Skip to main content

A simple llm router

Project description

GitHub stars License: MIT PyPI version Python 3.11+

llamux 🦙

A simple router to rotate across your configured LLM endpoints, balancing load and avoiding rate limits. The router selects your preferred provider-model pair based on an implicit preference list, ensuring token and request limits (day/hour/minute) aren't crossed. State persists across sessions, with quotas stored in local cache.

Install

Requires Python 3.11+

pip install llamux

Usage

You need first to set the list of endpoints you want to allow routing on;

$ > endpoints.csv

| provider | model                   | rpm | tpm   | rph   | tph     | rpd   | tpd     |
| -------- | ----------------------- | --- | ----- | ----- | ------- | ----- | ------- |
| cerebras | llama3.3-70b            | 30  | 60000 | 900   | 1000000 | 14400 | 1000000 |
| groq     | llama-3.3-70b-versatile | 30  | 6000  | 14400 |         | 14400 |         |

where rpm, tpm are requests and tokens limits per minutes, and the same follows for hours and days. Important note 🔊 Your implicit preference list is given by the ordering of the endpoints in this table. In the above, we'll always prefer Cerebras over Groq, as long as the quota limits are not exceeded.

Use it as a standalone router;

from llamux import Router

os.environ["CEREBRAS_API_KEY"] = "sk-..."
os.environ["GROQ_API_KEY"] = "sk-..."

router = Router.from_csv("endpoints.csv")
messages = [{"role": "user", "content": "Hello, how are you?"}]

provider, model, id, props = router.query(messages)
# provider: cerebras, model: llama3.3-70b

Or use it directly as a completion endpoint;

from llamux import Router

router = Router.from_csv("endpoints.csv")
messages = [{"role": "user", "content": "hey" * 59999}]

response = router.completion(messages) # calls cerebras
response = router.completion(messages) # calls groq (cerebras quota is out!)

The above builds upon litellm llm proxy

More features

Contributions are welcome :)

  • Add support for speed and cost routing preferences
  • Add other routing strategies (now preferencial ordering only)
  • Avoid getting preference listing from table ordering

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

llamux-0.1.5.tar.gz (4.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

llamux-0.1.5-py3-none-any.whl (5.3 kB view details)

Uploaded Python 3

File details

Details for the file llamux-0.1.5.tar.gz.

File metadata

  • Download URL: llamux-0.1.5.tar.gz
  • Upload date:
  • Size: 4.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/1.8.4 CPython/3.13.0 Darwin/22.6.0

File hashes

Hashes for llamux-0.1.5.tar.gz
Algorithm Hash digest
SHA256 240c3a3951bb1e376a6c17d327928c06bfe776cdac53cd95471548f6435d5b62
MD5 64babbf1e317857d5661aaf42c5a1e12
BLAKE2b-256 b11d7523fb015d2dc515beb4fa949f3c4b1e3220de19b4cd1f855ff02a303909

See more details on using hashes here.

File details

Details for the file llamux-0.1.5-py3-none-any.whl.

File metadata

  • Download URL: llamux-0.1.5-py3-none-any.whl
  • Upload date:
  • Size: 5.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/1.8.4 CPython/3.13.0 Darwin/22.6.0

File hashes

Hashes for llamux-0.1.5-py3-none-any.whl
Algorithm Hash digest
SHA256 02d5cc552209389a0ae2310caee864db04b901991efa6216c7f193732a3bd84c
MD5 a4879b1ea969ba08c33f6e8d9b838aad
BLAKE2b-256 9dd58f4259108a5e2da0048f4423fa1265c8944cf2fec6e945c251d77b4a82a9

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page