mirror of
https://github.com/katanemo/plano.git
synced 2026-04-25 00:36:34 +02:00
4182879717
* add precommit check
* remove check
* Revert "remove check"
This reverts commit 9987b62b9b.
* fix checks
* fix whitespace errors
159 lines
4.5 KiB
ReStructuredText
159 lines
4.5 KiB
ReStructuredText
.. _llms_in_arch:
|
||
|
||
LLMs
|
||
====
|
||
|
||
Arch utilizes purpose-built, industry leading, LLMs to handle the crufty and undifferentiated work around
|
||
accepting, handling and processing prompts. The following sections talk about some of the core models that
|
||
are built-in Arch.
|
||
|
||
Arch-Guard-v1
|
||
-------------
|
||
LLM-powered applications are susceptible to prompt attacks, which are prompts intentionally designed to
|
||
subvert the developer’s intended behavior of the LLM. Arch-Guard-v1 is a classifier model trained on a large
|
||
corpus of attacks, capable of detecting explicitly malicious prompts (and toxicity).
|
||
|
||
The model is useful as a starting point for identifying and guardrailing against the most risky realistic
|
||
inputs to LLM-powered applications. Our goal in embedding Arch-Guard in the Arch gateway is to enable developers
|
||
to focus on their business logic and factor out security and safety outside application logic. Wth Arch-Guard-v1
|
||
developers can take to significantly reduce prompt attack risk while maintaining control over the user experience.
|
||
|
||
Below is our test results of the strength of our model as compared to Prompt-Guard from `Meta LLama <https://huggingface.co/meta-llama/Prompt-Guard-86M>`_.
|
||
|
||
.. list-table::
|
||
:header-rows: 1
|
||
:widths: 15 15 10 15 15
|
||
|
||
* - Dataset
|
||
- Jailbreak (Yes/No)
|
||
- Samples
|
||
- Prompt-Guard Accuracy
|
||
- Arch-Guard Accuracy
|
||
* - casual_conversation
|
||
- 0
|
||
- 3725
|
||
- 1.00
|
||
- 1.00
|
||
* - commonqa
|
||
- 0
|
||
- 9741
|
||
- 1.00
|
||
- 1.00
|
||
* - financeqa
|
||
- 0
|
||
- 1585
|
||
- 1.00
|
||
- 1.00
|
||
* - instruction
|
||
- 0
|
||
- 5000
|
||
- 1.00
|
||
- 1.00
|
||
* - jailbreak_behavior_benign
|
||
- 0
|
||
- 100
|
||
- 0.10
|
||
- 0.20
|
||
* - jailbreak_behavior_harmful
|
||
- 1
|
||
- 100
|
||
- 0.30
|
||
- 0.52
|
||
* - jailbreak_judge
|
||
- 1
|
||
- 300
|
||
- 0.33
|
||
- 0.49
|
||
* - jailbreak_prompts
|
||
- 1
|
||
- 79
|
||
- 0.99
|
||
- 1.00
|
||
* - jailbreak_tweet
|
||
- 1
|
||
- 1282
|
||
- 0.16
|
||
- 0.35
|
||
* - jailbreak_v
|
||
- 1
|
||
- 20000
|
||
- 0.90
|
||
- 0.93
|
||
* - jailbreak_vigil
|
||
- 1
|
||
- 104
|
||
- 1.00
|
||
- 1.00
|
||
* - mental_health
|
||
- 0
|
||
- 3512
|
||
- 1.00
|
||
- 1.00
|
||
* - telecom
|
||
- 0
|
||
- 4000
|
||
- 1.00
|
||
- 1.00
|
||
* - truthqa
|
||
- 0
|
||
- 817
|
||
- 1.00
|
||
- 0.98
|
||
* - weather
|
||
- 0
|
||
- 3121
|
||
- 1.00
|
||
- 1.00
|
||
|
||
.. list-table::
|
||
:header-rows: 1
|
||
:widths: 15 20
|
||
|
||
* - Statistics
|
||
- Overall performance
|
||
* - Overall Accuracy
|
||
- 0.93568 (Prompt-Guard), 0.95267 (Arch-Guard)
|
||
* - True positives rate (TPR)
|
||
- 0.8468 (Prompt-Guard), 0.8887 (Arch-Guard)
|
||
* - True negative rate (TNR)
|
||
- 0.9972 (Prompt-Guard), 0.9970 (Arch-Guard)
|
||
* - False positive rate (FPR)
|
||
- 0.0028 (Prompt-Guard), 0.0030 (Arch-Guard)
|
||
* - False negative rate (FNR)
|
||
- 0.1532 (Prompt-Guard), 0.1113 (Arch-Guard)
|
||
|
||
.. list-table::
|
||
:header-rows: 1
|
||
:widths: 15 20
|
||
|
||
* - Metrics
|
||
- Values
|
||
* - AUC
|
||
- 0.857 (Prompt-Guard), 0.880 (Arch-Guard)
|
||
* - Precision
|
||
- 0.715 (Prompt-Guard), 0.761 (Arch-Guard)
|
||
* - Recall
|
||
- 0.999 (Prompt-Guard), 0.999 (Arch-Guard)
|
||
|
||
|
||
|
||
Arch-FC
|
||
-------
|
||
Arch-FC is a lean, powerful and cost-effective agentic model designed for function calling scenarios.
|
||
You can run Arch-FC locally, or use the cloud-hosted version for as little as $0.05/M token (100x cheaper
|
||
than GPT-4o), with a p50 latency of 200ms (5x faster than GPT-4o), while meeting frontier model performance.
|
||
|
||
.. Note::
|
||
Function calling helps you personalize the GenAI experience by calling application-specific operations via
|
||
prompts. This involves any predefined functions or APIs you want to expose to perform tasks, gather
|
||
information, or manipulate data - via prompts.
|
||
|
||
You can get started with function calling simply by configuring a prompt target with a name, description
|
||
and set of parameters needed by a specific backend function or a hosted API. The name, and description helps
|
||
Arch-FC match a user prompt to a function or API that can process it.
|
||
|
||
By using Arch-FC, Arch enables you to easily build agentic workflows tailored to domain-specific use cases -
|
||
from updating insurance claims to creating ad campaigns. Arch-FC analyzes prompts, extracts critical information
|
||
from prompts, engages in lightweight conversations with the user to gather any missing parameters need before
|
||
handling control back to Arch to make the API call to your hosted backend. Arch-FC handles the muck of information
|
||
extraction so that you can focus on the business logic of your application.
|