmgfMhNrTsr9uva+4hVvMFbWq8t/1nTt4u9tunXKdio2Zd1dsxRMbDqWEtDrLcNSSkxMuv2qpl+vG189Vfiffrru+EVnvdhS5x2vtuvN4wbukgohmpTUz8uydxE/K0ThnhyspVMi5uZ7M5m8NS0FV0vXlryjmpldSrYxRdODhSFl5cjW6EYKZ0YIp0xAQ4xxv1k4HV22gaxvtdqgWskhFAd35C2cHntXyDBjfISagvcY1SaWlIKqqbfQ4kjh7Ajh4EV8TFqb36653jWUhPFDpVWL18fY0e4auIRaMIwIAwseK6sGiSFviFwtxITMgdJI4dykwvmastc+G1bVBWVDSL4Ek4oXc4SZEqucqzccW8Boy7039rXg1VqxoZHC+V8vnHUJVqR1/h2aC74qD0NLmkzRAW2naE0vuiEyUmJLJsN3VN0w8qxiSF8xDmJGdA/OjhQuTKq5Vpy2DRKyYxMj7C0T1IPA03VuKXNW3DQjRhWXqtZBAZfEWrTqPhhfRwoXJxWOfI3kFQI/GWWs9ap2pwN1131sXpnqI0FGZ22zTjOFVlVF7DWpwqOUkcKlSc0SYEZ72BepmAr1WLWBl89VFbiWJgMuwEJNyy1HU6BYLjZoDLsCJ9TJjxQuT6o57eDyKqtoXEY4aBqOsvdSgd8AkEgF3210utraXKeovSdVm8J1AH3K9XHCaTWpcA7xqzUIRwrQwyaEaANxYROM23KGz8y6ateiA35VsMlcSgUNsFxqSyNDgdaTmqUnZ7V3ttteoBJtU3GthBQSwjlMM4daVSw+WQXfCSNOcKi9uaBTsRieI4Uz0wZxYzp1YEZAXqdSbAZ8wnefO5GDGjVDUyooXeFD4GZMLSYbyOxDsVGHkcJNi1AqqB68AxBkN9qUnkE8MN4ywBd0CWcSPANRVmd9jsSmWWUL/A7CO1jKWPil3aRmaRGiaysAqa20bBT+zsHArTn4+gS63ShBnyI+BNSxWcRyRAQFIgwKlkcK56fVXITfK67kkmvSoXCA9zBoNrRCoAeAnRpxvlEmkHpwtwQVI95X8gn4bKy3/A0Ixfx2ymMrldJcEzdobaxsdCcMM7gOQwaE3+emAc4CAnrFtS3rGKJzmuBA4X1GChcn9pY9Ku5aYpBJ2mHIeaM9ZXQZQpqQRbADDbjlO6AXJ1A9Br92RiJFGYktdZpUc1oDdqSevEoKUpjcSHUYowlgwhxVqdIUcoDWKiWKLXkPJ2nYBgb0HMnEdZ5UOI7w64jY0FPlVBIAc7YAxZwBJDXML0fI1p0S6ZNRwNelAqngNmvAascJZ9SkZlmZ4EpKNs1hYMGLwE8ErzRAV06kO9g3AkD0XGNBhAdD0B1sVVkVoFYeGeeMnlRzgBwYP67H1rU3JYDVZZ2zTTFzt/jJWZCn7a6YwKrDvRjuDZxVV2Y/EqEYM6lwPuD2ZgConCUQtQiS1vBDRYW4XcB1IoH2dJCGBAN1oWXVkgH/kSiuRgZxYyemPNpbRDVnilJVk/alACr36lsKrFU1GIOknQdSjuJkUoTDMQaRPnIrY8ecm9ahJJdtBJaEzhrF5LOi4ot1FaCyK+HdiGw69mgkL8weVtnwThyCsGnsmPPTBnHl4FFs8g2UBtCkKQTuiC72JcTQY+7ea+iLagALpwqwgkgvCcEGxGZHJohMmNQs4c6NZnBR9BOwv8/GJOWMR9DORKwJ2DKQ7VBfRigntoh5oUnmMjqY6kjh4qSay1qh/U7h354hQSsuBUPcCYrL8I6lxKDlHSVGYeqUKMdYuvUV/FyPFG5ahKLAeIqOkTK4QQb9Vj4haluEt9YR4GzLjQnuxHSANPEiHsgllRoqox9Gwi+Tp8WW1VhdvbUEj8+RWuEeGNCKrAOvaaoyGHghVmB8BRiUQh6SD9B2yW1knLNqUs1RNE6DFtRYrc2q2Fy46pDAZywgVugqG/BvkNhkAlClDmyst8VbDa9CaqRw0yIUvLCCjMIMYwul95Q9w7VQcg2UuxJCNki3MaW6UEDpHPSqQA0yggTY30jgbM2kZqkhhwH3BAEtwMu2IXzn3IKDZVZECbByDT7HxjDwmAJathZMDp7UA4zSWLO002ouZFdhcDo5mGTX8P2tOysIuhgSHkfaJA6hd2AZw8Ug5mdWYA0U62iznBah5JQMRlO1seFRVbcOctOsBZh0wP/kwA0gPVxJdeAH4OreIZ7naltJNo7MW9ppZ3nAvxG/wb1jEWkKjA5x2uIoFOa6TQjmMoEAhpO6TwwgDZEhoM255jQyh2KnzaH0CsdXwc04+B5c6dAawluKvQZg6d4KTDPVHEuOsujEMOhRIVtSg0MdLdy0CCUacNUMzcEuW1W6FJnq6AZPVmB4OQcXrTfUuIDElc45OyCZJsTTjs5+2WlnedhqlbuLqQB+wFUHOBWnU8uR4Dc8joKbFybXPITpBaHewY9YECLN2o8Vbtociks6FoZrD05gpIoNfLQlDY7HCXEdKMX6ECBlqTXlEkHMBYKaZqvhNHIKy02LUDrIaWuA+GClGEsZwLlFa1P01TKAbtMIekbJPDIioQVi1qlkBfmsCzmOJKtu2lmeDAwVbU4I366VnhOGWgB2DKB1NbJ1JRpfJbVnZcYRbN31poKrrWmreKRZumlzKBrW5xIUAviV0VWFSKnmOJQKVprYe8TzWmwgRY4QxS2COgK98rr20kZSHmcnBs7NWFU1h0gyG1eq7V0pT1Y4uIXFthAwKvGvtaGp3HRhbQZIllmPRChu2lkeeHNtOqCGjbaW7KAfGFtXjJiuXOiOTUoIFSbZoKpHWABU0xiR2hUufWSawU2bQwG+cg2gC34S3CaqRhpwWXGFRhXYKKEhGI7BtAb/6Cw18B+rWqox1dRHToS4iRFKJ9ZwDjmzCckrx6AFwYIGIGhXdtoAmymAS1sKdMUwS+AMDafptW801qHEiVcQaa2tbwzckXx0PetKDAaUXATApNg1d+PgIoMGWlEWxKHBqZCxCBrVjBRu2hyK6Spw0CCqreoEIlpTshSqBzWnYkBUc4Bf9AEk3HvfNORuLRuTC/jd2GljNy1CSSqx8sOSL1AbWWySeqxQZSiBPPwGdd0QA7JlsiA7GJQBylOhhwhMOjJB5Ked5RGdwcgUI8A6X6upwrC9rM0LgrqsQ7Qr1aTgbS4BFMJTgqcJ0cWCt48UbtocCnQFMJJVT1W5Dt8IUGKUrPYCwKwy9d+NaTDVys22zIgLMF3bM2IBOnZkUtZPi1BqdbkCjbQgHFTYmyohgXX3ki3UI6RHCcnJJqgC6MzaalyeFVUEhZEJIj/xLI+G07MSyBNidww5VIwwUFbvRDWWZN0UWtO0KWw9IEyWXC26g3WPfWSawU+bQynKZCNzcLoipCnE6qB96Rbe34KYA5U1bogIBfCsExmwItUi5VRMbT6MDAV+WoQCygY9OAArZasVwJhjZfwHEBZzcjpqw/AksVO2gJraO1myAmzWjAMzGinctLM8ASTcFTDr7HTupWTQUsCS1qj0ImsUOwahoo4o2MgGhmkCT2sLfmtiGB0Kps2hFFmjXTnCVcpkTsgmV7ImyTx4MRCcNRUva/oIOgyy7LkEUWXzNiBIjhRuWoTiDDQCdi1EPFhDFgyutkTgbJBMZrWcqUqSXWDrRUdGDDQVPlYrFbmP1dy0szyWWwdR9QZeHnARQkFCEIGkJPkKkEIO8rAyOkrQiDIDpAFeUq4yQTQyzoVpcyjAiK6RV9ZlsDQZYa2E3ICVm2KMxA6NWnZeJ07dJi3zINll642ssmkjvWWYFqEMCXLiVlKsEWMtGlldCWIOhw8rJNkE1oBPqtXNdEcNlEhHBImSKAFijhRu2lkeoBOfspK9PEAkppWWjGY4RRhgBRM3NsCHhOJikMUpMZLV0GJHQ13XZSRCCdPmUGRnpqRLUoSDjwrMVSHSsY8d0Ku51msysupXJZMb8JfMfzjtEShgtcaP9JZhWoTibG4VugHsr9mpLCBLdW+gpODIRVaIBly4AGBXylrBw+iWiiZduPHIIB6mneWxzsFDaGCp5kwjuIhUgZ6TTwzuWsBvwOECeDniWu69thq9rP0C+OScwsjUXpg2h5I8KLiXnaS5VC6UqXmA6ERKeAKbgkinZf8rLsAgMwF0zwTqIScAttFjbmKE0msHFo7ZuwbBEKhzYK9dt0qmBmSKBxBaRSN7rnIwsBNJODRYGBV2I6eNw7SzPB4u38TmKDvipDK5ZnzzsUUGP7XM0Wbwcq4Bob4gzDX4TTggRoyH+kYilDBtDiVqqrr0qK2Fo466RE9UVGiwQ5cbjJBliawr0QFfm+Rq8rYm1YHIADdH5i3jtAilYLwVCtEmr2BnPoPTcaISSwLKyg7YxACkuGgkJdvYxag7pFbaV2DPkd4y6onXW2bVqo0EIOmy4dq9T6ZAeQjhRiM+BN9l+6rL3uQYM5CJDUoZSYJ5NRJ+xWlzKFCWpIC8ldnTbmustSSDWO0AHbsOiO6hU6vAJxnARc5qKMCZZGDRbeSYixPP8lhZpg1okmAEFR6Sja9K7LD40GGFHMFbJbXA2fnkjGRxNXWlE/lCI+NcnHgvD1VAjqAhSNDehJQYSmtZy6pZRAnmjlBoEg4l4yOxklWnqsClMDjCyGnjOG0ORbGqqjQGKsmUAUhkZs7n2sC7weMsWaFuQFzg305WpmspJVKzbq70HEYG8TgtQlHwD1UBcphSMKZAxp3mqAC2igG4LFAWnD9H4LME9WZrTC3FyXLa5vTYNc5x4r08XJTnxLIvVztn8CzZNx4jHo1/4DJc4e6rLAHAKeMR3mC8XdWuyIzdYhbTxJsmsqyvBLeuQbvk4f0jNKNlTam1tciOY0CxUGt3rmeXonUE6uBqKECfY73ltAjFJ+U1yHcBtbZVdyNVTiRZiQgdSg0kVRlkSUPL8KsAKGBy0clKh2DTaOCc1MTp9DwAK4AwA0DcQUmz8Z40Z/BzHRG5e9DKwZu0Qsol2cwUsoodQTGEkTOradocCqiZbM4xMkdnAb7IIDw32CcT4GMITCbK1jOgaOfAfCwcDXc4GnhLIJiRoSBNi1CsJNEDV9n9WGpDFMhZ1jgXLbkVtjkS6SQQzflghTdU02UZQx+KU4w1y2lnedgFIMUmL/Y5VU1MnENv3jcD9l0hDLByaSCwgGnw5VXFFrwOLUmhnpE5lDRtDkW2U3l4EIyiDL7FoDy1l1JjMsykDQaZRmiPNrba4UpDNBm4GnBUSQZ67JibFqFoSWn1AB+CuKWYHVCxNgUD0VDO+FOongJrCPCY8I6d4IFU98lGqWQz1iynneUh2YvkbKXW4RF9Ju9aAA/SJHV7dGpN5QLWQ0CTyTKR1Ta16huGHzQ71qHEiaeNCSwV2L+HKFWUYg0pS2pPNnf6KHVBGqXuZT2mY2pJ0inwp9U5yNlGrv1KE+/lqa3LdGrK
B64

    Expected result: Source segment written; no model or native code executed.

  2. 2. Write native reader source bundle segment 2

    This is reversible compressed UTF-8 source, not model weights. Decode and inspect all exact reader files before use.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat >> reader-bundle.b64 <<'B64'
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
B64

    Expected result: Source segment written; no model or native code executed.

  3. 3. Write native reader source bundle segment 3

    This is reversible compressed UTF-8 source, not model weights. Decode and inspect all exact reader files before use.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat >> reader-bundle.b64 <<'B64'
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
B64

    Expected result: Source segment written; no model or native code executed.

  4. 4. Decode and verify all native assets

    Reader downloads full pinned native repository files only after inspecting the source bundle. Stop on any size/digest/path mismatch; never bypass. This does not accept terms on behalf of the reader.

    Command
    set -eu
cd "$HOME/hermes3-405b"
python3 - <<'PYBUNDLE'
import base64,zlib,json,hashlib,pathlib
b=zlib.decompress(base64.b64decode(pathlib.Path('reader-bundle.b64').read_text()))
assert len(b)<262144 and hashlib.sha256(b).hexdigest()=='ffa8bfa697d5c8ae74c373527a2a8a8cf57a90032f4e1a5da3e8fe53ba5c2c2d'
for n,r in json.loads(b).items():
 assert pathlib.PurePosixPath(n).name==n
 p=pathlib.Path(n);assert not p.is_symlink()
 raw=r['text'].encode();assert hashlib.sha256(raw).hexdigest()==r['sha256']
 if p.exists():assert p.read_bytes()==raw
 else:p.write_bytes(raw)
PYBUNDLE
python3 reader-download.v2.py

    Expected result: Complete regular-file asset checks pass, including all 191 shards, before native tokenizer loading.

Configure

  1. 1. Bind the existing private reader key

    Use an existing authorized ASCII Bearer key; no new account credentials are created. Keep the env file private; do not publish logs or process metadata containing secrets.

    Command
    set -eu
cd "$HOME/hermes3-405b"
test -n "${HERMES3_READER_API_KEY:-}"
python3 - <<'PYKEY'
import os,pathlib
k=os.environ['HERMES3_READER_API_KEY'];assert 32<=len(k)<=256 and all(33<=ord(c)<=126 for c in k)
p=pathlib.Path('.api-env');assert not p.is_symlink()
fd=os.open(p,os.O_WRONLY|os.O_CREAT|os.O_TRUNC,0o600)
os.fchmod(fd,0o600)
with os.fdopen(fd,'w')as f:f.write('VLLM_API_KEY='+k+'\nHERMES3_READER_API_KEY='+k+'\nHF_HUB_OFFLINE=1\nTRANSFORMERS_OFFLINE=1\n')
PYKEY

    Expected result: Private0600env contains the same existing server/client key and explicit offline environment.

  2. 2. Build the pinned native runtime layer

    Official vLLM amd64 image is immutable; the same source-qualified exact Transformers replacement used in prior native guides is explicit. This image is unbuilt here. Stop on pip check, version, CUDA, tensor-parallel or loader errors; record realized versions.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat > Dockerfile <<'DOCKERFILE'
FROM vllm/vllm-openai@sha256:c2f3b1b964e47809b722b5e75b61b1e7b39a50f70388cf2bf2418f16a9f31da2
RUN python3 -m pip install --no-deps 'https://files.pythonhosted.org/packages/fc/b1/8be7e7ef0b5200491312201918b6125ef9c9df9dd0f0240ccef9ac824e6b/transformers-5.8.1-py3-none-any.whl#sha256=5340fb95962162cdfdae5cc91d7f8fedd92ed75216c1154c5e1f590fcf56dd0e' && python3 -m pip check
RUN python3 -c "import vllm,transformers; assert vllm.__version__=='0.27.1'; assert transformers.__version__=='5.8.1'"
ENV HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1
ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]
DOCKERFILE
sudo docker build --pull -t hermes3-405b:reader .
sudo docker run --rm --gpus all --entrypoint nvidia-smi hermes3-405b:reader
sudo docker run --rm --entrypoint python3 hermes3-405b:reader -m pip freeze > resolved-container-environment.txt
sudo docker network inspect hermes3-private >/dev/null 2>&1 || sudo docker network create --internal hermes3-private

    Expected result: Only after actual reader execution: package/version guards and all eight GPU visibility pass; no model load implied.

Start the model

  1. 1. Start complete native BF16 with tensor parallel eight

    Foreground lifetime is 1,800 seconds including loading; it is not an inference deadline/SLA. Native full loader must account for all model parameters; do not bypass diagnostics, precision or architecture failure.

    Command
    set -eu
cd "$HOME/hermes3-405b"
python3 -c 'import importlib.util,json,pathlib; s=importlib.util.spec_from_file_location("r","reader-client.v3.py");m=importlib.util.module_from_spec(s);s.loader.exec_module(m);m.verify_assets(pathlib.Path("model"),json.loads(pathlib.Path("native-assets.v2.json").read_text()))'
timeout --signal=TERM --kill-after=10s 1800s sudo docker run --rm --name hermes3-405b --gpus all --network hermes3-private -p 127.0.0.1:8994:8000 --shm-size 64g --env-file .api-env -v "$PWD/model:/model:ro" -v "$PWD:/work" hermes3-405b:reader --model /model --served-model-name hermes3-405b-native --host 0.0.0.0 --port 8000 --dtype bfloat16 --kv-cache-dtype auto --tensor-parallel-size 8 --max-model-len 8192 --max-num-seqs 1 --max-num-batched-tokens 8192 --gpu-memory-utilization 0.95 --no-enable-prefix-caching --enforce-eager

    Expected result: Only after actual reader initialization: complete native model starts without missing-parameter/collective/CUDA errors and private health becomes ready. Unknown fit may prevent startup.

Connect to your endpoint

http://127.0.0.1:8994/v1/completions

Owned server loopback only. Run checks on this host; public port8994 remains closed. No cloud SSH or rental prerequisite applies.

Access: loopback. Stable private Bearer key protects native v1 routes. Native health/tokenize are unprotected and must remain private. The model server is trusted single-user research, not a hardened public endpoint; max-num-seqs 1 is scheduling, not admission or a bounded user queue. Only the fixed client is used.

Verify the setup

  1. 1. Verify native ChatML conversation and full inert tool history

    From a second host terminal check loopback access, unauthenticated v1 rejection and then run the exact native local tokenizer/client inside the serving container. All tool outputs are fixed synthetic data, not an executor.

    Command
    set -eu
cd "$HOME/hermes3-405b"
curl --fail --max-time 5 http://127.0.0.1:8994/health
test "$(curl -s --max-time 5 -o /dev/null -w '%{http_code}' http://127.0.0.1:8994/v1/models)" = 401
sudo docker exec hermes3-405b python3 /work/reader-client.v3.py

    Expected result: Successful client exit and populated parseable result.json show all three strict protocol turns passed. Manually inspect the actual call, synthetic seven-plus-one answer and independent default-template response; no accuracy or throughput guarantee.

Performance evidence

No compatible measurement available. Documented support and estimated memory do not establish response speed.

Limits and troubleshooting

  • No install, build, tokenizer execution, weight download, native inference, provisioning or spending occurred. Source compatibility is not measured runtime success.
  • Native config has 131,072 positions. Reduced walkthrough budgets do not establish full native window fit, quality or speed. Input count includes one BOS, tools and full retained native history.
  • Native card llama3 metadata and Llama 3.1 base identify distinct agreements; both actual term/AUP sets retained. No later-base license silently replaces the linked license or grants legal clearance.
  • Eight H200 per-device usable capacity, actual TP balance, kernel/collective buffers and peak are unknown. All weights remain; no smaller model or offload workaround is silently substituted.
  • Client 600-second socket timeout does not cancel GPU work. Foreground 1,800-second lifetime includes initialization; stop/kill the named experiment and cloud billing separately.
  • Docker workspace mount contains the private key; the container and Docker administrator are trusted. No arbitrary tools, uploads, URL fetch or remote model choices are provided.
  • Hermes 3 uses ChatML EOS 128039 and im_start128040; default and tool_use templates are independently selected from the exact native tokenizer. No Hermes4 thinking/control IDs apply.
  • The full serialized weight size is a storage/accounting lower bound, not a GPU or host minimum or a proof of impossibility on other allocations.
  • A result is successful only after the client exits successfully with populated parseable results. Human semantic correctness is a separate reader check.
  • Frozen free/provider identity does not prove API availability or serving-byte equivalence. All useful backends and family workloads remain open.
Asset or native initialization failure
Stop on any digest, missing file, unexpected loader key, tensor-parallel, CUDA, version or tokenizer mismatch. Recheck pinned original sources and device inventory; do not delete components or substitute quantization. No errors are treated as a healthy run.
Protocol or semantic failure
Do not strip malformed output or treat length finish as completion. Preserve actual raw response privately, stop the job and inspect native EOS/template/history and all token budgets. A valid EOS at the4096cap is accepted only with correct raw IDs, decode and usage. Reader records actual errors; protocol pass does not establish semantic quality.

Stop the model

  1. 1. Stop the named native experiment

    Use a second server terminal; foreground timeout is not proof the container exited. Kill only this dedicated experiment if it remains running.

    Command
    sudo docker stop --time 10 hermes3-405b || true
if [ "$(sudo docker inspect -f '{{.State.Running}}' hermes3-405b 2>/dev/null || true)" = true ]; then sudo docker kill hermes3-405b; fi
curl --max-time 3 http://127.0.0.1:8994/health || true

    Expected result: The exact container and endpoint stop. Existing host and electricity remain reader-managed; no cloud billing assumed.

Uninstall

  1. 1. Remove dedicated checkpoint, runtime tag and private files

    Preserve wanted results and legal/source records first. Confirm dedicated paths. Shared drivers/Docker and retained image layers remain; no broad prune.

    Command
    sudo docker rm -f hermes3-405b 2>/dev/null || true
sudo docker network rm hermes3-private
sudo docker image rm hermes3-405b:reader
rm -r -- "$HOME/hermes3-405b"

    Expected result: Dedicated files, results/key and image tag/network removed; shared packages/layers remain.

Sources and verification

Sources checked Oct 17, 2026. Verification source

Compare this configuration
Cloud serverDocumented setupSources checked Oct 17, 2026

Run complete native Hermes 3 405B ChatML conversation and inert tools on AWS p5en.48xlarge

Use all 191 native BF16 shards with the exact default and tool_use ChatML templates, full inert tool history and a separate conversation check. Native EOS is 128039. The bounded private reader uses 8,192 total positions and a 4,096-output cap. All-model fit, startup time, speed and quality remain unknown; this preserves the frozen free/provider catalog identity without asserting provider serving-byte equivalence.

Hardware
AWS p5en.48xlarge, eight H200 GPUs and 2 TiB host RAM
Runtime
Hermes3 native BF16 TP8 vLLM0.27.1 vLLM 0.27.1 pinned amd64 image plus exact Transformers 5.8.1 wheel; source-qualified, unbuilt · CUDA
Model format
Native HF sharded safetensors · Complete native BF16 dense model; no quantization or omitted tensors · Native unquantized BF16; all dense model tensors retained
Workload
8,192 token context · 1 concurrent

Based on source documentation; this setup has not been executed by OpenTools. Memory: unknown. No compatible measurement available.

Read the complete setup guide

Before you start

  • Use an existing p5en.48xlarge in us-east-1 with Ubuntu 24.04 x86_64, 192 vCPUs, 2 TiB RAM, eight H200 GPUs and an allocated 2 TiB persistent gp3 root disk. Keep TCP 8994 closed publicly; restrict existing SSH access to your client. Confirm quote, quota and capacity before any billable action.
  • Review the publisher license/base discrepancy and both complete official agreements/AUPs. Apply all applicable conditions, notices, attribution, commercial thresholds and prohibited-use limits. No terms acceptance, legal clearance or base-training revision proof is claimed.
  • Use the exact dense Hermes 3 Llama 3.1 405B checkpoint, not a 70B, FP8 or GGUF derivative. Retain all 191 BF16 shards, tokenizer assets and both native ChatML templates. Nominal device capacities do not prove fit.
  • The 8,192-position total and 4,096-output cap are bounded native conversation/tool fixtures, not full 131,072-position validation. Preserve complete history and refuse truncation. Additional useful tasks/backends/family scopes remain open.

Exact model and requirements

Artifact:
Nous Research / Hermes 3 Llama 3.1 405B full native BF16 dense checkpoint · checkpoint NousResearch/Hermes-3-Llama-3.1-405B
Publisher revision:
b9fe8ede1f5075f171b2c0a235e20ae4aba33767
Artifact digest:
Complete pinned all-asset manifest verifies every native repository file by published SHA256 or Git blob identity; no invented whole-bundle digest.
Download size:
755.96 GiB
Memory requirement (unknown):
Unknown — Full native BF16 target, TP8 partitions, KV, collective/kernel workspaces, runtime and peak per-GPU memory have not been measured.
System RAM / disk:
Unknown / Unknown
Placement:
supported sharding. Native vLLM Llama tensor parallel eight; 128 query heads and eight KV heads are divisible by eight. Per-device weight balance and peak usage are not measured. Full target on eight GPUs; no CPU/disk weight offload. Host staging and download buffers remain additional.
KV cache:
BF16 attention K/V via explicit model dtype and auto cache dtype
Workload:
8,192 token context; Unknown; Unknown; concurrency 1; batch 1; cache cold; repetitions Unknown.
Tokenizer:
Exact native Hermes 3 tokenizer with distinct embedded default and tool_use ChatML templates · b9fe8ede1f5075f171b2c0a235e20ae4aba33767
Sampling / reasoning:
Reader fixed deterministic protocol prescription temperature 0.0; not a publisher recommended sampler or quality proof. No Hermes4 thinking flags or parser. Native ChatML starts with one BOS and ends assistant turns with im_end128039. Complete raw tool-call assistant content and the fixed inert tool result remain in history. Exact native template selection, complete token IDs, decode identity and integer usage are checked; no external tool executes.
Runtime prerequisites:
Ubuntu 24.04 x86_64 host; pinned Ubuntu 22.04 / Python 3.12 defaults container; NVIDIA Linux driver >=580.126.20, Working Docker GPU support with NVIDIA Container Toolkit; actual visibility preflight required.

Official artifact · License

Publisher card metadata declares llama3 while naming the Llama 3.1 405B base. The native LICENSE probe returned scoped404; the preserved USE_POLICY evidence remains separately bound; this does not erase its explicit metadata-linked grant. Full official Meta Llama 3 and stated-base Llama 3.1 agreements/AUP are archived separately, with no equivalence or publisher-intent claim. Retain both applicable agreement copies, notices and attribution (Built with Meta Llama 3 / Built with Llama); comply with redistribution and derivative naming requirements and both AUPs. Each agreement has its own release-date 700-million-MAU prior-month commercial license trigger. Preserve Llama 3 restrictions on using materials/outputs to improve other large language models, subject to its stated derivative exception. Patent/IP litigation, termination, warranty and liability conditions apply. This inert own-use walkthrough provides no legal clearance and accepts no terms on the reader behalf.

  • 811,706,916,800 serialized weight bytes are a full-model storage lower bound, not total memory or peak. Tensor-index metadata and file sizes are separate accounting.
  • Eight advertised 141 GB H200 devices are not silently converted into GiB. Usable per-device bytes and runtime partition balance are unknown; no aggregate-fit claim.
  • Native Llama has 126 dense layers, width 16,384, 128 query/eight KV heads and no MoE active-parameter shortcut. Host staging, Docker image and tokenizer add overhead.

One-time setup and hardware

hardware
Unknown — Hardware acquisition is part of instance rental; separate purchase cost not applicable.
Hardware acquisition is part of instance rental; separate purchase cost not applicable. Tax: unknown.
setup
Unknown — Provisioning/imagebuild/labor cost unknown.
Provisioning/imagebuild/labor cost unknown. Tax: unknown.

Recurring costs

compute
$63.30 / hour
Exact dated us-east-1 Linux On-Demand p5en.48xlarge quote USD 63.296 per instance-hour. Eight hours compute USD 506.368; no purchase, reservation, discount or realized spend assumed. Running idle hours bill. Storage, transfer, tax and retained resources separate. Region: us-east-1. SKU: p5en.48xlarge Linux On-Demand. Tax: excluded. Price dated Oct 15, 2026. Price source
storage
$0.08 / GB month
Region-selected us-east-1 gp3 source captured Oct16: USD0.08 per binary GB-month, baseline3000 IOPS/125 MB/s. 2048 GiB costs USD163.84 for a full month; actual retained duration, snapshots, IP, transfer and tax are separate/unknown. Region: us-east-1. SKU: EBS gp3 baseline. Tax: excluded. Price dated Oct 16, 2026. Price source
egress
Unknown — Actual transfer, publicIPv4/IP and associated service charges depend on usage/resources; unknown.
Actual transfer, publicIPv4/IP and associated service charges depend on usage/resources; unknown. Tax: unknown.
administration
Unknown — Maintenance time/rate unknown.
Maintenance time/rate unknown. Tax: unknown.

Install the runtime

  1. 1. Prepare an already owned eight-H200 instance

    Use an existing Ubuntu 24.04 x86_64 p5en.48xlarge with eight H200 devices and working driver >=580.126.20. These commands install host utilities only; provisioning, quota, price and all spending are reader responsibilities.

    Command
    set -eu
sudo apt-get update
sudo apt-get install -y docker.io python3 curl ca-certificates openssl
sudo systemctl enable --now docker
mkdir -p "$HOME/hermes3-405b"
cd "$HOME/hermes3-405b"
nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv,noheader
python3 - <<'PYPREFLIGHT'
import csv,subprocess,shutil
r=list(csv.reader(subprocess.check_output(["nvidia-smi","--query-gpu=name,memory.total,driver_version","--format=csv,noheader,nounits"],text=True).splitlines()))
assert len(r)==8 and all("H200" in x[0] for x in r)
assert all(tuple(map(int,x[2].strip().split("."))) >= (580,126,20) for x in r)
assert shutil.disk_usage(".").free >=1800*1024**3
print("Eight H200 devices and prescribed disk/driver checks passed; no model fit measured.")
PYPREFLIGHT
sudo docker version

    Expected result: Record eight actual GPU identities, usable memory and drivers. Stop on missing devices, driver/GPU Docker failure or inadequate prescribed disk. No native fit claim.

  2. 2. Configure NVIDIA GPU containers on this dedicated host

    Use the official production repository and exact documented Container Toolkit 1.20.1-1 packages. This modifies Docker daemon configuration and restarts Docker on this dedicated host; preserve its original configuration and do not apply on a shared production daemon without its owner review. Stop on repository/signature/package errors.

    Command
    set -euo pipefail
sudo apt-get install -y gnupg2
curl --fail --silent --show-error --location https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --batch --yes --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
curl --fail --silent --show-error --location https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list >/dev/null
sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit=1.20.1-1 nvidia-container-toolkit-base=1.20.1-1 libnvidia-container-tools=1.20.1-1 libnvidia-container1=1.20.1-1
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

    Expected result: The exact GPU container packages install and Docker is configured. The subsequent container nvidia-smi check must pass; this is not model inference or fit proof.

Download the model

  1. 1. Write native reader source bundle segment 1

    This is reversible compressed UTF-8 source, not model weights. Decode and inspect all exact reader files before use.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat > reader-bundle.b64 <<'B64'
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
B64

    Expected result: Source segment written; no model or native code executed.

  2. 2. Write native reader source bundle segment 2

    This is reversible compressed UTF-8 source, not model weights. Decode and inspect all exact reader files before use.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat >> reader-bundle.b64 <<'B64'
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
B64

    Expected result: Source segment written; no model or native code executed.

  3. 3. Write native reader source bundle segment 3

    This is reversible compressed UTF-8 source, not model weights. Decode and inspect all exact reader files before use.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat >> reader-bundle.b64 <<'B64'
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
B64

    Expected result: Source segment written; no model or native code executed.

  4. 4. Decode and verify all native assets

    Reader downloads full pinned native repository files only after inspecting the source bundle. Stop on any size/digest/path mismatch; never bypass. This does not accept terms on behalf of the reader.

    Command
    set -eu
cd "$HOME/hermes3-405b"
python3 - <<'PYBUNDLE'
import base64,zlib,json,hashlib,pathlib
b=zlib.decompress(base64.b64decode(pathlib.Path('reader-bundle.b64').read_text()))
assert len(b)<262144 and hashlib.sha256(b).hexdigest()=='ffa8bfa697d5c8ae74c373527a2a8a8cf57a90032f4e1a5da3e8fe53ba5c2c2d'
for n,r in json.loads(b).items():
 assert pathlib.PurePosixPath(n).name==n
 p=pathlib.Path(n);assert not p.is_symlink()
 raw=r['text'].encode();assert hashlib.sha256(raw).hexdigest()==r['sha256']
 if p.exists():assert p.read_bytes()==raw
 else:p.write_bytes(raw)
PYBUNDLE
python3 reader-download.v2.py

    Expected result: Complete regular-file asset checks pass, including all 191 shards, before native tokenizer loading.

Configure

  1. 1. Bind the existing private reader key

    Use an existing authorized ASCII Bearer key; no new account credentials are created. Keep the env file private; do not publish logs or process metadata containing secrets.

    Command
    set -eu
cd "$HOME/hermes3-405b"
test -n "${HERMES3_READER_API_KEY:-}"
python3 - <<'PYKEY'
import os,pathlib
k=os.environ['HERMES3_READER_API_KEY'];assert 32<=len(k)<=256 and all(33<=ord(c)<=126 for c in k)
p=pathlib.Path('.api-env');assert not p.is_symlink()
fd=os.open(p,os.O_WRONLY|os.O_CREAT|os.O_TRUNC,0o600)
os.fchmod(fd,0o600)
with os.fdopen(fd,'w')as f:f.write('VLLM_API_KEY='+k+'\nHERMES3_READER_API_KEY='+k+'\nHF_HUB_OFFLINE=1\nTRANSFORMERS_OFFLINE=1\n')
PYKEY

    Expected result: Private0600env contains the same existing server/client key and explicit offline environment.

  2. 2. Build the pinned native runtime layer

    Official vLLM amd64 image is immutable; the same source-qualified exact Transformers replacement used in prior native guides is explicit. This image is unbuilt here. Stop on pip check, version, CUDA, tensor-parallel or loader errors; record realized versions.

    Command
    set -eu
cd "$HOME/hermes3-405b"
cat > Dockerfile <<'DOCKERFILE'
FROM vllm/vllm-openai@sha256:c2f3b1b964e47809b722b5e75b61b1e7b39a50f70388cf2bf2418f16a9f31da2
RUN python3 -m pip install --no-deps 'https://files.pythonhosted.org/packages/fc/b1/8be7e7ef0b5200491312201918b6125ef9c9df9dd0f0240ccef9ac824e6b/transformers-5.8.1-py3-none-any.whl#sha256=5340fb95962162cdfdae5cc91d7f8fedd92ed75216c1154c5e1f590fcf56dd0e' && python3 -m pip check
RUN python3 -c "import vllm,transformers; assert vllm.__version__=='0.27.1'; assert transformers.__version__=='5.8.1'"
ENV HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1
ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]
DOCKERFILE
sudo docker build --pull -t hermes3-405b:reader .
sudo docker run --rm --gpus all --entrypoint nvidia-smi hermes3-405b:reader
sudo docker run --rm --entrypoint python3 hermes3-405b:reader -m pip freeze > resolved-container-environment.txt
sudo docker network inspect hermes3-private >/dev/null 2>&1 || sudo docker network create --internal hermes3-private

    Expected result: Only after actual reader execution: package/version guards and all eight GPU visibility pass; no model load implied.

Start the model

  1. 1. Start complete native BF16 with tensor parallel eight

    Foreground lifetime is 1,800 seconds including loading; it is not an inference deadline/SLA. Native full loader must account for all model parameters; do not bypass diagnostics, precision or architecture failure.

    Command
    set -eu
cd "$HOME/hermes3-405b"
python3 -c 'import importlib.util,json,pathlib; s=importlib.util.spec_from_file_location("r","reader-client.v3.py");m=importlib.util.module_from_spec(s);s.loader.exec_module(m);m.verify_assets(pathlib.Path("model"),json.loads(pathlib.Path("native-assets.v2.json").read_text()))'
timeout --signal=TERM --kill-after=10s 1800s sudo docker run --rm --name hermes3-405b --gpus all --network hermes3-private -p 127.0.0.1:8994:8000 --shm-size 64g --env-file .api-env -v "$PWD/model:/model:ro" -v "$PWD:/work" hermes3-405b:reader --model /model --served-model-name hermes3-405b-native --host 0.0.0.0 --port 8000 --dtype bfloat16 --kv-cache-dtype auto --tensor-parallel-size 8 --max-model-len 8192 --max-num-seqs 1 --max-num-batched-tokens 8192 --gpu-memory-utilization 0.95 --no-enable-prefix-caching --enforce-eager

    Expected result: Only after actual reader initialization: complete native model starts without missing-parameter/collective/CUDA errors and private health becomes ready. Unknown fit may prevent startup.

Connect to your endpoint

http://127.0.0.1:8994/v1/completions

VM loopback only; shown checks run on the server. Optional existing SSH tunnel: ssh -N -L 8994:127.0.0.1:8994 USER@VM_IP. Keep public service port 8994 closed.

Access: loopback. Stable private Bearer key protects native v1 routes. Native health/tokenize are unprotected and must remain private. The model server is trusted single-user research, not a hardened public endpoint; max-num-seqs 1 is scheduling, not admission or a bounded user queue. Only the fixed client is used.

Verify the setup

  1. 1. Verify native ChatML conversation and full inert tool history

    From a second host terminal check loopback access, unauthenticated v1 rejection and then run the exact native local tokenizer/client inside the serving container. All tool outputs are fixed synthetic data, not an executor.

    Command
    set -eu
cd "$HOME/hermes3-405b"
curl --fail --max-time 5 http://127.0.0.1:8994/health
test "$(curl -s --max-time 5 -o /dev/null -w '%{http_code}' http://127.0.0.1:8994/v1/models)" = 401
sudo docker exec hermes3-405b python3 /work/reader-client.v3.py

    Expected result: Successful client exit and populated parseable result.json show all three strict protocol turns passed. Manually inspect the actual call, synthetic seven-plus-one answer and independent default-template response; no accuracy or throughput guarantee.

Performance evidence

No compatible measurement available. Documented support and estimated memory do not establish response speed.

Limits and troubleshooting

  • No install, build, tokenizer execution, weight download, native inference, provisioning or spending occurred. Source compatibility is not measured runtime success.
  • Native config has 131,072 positions. Reduced walkthrough budgets do not establish full native window fit, quality or speed. Input count includes one BOS, tools and full retained native history.
  • Native card llama3 metadata and Llama 3.1 base identify distinct agreements; both actual term/AUP sets retained. No later-base license silently replaces the linked license or grants legal clearance.
  • Eight H200 per-device usable capacity, actual TP balance, kernel/collective buffers and peak are unknown. All weights remain; no smaller model or offload workaround is silently substituted.
  • Client 600-second socket timeout does not cancel GPU work. Foreground 1,800-second lifetime includes initialization; stop/kill the named experiment and cloud billing separately.
  • Docker workspace mount contains the private key; the container and Docker administrator are trusted. No arbitrary tools, uploads, URL fetch or remote model choices are provided.
  • Hermes 3 uses ChatML EOS 128039 and im_start128040; default and tool_use templates are independently selected from the exact native tokenizer. No Hermes4 thinking/control IDs apply.
  • The full serialized weight size is a storage/accounting lower bound, not a GPU or host minimum or a proof of impossibility on other allocations.
  • A result is successful only after the client exits successfully with populated parseable results. Human semantic correctness is a separate reader check.
  • Frozen free/provider identity does not prove API availability or serving-byte equivalence. All useful backends and family workloads remain open.
Asset or native initialization failure
Stop on any digest, missing file, unexpected loader key, tensor-parallel, CUDA, version or tokenizer mismatch. Recheck pinned original sources and device inventory; do not delete components or substitute quantization. No errors are treated as a healthy run.
Protocol or semantic failure
Do not strip malformed output or treat length finish as completion. Preserve actual raw response privately, stop the job and inspect native EOS/template/history and all token budgets. A valid EOS at the4096cap is accepted only with correct raw IDs, decode and usage. Reader records actual errors; protocol pass does not establish semantic quality.

Stop the model

  1. 1. Stop the named native experiment

    Use a second server terminal; foreground timeout is not proof the container exited. Kill only this dedicated experiment if it remains running.

    Command
    sudo docker stop --time 10 hermes3-405b || true
if [ "$(sudo docker inspect -f '{{.State.Running}}' hermes3-405b 2>/dev/null || true)" = true ]; then sudo docker kill hermes3-405b; fi
curl --max-time 3 http://127.0.0.1:8994/health || true

    Expected result: The exact container and endpoint stop; instance/storage billing continues until separately managed.

Uninstall

  1. 1. Remove dedicated checkpoint, runtime tag and private files

    Preserve wanted results and legal/source records first. Confirm dedicated paths. Shared drivers/Docker and retained image layers remain; no broad prune.

    Command
    sudo docker rm -f hermes3-405b 2>/dev/null || true
sudo docker network rm hermes3-private
sudo docker image rm hermes3-405b:reader
rm -r -- "$HOME/hermes3-405b"

    Expected result: Dedicated files, results/key and image tag/network removed; shared packages/layers remain.

Shut down cloud resources

  1. 1. End instance and retained-resource billing

    In EC2 us-east-1 stop for reuse or terminate this exact instance. Review Delete on termination for the 2 TiB root volume; delete unwanted detached EBS volumes, snapshots and retained public/Elastic IP resources. Included local NVMe is ephemeral. Check Bills/Cost Explorer for actual final usage.

    Expected result: Compute ends only after the actual instance lifecycle transition. Retained volumes/snapshots/IPs may still bill; stopping the model container is insufficient.

Sources and verification

Sources checked Oct 17, 2026. Verification source

Compare this configuration

Benchmarks

Performance scores for Hermes 3 405B Instruct across standard benchmarks.

MMLUOpenRouter · Jul 2024
87.3
HumanEvalOpenRouter · Jul 2024
84.9

Pricing and access

Free tier available on OpenRouter with rate limits

Nous Research logo
Nous Researchai lab
4 Models
View full profile
4 Models

AI Models by Nous Research

AI models from the same organization.

ModelContext WindowPrice (In / Out per M)
Hermes 4 70BCurrent131K$0.13 / $0.40
Hermes 4 405BCurrent131K$1.00 / $3.00
Hermes 3 70B InstructCurrent131K$0.30 / $0.30