Repository navigation
Conversation
|
|
Greptile SummaryThis PR adds Wallaby as a declarative OpenAI-compatible provider and registers pricing, limits, and capabilities for
Confidence Score: 3/5The PR is not safe to merge until Wallaby supports advertised tool calls and the bundled cost map is synchronized Wallaby tool calls are rejected before reaching the provider, while CLI and offline deployments omit its pricing because the bundled cost map was not updated Files Needing Attention: litellm/llms/openai_like/providers.json, model_prices_and_context_window.json
|
| Filename | Overview |
|---|---|
| litellm/llms/openai_like/providers.json | Adds Wallaby configuration, but omits compatibility-list registration needed for tool calls and lacks required focused tests |
| model_prices_and_context_window.json | Adds Wallaby Kimi K3 metadata without synchronizing the bundled backup cost map |
Reviews (1): Last reviewed commit: "Add wallaby provider and kimi-k3 pricing" | Re-trigger Greptile
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| "wallaby": { | ||
| "base_url": "https://api.wallabytoken.com/v1", | ||
| "api_key_env": "WALLABY_API_KEY", | ||
| "api_base_env": "WALLABY_API_BASE" |
There was a problem hiding this comment.
Wallaby is absent from openai_compatible_providers, so requests with tools fail before reaching the provider despite advertised function-calling support
Knowledge Base Used: Provider adapters and capabilities
| "wallaby/kimi-k3": { | ||
| "cache_read_input_token_cost": 2.7e-07, | ||
| "input_cost_per_token": 2.7e-06, | ||
| "litellm_provider": "wallaby", | ||
| "max_input_tokens": 1048576, | ||
| "max_output_tokens": 131072, | ||
| "max_tokens": 131072, | ||
| "mode": "chat", | ||
| "output_cost_per_token": 1.35e-05, | ||
| "source": "https://wallabytoken.com/pricing.json", | ||
| "supports_function_calling": true, | ||
| "supports_prompt_caching": true, | ||
| "supports_reasoning": true, | ||
| "supports_response_schema": true, | ||
| "supports_tool_choice": true | ||
| } |
There was a problem hiding this comment.
The backup cost map lacks Wallaby, so CLI and offline loads omit its pricing, and the cost-map guard fails
Knowledge Base Used: Provider adapters and capabilities
| "wallaby": { | ||
| "base_url": "https://api.wallabytoken.com/v1", | ||
| "api_key_env": "WALLABY_API_KEY", | ||
| "api_base_env": "WALLABY_API_BASE" |
There was a problem hiding this comment.
This new provider has no focused registry or routing test, violating the repository requirement for meaningful new-feature tests
Context Used: CLAUDE.md (source)
Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
|
Thanks for the review! Both P1s addressed:
Re: tests — this is a declarative JSON-configured provider following the same |
Dawncoral
changed the title
- provider registry, pricing (root + backup cost maps, byte-synced) - openai_compatible_endpoints base URL for bare-URL autodetect - LlmProviders.WALLABY enum, dashboard provider_create_fields entry - provider_endpoints_support registration (root + backup) - tests: tests/unit/llms/openai_like/test_wallaby_provider.py (9 tests)
2d0fac0 to
c53b3c9
Compare
|
Synced to latest main (resolves the merge conflicts) and added the focused
Ready for review whenever you have a moment. Thanks for maintaining litellm! |
sync: reapply wallaby cost map + endpoints on latest main
sync: reapply wallaby constants/enum/backups on latest main
sync: reapply wallaby providers.json on latest main
sync: reapply wallaby provider create fields on latest main
sync: reapply wallaby enum on latest main
chore: remove misplaced utils.py
sync: reapply wallaby enum on latest main
fix: restore litellm/utils.py to match main
chore: remove misnamed _utils.py
fix: restore litellm/utils.py to match main
|
Closing in favor of #43836 — same content, rebased onto latest main as a clean branch. The recurring cost-map conflicts couldn't be cleared through web uploads (they can't create merge commits). All review feedback from this thread is incorporated in the new PR. Thanks! |
3 participants
What this adds
Wallaby as an OpenAI-like provider, with pricing for
wallaby/kimi-k3.https://api.wallabytoken.com/v1Changes
litellm/llms/openai_like/providers.json— addwallabyentry (base_url +WALLABY_API_KEY/WALLABY_API_BASEenv vars)model_prices_and_context_window.json— addwallaby/kimi-k3($2.70 / $13.50 / $0.27 cache-read per 1M tokens; 1M context; supports reasoning, function calling, response schema, prompt caching)Verification
Tested locally against the live endpoint with the providers.json entry applied:
completion(model="wallaby/kimi-k3", ...)non-streaming ✅reasoning_contentthencontent)temperature=0.7, top_p=0.9from the client ✅ (the serving stack fixes sampling params; the gateway normalizes them instead of erroring)response_format={"type": "json_object"}✅Happy to provide a test API key for maintainer verification — just say the word.