[Feat] Add Isaacus embeddings provider

[adlumal]

Title

Add Isaacus embeddings provider with kanon-2-embedder support

Relevant issues

N/A - New provider integration

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
  • Added comprehensive tests in tests/llm_translation/test_isaacus_embedding.py with 14 unit tests and 6 e2e tests
• I have added a screenshot of my new test passing locally

• My PR passes all unit tests on make test-unit - Isaacus-specific and related tests pass, there are 3 unrelated tests that do not pass at the time of this PR on my machine
• My PR's scope is as isolated as possible, it only solves 1 specific problem - Only adds Isaacus embeddings provider, no other modifications

Type

🆕 New Feature

Changes

This PR adds support for the Isaacus embeddings provider (kanon-2-embedder model) to LiteLLM.

Implementation:

• Created litellm/llms/isaacus/embedding/transformation.py with IsaacusEmbeddingConfig class
• Implements request transformation: OpenAI input parameter → Isaacus texts parameter
• Implements response transformation: Isaacus response format → OpenAI-compatible format
• Registered provider in core files: __init__.py, constants.py, main.py, types/utils.py, utils.py, get_llm_provider_logic.py

Features:

• Standard OpenAI embeddings syntax compatibility (drop-in replacement)
• Support for single text and batch embedding requests
• Support for async embedding via aembedding()
• Isaacus-specific optional parameters:
  • task - Optimize for "retrieval/query" or "retrieval/document"
  • dimensions - Optionally reduce embedding dimensionality
  • overflow_strategy - Handle text overflow with "drop_end"

Testing:

• 14 unit tests with mocked responses covering:
  • Different input formats (string, list)
  • Optional parameters handling
  • Request/response transformation
• 6 end-to-end tests covering:
  • Basic embedding
  • Batch embedding
  • Optional parameters
  • Async embedding

Documentation:

• Added Isaacus section to docs/my-website/docs/embedding/supported_embedding.md
• Included usage examples and parameter descriptions

Files changed:

• litellm/llms/isaacus/__init__.py (new)
• litellm/llms/isaacus/embedding/__init__.py (new)
• litellm/llms/isaacus/embedding/transformation.py (new)
• litellm/__init__.py (added isaacus_key, imported IsaacusEmbeddingConfig)
• litellm/constants.py (registered isaacus provider)
• litellm/main.py (added isaacus routing logic)
• litellm/types/utils.py (added ISAACUS to LlmProviders enum)
• litellm/utils.py (registered IsaacusEmbeddingConfig in ProviderConfigManager)
• litellm/litellm_core_utils/get_llm_provider_logic.py (added isaacus provider detection)
• tests/llm_translation/test_isaacus_embedding.py (new - 14 unit tests, 6 e2e tests requiring API key, skipped)
• tests/test_litellm/llms/isaacus/embedding/__init__.py (new)
• tests/test_litellm/llms/isaacus/embedding/test_transformation.py (new)
• docs/my-website/docs/embedding/supported_embedding.md (added Isaacus documentation)

Example usage:

import litellm

response = litellm.embedding(
    model="isaacus/kanon-2-embedder",
    input="Legal text to embed",
    api_key="your-isaacus-api-key"
)

This is my first time contributing to LiteLLM so please let me know of any errors or mistakes.

[vercel]

@adlumal is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[regismesquita]

Looking forward to have this merged :)