@homeflare/distilled-litellm 0.2.0 → 0.4.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +48 -12
- package/dist/services/a2a.js +1 -1
- package/dist/services/a2a_registration.js +1 -1
- package/dist/services/access_groups.d.ts +18 -0
- package/dist/services/access_groups.d.ts.map +1 -1
- package/dist/services/access_groups.js +15 -1
- package/dist/services/access_groups.js.map +1 -1
- package/dist/services/adaptive_router.js +1 -1
- package/dist/services/agents.d.ts +10 -0
- package/dist/services/agents.d.ts.map +1 -1
- package/dist/services/agents.js +10 -1
- package/dist/services/agents.js.map +1 -1
- package/dist/services/alerting.js +1 -1
- package/dist/services/anthropic_passthrough.js +1 -1
- package/dist/services/anthropic_skills.d.ts +7 -1
- package/dist/services/anthropic_skills.d.ts.map +1 -1
- package/dist/services/anthropic_skills.js +6 -2
- package/dist/services/anthropic_skills.js.map +1 -1
- package/dist/services/assistants.js +1 -1
- package/dist/services/audio.js +1 -1
- package/dist/services/audit_logging.d.ts +2 -0
- package/dist/services/audit_logging.d.ts.map +1 -1
- package/dist/services/audit_logging.js +2 -1
- package/dist/services/audit_logging.js.map +1 -1
- package/dist/services/auto_router.d.ts +162 -62
- package/dist/services/auto_router.d.ts.map +1 -1
- package/dist/services/auto_router.js +92 -31
- package/dist/services/auto_router.js.map +1 -1
- package/dist/services/batch.js +1 -1
- package/dist/services/beta_agents.js +1 -1
- package/dist/services/beta_mcp.js +1 -1
- package/dist/services/budget_management.d.ts +8 -3
- package/dist/services/budget_management.d.ts.map +1 -1
- package/dist/services/budget_management.js +7 -4
- package/dist/services/budget_management.js.map +1 -1
- package/dist/services/budget_spend_tracking.d.ts +24 -5
- package/dist/services/budget_spend_tracking.d.ts.map +1 -1
- package/dist/services/budget_spend_tracking.js +17 -4
- package/dist/services/budget_spend_tracking.js.map +1 -1
- package/dist/services/cache_settings.js +1 -1
- package/dist/services/caching.js +1 -1
- package/dist/services/chat_completions.js +1 -1
- package/dist/services/claude_code_marketplace.d.ts +11 -10
- package/dist/services/claude_code_marketplace.d.ts.map +1 -1
- package/dist/services/claude_code_marketplace.js +8 -6
- package/dist/services/claude_code_marketplace.js.map +1 -1
- package/dist/services/cloudzero.js +1 -1
- package/dist/services/completions.js +1 -1
- package/dist/services/compliance.js +1 -1
- package/dist/services/config_overrides.d.ts +5 -1
- package/dist/services/config_overrides.d.ts.map +1 -1
- package/dist/services/config_overrides.js +3 -1
- package/dist/services/config_overrides.js.map +1 -1
- package/dist/services/config_yaml.d.ts +120 -6
- package/dist/services/config_yaml.d.ts.map +1 -1
- package/dist/services/config_yaml.js +67 -1
- package/dist/services/config_yaml.js.map +1 -1
- package/dist/services/containers.d.ts +8 -2
- package/dist/services/containers.d.ts.map +1 -1
- package/dist/services/containers.js +13 -5
- package/dist/services/containers.js.map +1 -1
- package/dist/services/coordination_redis_settings.js +1 -1
- package/dist/services/cost_tracking.d.ts +91 -1
- package/dist/services/cost_tracking.d.ts.map +1 -1
- package/dist/services/cost_tracking.js +82 -2
- package/dist/services/cost_tracking.js.map +1 -1
- package/dist/services/credential_management.d.ts +16 -3
- package/dist/services/credential_management.d.ts.map +1 -1
- package/dist/services/credential_management.js +13 -4
- package/dist/services/credential_management.js.map +1 -1
- package/dist/services/customer_management.d.ts +25 -2
- package/dist/services/customer_management.d.ts.map +1 -1
- package/dist/services/customer_management.js +22 -3
- package/dist/services/customer_management.js.map +1 -1
- package/dist/services/email_management.js +1 -1
- package/dist/services/embeddings.js +1 -1
- package/dist/services/evals.js +1 -1
- package/dist/services/experimental.js +1 -1
- package/dist/services/fallback_management.js +1 -1
- package/dist/services/files.js +1 -1
- package/dist/services/fine_tuning.js +1 -1
- package/dist/services/gemini_agents.js +1 -1
- package/dist/services/google_genai_endpoints.js +1 -1
- package/dist/services/guardrails.d.ts +80 -7
- package/dist/services/guardrails.d.ts.map +1 -1
- package/dist/services/guardrails.js +37 -1
- package/dist/services/guardrails.js.map +1 -1
- package/dist/services/health.d.ts +2 -2
- package/dist/services/health.d.ts.map +1 -1
- package/dist/services/health.js +2 -2
- package/dist/services/health.js.map +1 -1
- package/dist/services/images.js +1 -1
- package/dist/services/index.d.ts +1 -15
- package/dist/services/index.d.ts.map +1 -1
- package/dist/services/index.js +1 -15
- package/dist/services/index.js.map +1 -1
- package/dist/services/internal_user_management.d.ts +227 -39
- package/dist/services/internal_user_management.d.ts.map +1 -1
- package/dist/services/internal_user_management.js +194 -37
- package/dist/services/internal_user_management.js.map +1 -1
- package/dist/services/invite_links.js +1 -1
- package/dist/services/jwt_mappings.d.ts +3 -0
- package/dist/services/jwt_mappings.d.ts.map +1 -1
- package/dist/services/jwt_mappings.js +4 -1
- package/dist/services/jwt_mappings.js.map +1 -1
- package/dist/services/key_management.d.ts +116 -50
- package/dist/services/key_management.d.ts.map +1 -1
- package/dist/services/key_management.js +95 -40
- package/dist/services/key_management.js.map +1 -1
- package/dist/services/langfuse_passthrough.js +1 -1
- package/dist/services/llm_passthrough.d.ts +1199 -0
- package/dist/services/llm_passthrough.d.ts.map +1 -0
- package/dist/services/llm_passthrough.js +2150 -0
- package/dist/services/llm_passthrough.js.map +1 -0
- package/dist/services/llm_utils.js +1 -1
- package/dist/services/logging_callbacks.js +1 -1
- package/dist/services/mcp_byok_oauth.d.ts +18 -11
- package/dist/services/mcp_byok_oauth.d.ts.map +1 -1
- package/dist/services/mcp_byok_oauth.js +27 -24
- package/dist/services/mcp_byok_oauth.js.map +1 -1
- package/dist/services/mcp_discoverable.d.ts +34 -0
- package/dist/services/mcp_discoverable.d.ts.map +1 -1
- package/dist/services/mcp_discoverable.js +51 -1
- package/dist/services/mcp_discoverable.js.map +1 -1
- package/dist/services/mcp_management.d.ts +90 -2
- package/dist/services/mcp_management.d.ts.map +1 -1
- package/dist/services/mcp_management.js +115 -3
- package/dist/services/mcp_management.js.map +1 -1
- package/dist/services/mcp_rest.d.ts +13 -0
- package/dist/services/mcp_rest.d.ts.map +1 -1
- package/dist/services/mcp_rest.js +10 -2
- package/dist/services/mcp_rest.js.map +1 -1
- package/dist/services/memory_management.d.ts +2 -0
- package/dist/services/memory_management.d.ts.map +1 -1
- package/dist/services/memory_management.js +2 -1
- package/dist/services/memory_management.js.map +1 -1
- package/dist/services/misc.d.ts +86 -1
- package/dist/services/misc.d.ts.map +1 -1
- package/dist/services/misc.js +126 -2
- package/dist/services/misc.js.map +1 -1
- package/dist/services/model_management.d.ts +310 -19
- package/dist/services/model_management.d.ts.map +1 -1
- package/dist/services/model_management.js +240 -7
- package/dist/services/model_management.js.map +1 -1
- package/dist/services/moderations.js +1 -1
- package/dist/services/ocr.js +1 -1
- package/dist/services/open_ai_pass_through.d.ts +35 -90
- package/dist/services/open_ai_pass_through.d.ts.map +1 -1
- package/dist/services/open_ai_pass_through.js +41 -139
- package/dist/services/open_ai_pass_through.js.map +1 -1
- package/dist/services/organization_management.d.ts +21 -1
- package/dist/services/organization_management.d.ts.map +1 -1
- package/dist/services/organization_management.js +20 -2
- package/dist/services/organization_management.js.map +1 -1
- package/dist/services/plugins.js +1 -1
- package/dist/services/policies.d.ts +2 -0
- package/dist/services/policies.d.ts.map +1 -1
- package/dist/services/policies.js +3 -1
- package/dist/services/policies.js.map +1 -1
- package/dist/services/policy_engine.d.ts +6 -0
- package/dist/services/policy_engine.d.ts.map +1 -1
- package/dist/services/policy_engine.js +4 -1
- package/dist/services/policy_engine.js.map +1 -1
- package/dist/services/project_management.d.ts +15 -0
- package/dist/services/project_management.d.ts.map +1 -1
- package/dist/services/project_management.js +14 -1
- package/dist/services/project_management.js.map +1 -1
- package/dist/services/prompts.js +1 -1
- package/dist/services/public.d.ts +124 -0
- package/dist/services/public.d.ts.map +1 -1
- package/dist/services/public.js +158 -1
- package/dist/services/public.js.map +1 -1
- package/dist/services/rag.js +1 -1
- package/dist/services/realtime.d.ts +4 -26
- package/dist/services/realtime.d.ts.map +1 -1
- package/dist/services/realtime.js +7 -57
- package/dist/services/realtime.js.map +1 -1
- package/dist/services/rerank.d.ts +72 -0
- package/dist/services/rerank.d.ts.map +1 -1
- package/dist/services/rerank.js +61 -4
- package/dist/services/rerank.js.map +1 -1
- package/dist/services/responses.d.ts +30 -0
- package/dist/services/responses.d.ts.map +1 -1
- package/dist/services/responses.js +59 -1
- package/dist/services/responses.js.map +1 -1
- package/dist/services/router_settings.js +1 -1
- package/dist/services/rust_control_plane.js +1 -1
- package/dist/services/scim.d.ts +41 -1
- package/dist/services/scim.d.ts.map +1 -1
- package/dist/services/scim.js +60 -2
- package/dist/services/scim.js.map +1 -1
- package/dist/services/search.js +1 -1
- package/dist/services/search_tools.js +1 -1
- package/dist/services/settings.d.ts +88 -0
- package/dist/services/settings.d.ts.map +1 -1
- package/dist/services/settings.js +113 -1
- package/dist/services/settings.js.map +1 -1
- package/dist/services/sso_settings.d.ts +1 -1
- package/dist/services/sso_settings.d.ts.map +1 -1
- package/dist/services/sso_settings.js +1 -1
- package/dist/services/sso_settings.js.map +1 -1
- package/dist/services/tag_management.d.ts +7 -0
- package/dist/services/tag_management.d.ts.map +1 -1
- package/dist/services/tag_management.js +8 -1
- package/dist/services/tag_management.js.map +1 -1
- package/dist/services/team_management.d.ts +265 -17
- package/dist/services/team_management.d.ts.map +1 -1
- package/dist/services/team_management.js +247 -12
- package/dist/services/team_management.js.map +1 -1
- package/dist/services/tools.js +1 -1
- package/dist/services/ui_settings.js +1 -1
- package/dist/services/ui_theme_settings.js +1 -1
- package/dist/services/usage_ai.js +1 -1
- package/dist/services/vantage.js +1 -1
- package/dist/services/vector_store_management.js +1 -1
- package/dist/services/vector_stores.js +1 -1
- package/dist/services/videos.js +1 -1
- package/dist/services/web_socket.d.ts +0 -18
- package/dist/services/web_socket.d.ts.map +1 -1
- package/dist/services/web_socket.js +1 -33
- package/dist/services/web_socket.js.map +1 -1
- package/dist/services/workflow_management.js +1 -1
- package/package.json +1 -1
- package/src/services/a2a.ts +1 -1
- package/src/services/a2a_registration.ts +1 -1
- package/src/services/access_groups.ts +44 -1
- package/src/services/adaptive_router.ts +1 -1
- package/src/services/agents.ts +22 -1
- package/src/services/alerting.ts +1 -1
- package/src/services/anthropic_passthrough.ts +1 -1
- package/src/services/anthropic_skills.ts +12 -2
- package/src/services/assistants.ts +1 -1
- package/src/services/audio.ts +1 -1
- package/src/services/audit_logging.ts +4 -1
- package/src/services/auto_router.ts +303 -93
- package/src/services/batch.ts +1 -1
- package/src/services/beta_agents.ts +1 -1
- package/src/services/beta_mcp.ts +1 -1
- package/src/services/budget_management.ts +12 -4
- package/src/services/budget_spend_tracking.ts +38 -6
- package/src/services/cache_settings.ts +1 -1
- package/src/services/caching.ts +1 -1
- package/src/services/chat_completions.ts +1 -1
- package/src/services/claude_code_marketplace.ts +20 -14
- package/src/services/cloudzero.ts +1 -1
- package/src/services/completions.ts +1 -1
- package/src/services/compliance.ts +1 -1
- package/src/services/config_overrides.ts +8 -2
- package/src/services/config_yaml.ts +251 -6
- package/src/services/containers.ts +29 -11
- package/src/services/coordination_redis_settings.ts +1 -1
- package/src/services/cost_tracking.ts +198 -2
- package/src/services/credential_management.ts +25 -6
- package/src/services/customer_management.ts +49 -3
- package/src/services/email_management.ts +1 -1
- package/src/services/embeddings.ts +1 -1
- package/src/services/evals.ts +1 -1
- package/src/services/experimental.ts +1 -1
- package/src/services/fallback_management.ts +1 -1
- package/src/services/files.ts +1 -1
- package/src/services/fine_tuning.ts +1 -1
- package/src/services/gemini_agents.ts +1 -1
- package/src/services/google_genai_endpoints.ts +1 -1
- package/src/services/guardrails.ts +198 -8
- package/src/services/health.ts +3 -2
- package/src/services/images.ts +1 -1
- package/src/services/index.ts +1 -15
- package/src/services/internal_user_management.ts +565 -103
- package/src/services/invite_links.ts +1 -1
- package/src/services/jwt_mappings.ts +7 -1
- package/src/services/key_management.ts +257 -127
- package/src/services/langfuse_passthrough.ts +1 -1
- package/src/services/llm_passthrough.ts +4542 -0
- package/src/services/llm_utils.ts +1 -1
- package/src/services/logging_callbacks.ts +1 -1
- package/src/services/mcp_byok_oauth.ts +65 -45
- package/src/services/mcp_discoverable.ts +109 -1
- package/src/services/mcp_management.ts +258 -3
- package/src/services/mcp_rest.ts +30 -5
- package/src/services/memory_management.ts +4 -1
- package/src/services/misc.ts +269 -2
- package/src/services/model_management.ts +666 -21
- package/src/services/moderations.ts +1 -1
- package/src/services/ocr.ts +1 -1
- package/src/services/open_ai_pass_through.ts +81 -286
- package/src/services/organization_management.ts +44 -2
- package/src/services/plugins.ts +1 -1
- package/src/services/policies.ts +5 -1
- package/src/services/policy_engine.ts +10 -1
- package/src/services/project_management.ts +33 -1
- package/src/services/prompts.ts +1 -1
- package/src/services/public.ts +356 -1
- package/src/services/rag.ts +1 -1
- package/src/services/realtime.ts +11 -111
- package/src/services/rerank.ts +187 -7
- package/src/services/responses.ts +117 -1
- package/src/services/router_settings.ts +1 -1
- package/src/services/rust_control_plane.ts +1 -1
- package/src/services/scim.ts +134 -3
- package/src/services/search.ts +1 -1
- package/src/services/search_tools.ts +1 -1
- package/src/services/settings.ts +278 -1
- package/src/services/sso_settings.ts +2 -1
- package/src/services/tag_management.ts +15 -1
- package/src/services/team_management.ts +676 -37
- package/src/services/tools.ts +1 -1
- package/src/services/ui_settings.ts +1 -1
- package/src/services/ui_theme_settings.ts +1 -1
- package/src/services/usage_ai.ts +1 -1
- package/src/services/vantage.ts +1 -1
- package/src/services/vector_store_management.ts +1 -1
- package/src/services/vector_stores.ts +1 -1
- package/src/services/videos.ts +1 -1
- package/src/services/web_socket.ts +1 -61
- package/src/services/workflow_management.ts +1 -1
- package/dist/services/anthropic_pass_through.d.ts +0 -70
- package/dist/services/anthropic_pass_through.d.ts.map +0 -1
- package/dist/services/anthropic_pass_through.js +0 -114
- package/dist/services/anthropic_pass_through.js.map +0 -1
- package/dist/services/assembly_ai_eu_pass_through.d.ts +0 -70
- package/dist/services/assembly_ai_eu_pass_through.d.ts.map +0 -1
- package/dist/services/assembly_ai_eu_pass_through.js +0 -114
- package/dist/services/assembly_ai_eu_pass_through.js.map +0 -1
- package/dist/services/assembly_ai_pass_through.d.ts +0 -70
- package/dist/services/assembly_ai_pass_through.d.ts.map +0 -1
- package/dist/services/assembly_ai_pass_through.js +0 -114
- package/dist/services/assembly_ai_pass_through.js.map +0 -1
- package/dist/services/aws_comprehend_medical_pass_through.d.ts +0 -36
- package/dist/services/aws_comprehend_medical_pass_through.d.ts.map +0 -1
- package/dist/services/aws_comprehend_medical_pass_through.js +0 -56
- package/dist/services/aws_comprehend_medical_pass_through.js.map +0 -1
- package/dist/services/azure_ai_pass_through.d.ts +0 -70
- package/dist/services/azure_ai_pass_through.d.ts.map +0 -1
- package/dist/services/azure_ai_pass_through.js +0 -112
- package/dist/services/azure_ai_pass_through.js.map +0 -1
- package/dist/services/azure_pass_through.d.ts +0 -70
- package/dist/services/azure_pass_through.d.ts.map +0 -1
- package/dist/services/azure_pass_through.js +0 -107
- package/dist/services/azure_pass_through.js.map +0 -1
- package/dist/services/bedrock_pass_through.d.ts +0 -70
- package/dist/services/bedrock_pass_through.d.ts.map +0 -1
- package/dist/services/bedrock_pass_through.js +0 -114
- package/dist/services/bedrock_pass_through.js.map +0 -1
- package/dist/services/cohere_pass_through.d.ts +0 -70
- package/dist/services/cohere_pass_through.d.ts.map +0 -1
- package/dist/services/cohere_pass_through.js +0 -112
- package/dist/services/cohere_pass_through.js.map +0 -1
- package/dist/services/cursor_pass_through.d.ts +0 -70
- package/dist/services/cursor_pass_through.d.ts.map +0 -1
- package/dist/services/cursor_pass_through.js +0 -112
- package/dist/services/cursor_pass_through.js.map +0 -1
- package/dist/services/google_ai_studio_pass_through.d.ts +0 -70
- package/dist/services/google_ai_studio_pass_through.d.ts.map +0 -1
- package/dist/services/google_ai_studio_pass_through.js +0 -112
- package/dist/services/google_ai_studio_pass_through.js.map +0 -1
- package/dist/services/milvus_pass_through.d.ts +0 -70
- package/dist/services/milvus_pass_through.d.ts.map +0 -1
- package/dist/services/milvus_pass_through.js +0 -112
- package/dist/services/milvus_pass_through.js.map +0 -1
- package/dist/services/mistral_pass_through.d.ts +0 -70
- package/dist/services/mistral_pass_through.d.ts.map +0 -1
- package/dist/services/mistral_pass_through.js +0 -114
- package/dist/services/mistral_pass_through.js.map +0 -1
- package/dist/services/vertex_ai_pass_through.d.ts +0 -180
- package/dist/services/vertex_ai_pass_through.d.ts.map +0 -1
- package/dist/services/vertex_ai_pass_through.js +0 -334
- package/dist/services/vertex_ai_pass_through.js.map +0 -1
- package/dist/services/vllm_pass_through.d.ts +0 -70
- package/dist/services/vllm_pass_through.d.ts.map +0 -1
- package/dist/services/vllm_pass_through.js +0 -104
- package/dist/services/vllm_pass_through.js.map +0 -1
- package/dist/services/watsonx_pass_through.d.ts +0 -70
- package/dist/services/watsonx_pass_through.d.ts.map +0 -1
- package/dist/services/watsonx_pass_through.js +0 -114
- package/dist/services/watsonx_pass_through.js.map +0 -1
- package/src/services/anthropic_pass_through.ts +0 -233
- package/src/services/assembly_ai_eu_pass_through.ts +0 -237
- package/src/services/assembly_ai_pass_through.ts +0 -237
- package/src/services/aws_comprehend_medical_pass_through.ts +0 -109
- package/src/services/azure_ai_pass_through.ts +0 -231
- package/src/services/azure_pass_through.ts +0 -227
- package/src/services/bedrock_pass_through.ts +0 -229
- package/src/services/cohere_pass_through.ts +0 -227
- package/src/services/cursor_pass_through.ts +0 -227
- package/src/services/google_ai_studio_pass_through.ts +0 -227
- package/src/services/milvus_pass_through.ts +0 -227
- package/src/services/mistral_pass_through.ts +0 -229
- package/src/services/vertex_ai_pass_through.ts +0 -684
- package/src/services/vllm_pass_through.ts +0 -227
- package/src/services/watsonx_pass_through.ts +0 -229
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.
|
|
1
|
+
// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.103.0 — see README). Do not edit.
|
|
2
2
|
import * as S from "@distilled.cloud/core/schema";
|
|
3
3
|
import * as API from "@distilled.cloud/core/api";
|
|
4
4
|
import * as C from "@distilled.cloud/core/category";
|
|
@@ -18,6 +18,11 @@ export class UnprocessableEntity extends /*@__PURE__*/ /*@__PURE__*/ T.applyErro
|
|
|
18
18
|
}).pipe(C.withBadRequestError), [{ status: 422 }]) {
|
|
19
19
|
}
|
|
20
20
|
export const LiteLLMParamsAdaptiveRouterConfigMap = /*@__PURE__*/ S.Record(S.String, S.Unknown);
|
|
21
|
+
export const AwsSessionTag = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
22
|
+
Key: S.String,
|
|
23
|
+
Value: S.String,
|
|
24
|
+
})).annotate({ identifier: "AwsSessionTag" });
|
|
25
|
+
export const LiteLLMParamsAwsSessionTagsList = /*@__PURE__*/ S.Array(AwsSessionTag);
|
|
21
26
|
export const LiteLLMParamsBedrockTagsList = /*@__PURE__*/ S.Array(S.Unknown);
|
|
22
27
|
export const LiteLLMParamsComplexityRouterConfigMap = /*@__PURE__*/ S.Record(S.String, S.Unknown);
|
|
23
28
|
export const ConfigurableClientsideParamsCustomAuthInput =
|
|
@@ -29,6 +34,7 @@ export const ConfigurableClientsideParamsCustomAuthInput =
|
|
|
29
34
|
export const LiteLLMParamsConfigurableClientsideAuthParamsItem = S.Unknown;
|
|
30
35
|
export const LiteLLMParamsConfigurableClientsideAuthParamsList =
|
|
31
36
|
/*@__PURE__*/ S.Array(LiteLLMParamsConfigurableClientsideAuthParamsItem);
|
|
37
|
+
export const LiteLLMParamsDropParams = S.Unknown;
|
|
32
38
|
export const LiteLLMParamsMilvusPartitionNamesList = /*@__PURE__*/ S.Array(S.String);
|
|
33
39
|
export const ChoicesFinishReason = S.String;
|
|
34
40
|
export const ChatCompletionTokenLogprobBytesList = /*@__PURE__*/ S.Array(S.Number);
|
|
@@ -202,6 +208,7 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
202
208
|
adaptive_router_default_model: S.optional(S.NullOr(S.String)),
|
|
203
209
|
allow_client_keepalive_override: S.optional(S.NullOr(S.Boolean)),
|
|
204
210
|
annotation_cost_per_page: S.optional(S.NullOr(S.Number)),
|
|
211
|
+
annotation_cost_per_page_batches: S.optional(S.NullOr(S.Number)),
|
|
205
212
|
api_base: S.optional(S.NullOr(S.String)),
|
|
206
213
|
api_key: S.optional(S.NullOr(S.String)),
|
|
207
214
|
api_version: S.optional(S.NullOr(S.String)),
|
|
@@ -210,6 +217,8 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
210
217
|
auto_router_default_model: S.optional(S.NullOr(S.String)),
|
|
211
218
|
auto_router_embedding_model: S.optional(S.NullOr(S.String)),
|
|
212
219
|
auto_router_max_input_chars: S.optional(S.NullOr(S.Number)),
|
|
220
|
+
auto_router_model_compression: S.optional(S.NullOr(S.String)),
|
|
221
|
+
auto_router_routing_compression: S.optional(S.NullOr(S.String)),
|
|
213
222
|
aws_access_key_id: S.optional(S.NullOr(S.String)),
|
|
214
223
|
aws_batch_role_arn: S.optional(S.NullOr(S.String)),
|
|
215
224
|
aws_bedrock_project_id: S.optional(S.NullOr(S.String)),
|
|
@@ -220,10 +229,14 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
220
229
|
aws_role_name: S.optional(S.NullOr(S.String)),
|
|
221
230
|
aws_secret_access_key: S.optional(S.NullOr(S.String)),
|
|
222
231
|
aws_session_name: S.optional(S.NullOr(S.String)),
|
|
232
|
+
aws_session_tags: S.optional(S.NullOr(LiteLLMParamsAwsSessionTagsList)),
|
|
223
233
|
aws_session_token: S.optional(S.NullOr(S.String)),
|
|
224
234
|
aws_sts_endpoint: S.optional(S.NullOr(S.String)),
|
|
225
235
|
aws_web_identity_token: S.optional(S.NullOr(S.String)),
|
|
226
236
|
azure_ad_token: S.optional(S.NullOr(S.String)),
|
|
237
|
+
azure_password: S.optional(S.NullOr(S.String)),
|
|
238
|
+
azure_scope: S.optional(S.NullOr(S.String)),
|
|
239
|
+
azure_username: S.optional(S.NullOr(S.String)),
|
|
227
240
|
bedrock_tags: S.optional(S.NullOr(LiteLLMParamsBedrockTagsList)),
|
|
228
241
|
budget_duration: S.optional(S.NullOr(S.String)),
|
|
229
242
|
cache_creation_input_audio_token_cost: S.optional(S.NullOr(S.Number)),
|
|
@@ -248,22 +261,27 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
248
261
|
cache_read_input_token_cost_priority: S.optional(S.NullOr(S.Number)),
|
|
249
262
|
cache_read_input_token_cost_ultrafast: S.optional(S.NullOr(S.Number)),
|
|
250
263
|
citation_cost_per_token: S.optional(S.NullOr(S.Number)),
|
|
264
|
+
client_id: S.optional(S.NullOr(S.String)),
|
|
265
|
+
client_secret: S.optional(S.NullOr(S.String)),
|
|
251
266
|
complexity_router_config: S.optional(S.NullOr(LiteLLMParamsComplexityRouterConfigMap)),
|
|
252
267
|
complexity_router_default_model: S.optional(S.NullOr(S.String)),
|
|
253
268
|
configurable_clientside_auth_params: S.optional(S.NullOr(LiteLLMParamsConfigurableClientsideAuthParamsList)),
|
|
254
269
|
custom_llm_provider: S.optional(S.NullOr(S.String)),
|
|
255
270
|
default_api_key_rpm_limit: S.optional(S.NullOr(S.Number)),
|
|
256
271
|
default_api_key_tpm_limit: S.optional(S.NullOr(S.Number)),
|
|
272
|
+
drop_params: S.optional(S.NullOr(LiteLLMParamsDropParams)),
|
|
257
273
|
gcs_bucket_name: S.optional(S.NullOr(S.String)),
|
|
258
274
|
google_maps_grounding_cost_per_query: S.optional(S.NullOr(S.Number)),
|
|
259
275
|
input_cost_per_audio_per_second: S.optional(S.NullOr(S.Number)),
|
|
260
276
|
input_cost_per_audio_per_second_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
261
277
|
input_cost_per_audio_token: S.optional(S.NullOr(S.Number)),
|
|
278
|
+
input_cost_per_audio_token_batches: S.optional(S.NullOr(S.Number)),
|
|
262
279
|
input_cost_per_character: S.optional(S.NullOr(S.Number)),
|
|
263
280
|
input_cost_per_character_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
264
281
|
input_cost_per_image: S.optional(S.NullOr(S.Number)),
|
|
265
282
|
input_cost_per_image_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
266
283
|
input_cost_per_image_token: S.optional(S.NullOr(S.Number)),
|
|
284
|
+
input_cost_per_image_token_batches: S.optional(S.NullOr(S.Number)),
|
|
267
285
|
input_cost_per_pixel: S.optional(S.NullOr(S.Number)),
|
|
268
286
|
input_cost_per_query: S.optional(S.NullOr(S.Number)),
|
|
269
287
|
input_cost_per_second: S.optional(S.NullOr(S.Number)),
|
|
@@ -285,6 +303,7 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
285
303
|
input_cost_per_video_per_second_above_15s_interval: S.optional(S.NullOr(S.Number)),
|
|
286
304
|
input_cost_per_video_per_second_above_8s_interval: S.optional(S.NullOr(S.Number)),
|
|
287
305
|
input_cost_per_video_token: S.optional(S.NullOr(S.Number)),
|
|
306
|
+
input_cost_per_video_token_batches: S.optional(S.NullOr(S.Number)),
|
|
288
307
|
itpm: S.optional(S.NullOr(S.Number)),
|
|
289
308
|
keepalive_seconds: S.optional(S.NullOr(S.Number)),
|
|
290
309
|
litellm_credential_name: S.optional(S.NullOr(S.String)),
|
|
@@ -301,6 +320,7 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
301
320
|
model_info: S.optional(S.NullOr(LiteLLMParamsModelInfoMap)),
|
|
302
321
|
ocr_cost_per_credit: S.optional(S.NullOr(S.Number)),
|
|
303
322
|
ocr_cost_per_page: S.optional(S.NullOr(S.Number)),
|
|
323
|
+
ocr_cost_per_page_batches: S.optional(S.NullOr(S.Number)),
|
|
304
324
|
organization: S.optional(S.NullOr(S.String)),
|
|
305
325
|
otpm: S.optional(S.NullOr(S.Number)),
|
|
306
326
|
output_cost_per_audio_per_second: S.optional(S.NullOr(S.Number)),
|
|
@@ -317,6 +337,7 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
317
337
|
output_cost_per_second_1080p: S.optional(S.NullOr(S.Number)),
|
|
318
338
|
output_cost_per_second_480p: S.optional(S.NullOr(S.Number)),
|
|
319
339
|
output_cost_per_second_4k: S.optional(S.NullOr(S.Number)),
|
|
340
|
+
output_cost_per_second_720p: S.optional(S.NullOr(S.Number)),
|
|
320
341
|
output_cost_per_token: S.optional(S.NullOr(S.Number)),
|
|
321
342
|
output_cost_per_token_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
322
343
|
output_cost_per_token_above_200k_tokens: S.optional(S.NullOr(S.Number)),
|
|
@@ -341,12 +362,14 @@ export const LiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
341
362
|
rpm: S.optional(S.NullOr(S.Number)),
|
|
342
363
|
s3_bucket_name: S.optional(S.NullOr(S.String)),
|
|
343
364
|
s3_encryption_key_id: S.optional(S.NullOr(S.String)),
|
|
365
|
+
s3_endpoint_url: S.optional(S.NullOr(S.String)),
|
|
344
366
|
s3_output_bucket_name: S.optional(S.NullOr(S.String)),
|
|
345
367
|
s3_region_name: S.optional(S.NullOr(S.String)),
|
|
346
368
|
search_context_cost_per_query: S.optional(S.NullOr(LiteLLMParamsSearchContextCostPerQueryMap)),
|
|
347
369
|
stream_timeout: S.optional(S.NullOr(LiteLLMParamsStreamTimeout)),
|
|
348
370
|
tag_regex: S.optional(S.NullOr(LiteLLMParamsTagRegexList)),
|
|
349
371
|
tags: S.optional(S.NullOr(LiteLLMParamsTagsList)),
|
|
372
|
+
tenant_id: S.optional(S.NullOr(S.String)),
|
|
350
373
|
tiered_pricing: S.optional(S.NullOr(LiteLLMParamsTieredPricingList)),
|
|
351
374
|
timeout: S.optional(S.NullOr(LiteLLMParamsTimeout)),
|
|
352
375
|
tpm: S.optional(S.NullOr(S.Number)),
|
|
@@ -385,6 +408,8 @@ export const LitellmTypesRouterModelInfo = /*@__PURE__*/ S.suspend(() => S.Struc
|
|
|
385
408
|
id: S.NullOr(S.String),
|
|
386
409
|
input_cost_per_character: S.optional(S.NullOr(S.Number)),
|
|
387
410
|
input_cost_per_token: S.optional(S.NullOr(S.Number)),
|
|
411
|
+
internal_router_model: S.optional(S.NullOr(S.Boolean)),
|
|
412
|
+
member_auto_router: S.optional(S.Boolean),
|
|
388
413
|
output_cost_per_character: S.optional(S.NullOr(S.Number)),
|
|
389
414
|
output_cost_per_token: S.optional(S.NullOr(S.Number)),
|
|
390
415
|
ptu_count: S.optional(S.NullOr(S.Number)),
|
|
@@ -716,6 +741,8 @@ export const GetModelInfoV2V2ModelInfoRequest = /*@__PURE__*/ S.suspend(() => S.
|
|
|
716
741
|
sortBy: S.optional(S.String.pipe(T.Query())),
|
|
717
742
|
sortOrder: S.optional(S.String.pipe(T.Query())),
|
|
718
743
|
exclude_auto_routers: S.optional(S.Boolean.pipe(T.Query())),
|
|
744
|
+
access_group: S.optional(S.String.pipe(T.Query())),
|
|
745
|
+
wildcard_only: S.optional(S.Boolean.pipe(T.Query())),
|
|
719
746
|
}).pipe(T.Http({ method: "GET", uri: "/v2/model/info", code: 200 }))).annotate({
|
|
720
747
|
identifier: "GetModelInfoV2V2ModelInfoRequest",
|
|
721
748
|
});
|
|
@@ -842,12 +869,14 @@ export const ModelListV1ModelsGetResponse = /*@__PURE__*/ S.suspend(() => S.Stru
|
|
|
842
869
|
});
|
|
843
870
|
export const UpdateLiteLLMParamsAdaptiveRouterConfigMap =
|
|
844
871
|
/*@__PURE__*/ S.Record(S.String, S.Unknown);
|
|
872
|
+
export const UpdateLiteLLMParamsAwsSessionTagsList = /*@__PURE__*/ S.Array(AwsSessionTag);
|
|
845
873
|
export const UpdateLiteLLMParamsBedrockTagsList = /*@__PURE__*/ S.Array(S.Unknown);
|
|
846
874
|
export const UpdateLiteLLMParamsComplexityRouterConfigMap =
|
|
847
875
|
/*@__PURE__*/ S.Record(S.String, S.Unknown);
|
|
848
876
|
export const UpdateLiteLLMParamsConfigurableClientsideAuthParamsItem = S.Unknown;
|
|
849
877
|
export const UpdateLiteLLMParamsConfigurableClientsideAuthParamsList =
|
|
850
878
|
/*@__PURE__*/ S.Array(UpdateLiteLLMParamsConfigurableClientsideAuthParamsItem);
|
|
879
|
+
export const UpdateLiteLLMParamsDropParams = S.Unknown;
|
|
851
880
|
export const UpdateLiteLLMParamsMilvusPartitionNamesList =
|
|
852
881
|
/*@__PURE__*/ S.Array(S.String);
|
|
853
882
|
export const UpdateLiteLLMParamsMockResponse = S.Unknown;
|
|
@@ -869,6 +898,7 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
869
898
|
adaptive_router_default_model: S.optional(S.NullOr(S.String)),
|
|
870
899
|
allow_client_keepalive_override: S.optional(S.NullOr(S.Boolean)),
|
|
871
900
|
annotation_cost_per_page: S.optional(S.NullOr(S.Number)),
|
|
901
|
+
annotation_cost_per_page_batches: S.optional(S.NullOr(S.Number)),
|
|
872
902
|
api_base: S.optional(S.NullOr(S.String)),
|
|
873
903
|
api_key: S.optional(S.NullOr(S.String)),
|
|
874
904
|
api_version: S.optional(S.NullOr(S.String)),
|
|
@@ -877,6 +907,8 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
877
907
|
auto_router_default_model: S.optional(S.NullOr(S.String)),
|
|
878
908
|
auto_router_embedding_model: S.optional(S.NullOr(S.String)),
|
|
879
909
|
auto_router_max_input_chars: S.optional(S.NullOr(S.Number)),
|
|
910
|
+
auto_router_model_compression: S.optional(S.NullOr(S.String)),
|
|
911
|
+
auto_router_routing_compression: S.optional(S.NullOr(S.String)),
|
|
880
912
|
aws_access_key_id: S.optional(S.NullOr(S.String)),
|
|
881
913
|
aws_batch_role_arn: S.optional(S.NullOr(S.String)),
|
|
882
914
|
aws_bedrock_project_id: S.optional(S.NullOr(S.String)),
|
|
@@ -887,10 +919,14 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
887
919
|
aws_role_name: S.optional(S.NullOr(S.String)),
|
|
888
920
|
aws_secret_access_key: S.optional(S.NullOr(S.String)),
|
|
889
921
|
aws_session_name: S.optional(S.NullOr(S.String)),
|
|
922
|
+
aws_session_tags: S.optional(S.NullOr(UpdateLiteLLMParamsAwsSessionTagsList)),
|
|
890
923
|
aws_session_token: S.optional(S.NullOr(S.String)),
|
|
891
924
|
aws_sts_endpoint: S.optional(S.NullOr(S.String)),
|
|
892
925
|
aws_web_identity_token: S.optional(S.NullOr(S.String)),
|
|
893
926
|
azure_ad_token: S.optional(S.NullOr(S.String)),
|
|
927
|
+
azure_password: S.optional(S.NullOr(S.String)),
|
|
928
|
+
azure_scope: S.optional(S.NullOr(S.String)),
|
|
929
|
+
azure_username: S.optional(S.NullOr(S.String)),
|
|
894
930
|
bedrock_tags: S.optional(S.NullOr(UpdateLiteLLMParamsBedrockTagsList)),
|
|
895
931
|
budget_duration: S.optional(S.NullOr(S.String)),
|
|
896
932
|
cache_creation_input_audio_token_cost: S.optional(S.NullOr(S.Number)),
|
|
@@ -915,22 +951,27 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
915
951
|
cache_read_input_token_cost_priority: S.optional(S.NullOr(S.Number)),
|
|
916
952
|
cache_read_input_token_cost_ultrafast: S.optional(S.NullOr(S.Number)),
|
|
917
953
|
citation_cost_per_token: S.optional(S.NullOr(S.Number)),
|
|
954
|
+
client_id: S.optional(S.NullOr(S.String)),
|
|
955
|
+
client_secret: S.optional(S.NullOr(S.String)),
|
|
918
956
|
complexity_router_config: S.optional(S.NullOr(UpdateLiteLLMParamsComplexityRouterConfigMap)),
|
|
919
957
|
complexity_router_default_model: S.optional(S.NullOr(S.String)),
|
|
920
958
|
configurable_clientside_auth_params: S.optional(S.NullOr(UpdateLiteLLMParamsConfigurableClientsideAuthParamsList)),
|
|
921
959
|
custom_llm_provider: S.optional(S.NullOr(S.String)),
|
|
922
960
|
default_api_key_rpm_limit: S.optional(S.NullOr(S.Number)),
|
|
923
961
|
default_api_key_tpm_limit: S.optional(S.NullOr(S.Number)),
|
|
962
|
+
drop_params: S.optional(S.NullOr(UpdateLiteLLMParamsDropParams)),
|
|
924
963
|
gcs_bucket_name: S.optional(S.NullOr(S.String)),
|
|
925
964
|
google_maps_grounding_cost_per_query: S.optional(S.NullOr(S.Number)),
|
|
926
965
|
input_cost_per_audio_per_second: S.optional(S.NullOr(S.Number)),
|
|
927
966
|
input_cost_per_audio_per_second_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
928
967
|
input_cost_per_audio_token: S.optional(S.NullOr(S.Number)),
|
|
968
|
+
input_cost_per_audio_token_batches: S.optional(S.NullOr(S.Number)),
|
|
929
969
|
input_cost_per_character: S.optional(S.NullOr(S.Number)),
|
|
930
970
|
input_cost_per_character_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
931
971
|
input_cost_per_image: S.optional(S.NullOr(S.Number)),
|
|
932
972
|
input_cost_per_image_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
933
973
|
input_cost_per_image_token: S.optional(S.NullOr(S.Number)),
|
|
974
|
+
input_cost_per_image_token_batches: S.optional(S.NullOr(S.Number)),
|
|
934
975
|
input_cost_per_pixel: S.optional(S.NullOr(S.Number)),
|
|
935
976
|
input_cost_per_query: S.optional(S.NullOr(S.Number)),
|
|
936
977
|
input_cost_per_second: S.optional(S.NullOr(S.Number)),
|
|
@@ -952,6 +993,7 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
952
993
|
input_cost_per_video_per_second_above_15s_interval: S.optional(S.NullOr(S.Number)),
|
|
953
994
|
input_cost_per_video_per_second_above_8s_interval: S.optional(S.NullOr(S.Number)),
|
|
954
995
|
input_cost_per_video_token: S.optional(S.NullOr(S.Number)),
|
|
996
|
+
input_cost_per_video_token_batches: S.optional(S.NullOr(S.Number)),
|
|
955
997
|
itpm: S.optional(S.NullOr(S.Number)),
|
|
956
998
|
keepalive_seconds: S.optional(S.NullOr(S.Number)),
|
|
957
999
|
litellm_credential_name: S.optional(S.NullOr(S.String)),
|
|
@@ -968,6 +1010,7 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
968
1010
|
model_info: S.optional(S.NullOr(UpdateLiteLLMParamsModelInfoMap)),
|
|
969
1011
|
ocr_cost_per_credit: S.optional(S.NullOr(S.Number)),
|
|
970
1012
|
ocr_cost_per_page: S.optional(S.NullOr(S.Number)),
|
|
1013
|
+
ocr_cost_per_page_batches: S.optional(S.NullOr(S.Number)),
|
|
971
1014
|
organization: S.optional(S.NullOr(S.String)),
|
|
972
1015
|
otpm: S.optional(S.NullOr(S.Number)),
|
|
973
1016
|
output_cost_per_audio_per_second: S.optional(S.NullOr(S.Number)),
|
|
@@ -984,6 +1027,7 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
984
1027
|
output_cost_per_second_1080p: S.optional(S.NullOr(S.Number)),
|
|
985
1028
|
output_cost_per_second_480p: S.optional(S.NullOr(S.Number)),
|
|
986
1029
|
output_cost_per_second_4k: S.optional(S.NullOr(S.Number)),
|
|
1030
|
+
output_cost_per_second_720p: S.optional(S.NullOr(S.Number)),
|
|
987
1031
|
output_cost_per_token: S.optional(S.NullOr(S.Number)),
|
|
988
1032
|
output_cost_per_token_above_128k_tokens: S.optional(S.NullOr(S.Number)),
|
|
989
1033
|
output_cost_per_token_above_200k_tokens: S.optional(S.NullOr(S.Number)),
|
|
@@ -1008,12 +1052,14 @@ export const UpdateLiteLLMParams = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
1008
1052
|
rpm: S.optional(S.NullOr(S.Number)),
|
|
1009
1053
|
s3_bucket_name: S.optional(S.NullOr(S.String)),
|
|
1010
1054
|
s3_encryption_key_id: S.optional(S.NullOr(S.String)),
|
|
1055
|
+
s3_endpoint_url: S.optional(S.NullOr(S.String)),
|
|
1011
1056
|
s3_output_bucket_name: S.optional(S.NullOr(S.String)),
|
|
1012
1057
|
s3_region_name: S.optional(S.NullOr(S.String)),
|
|
1013
1058
|
search_context_cost_per_query: S.optional(S.NullOr(UpdateLiteLLMParamsSearchContextCostPerQueryMap)),
|
|
1014
1059
|
stream_timeout: S.optional(S.NullOr(UpdateLiteLLMParamsStreamTimeout)),
|
|
1015
1060
|
tag_regex: S.optional(S.NullOr(UpdateLiteLLMParamsTagRegexList)),
|
|
1016
1061
|
tags: S.optional(S.NullOr(UpdateLiteLLMParamsTagsList)),
|
|
1062
|
+
tenant_id: S.optional(S.NullOr(S.String)),
|
|
1017
1063
|
tiered_pricing: S.optional(S.NullOr(UpdateLiteLLMParamsTieredPricingList)),
|
|
1018
1064
|
timeout: S.optional(S.NullOr(UpdateLiteLLMParamsTimeout)),
|
|
1019
1065
|
tpm: S.optional(S.NullOr(S.Number)),
|
|
@@ -1081,11 +1127,16 @@ export const TierDefinition = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
1081
1127
|
})).annotate({ identifier: "TierDefinition" });
|
|
1082
1128
|
export const PostPreviewAutoRouterClassifierPromptAutoRouterClassifierDefaultPromptRequestTierDefinitionsList =
|
|
1083
1129
|
/*@__PURE__*/ S.Array(TierDefinition);
|
|
1130
|
+
export const PostPreviewAutoRouterClassifierPromptAutoRouterClassifierDefaultPromptRequestTierLabelsMap =
|
|
1131
|
+
/*@__PURE__*/ S.Record(S.String, S.String);
|
|
1084
1132
|
export const PostPreviewAutoRouterClassifierPromptAutoRouterClassifierDefaultPromptRequest =
|
|
1085
1133
|
/*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1134
|
+
classification_examples: S.optional(S.NullOr(S.String)),
|
|
1086
1135
|
classification_prompt: S.optional(S.NullOr(S.String)),
|
|
1136
|
+
classification_rubric: S.optional(S.NullOr(ClassificationRubric)),
|
|
1087
1137
|
context_window_size: S.optional(S.Number),
|
|
1088
|
-
tier_definitions: PostPreviewAutoRouterClassifierPromptAutoRouterClassifierDefaultPromptRequestTierDefinitionsList,
|
|
1138
|
+
tier_definitions: S.optional(S.NullOr(PostPreviewAutoRouterClassifierPromptAutoRouterClassifierDefaultPromptRequestTierDefinitionsList)),
|
|
1139
|
+
tier_labels: S.optional(S.NullOr(PostPreviewAutoRouterClassifierPromptAutoRouterClassifierDefaultPromptRequestTierLabelsMap)),
|
|
1089
1140
|
}).pipe(T.Http({
|
|
1090
1141
|
method: "POST",
|
|
1091
1142
|
uri: "/auto_router/classifier/default_prompt",
|
|
@@ -1100,26 +1151,132 @@ export const AdaptiveRouterWeights = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
1100
1151
|
})).annotate({
|
|
1101
1152
|
identifier: "AdaptiveRouterWeights",
|
|
1102
1153
|
});
|
|
1154
|
+
export const CapabilityCalibrationConfig = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1155
|
+
intercept: S.Number,
|
|
1156
|
+
slope: S.Number,
|
|
1157
|
+
version: S.String,
|
|
1158
|
+
})).annotate({
|
|
1159
|
+
identifier: "CapabilityCalibrationConfig",
|
|
1160
|
+
});
|
|
1161
|
+
export const CapabilityClassifierConfigResponseFormat = S.String;
|
|
1162
|
+
export const CapabilityClassifierConfig = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1163
|
+
base_threshold: S.Number,
|
|
1164
|
+
calibration: S.optional(S.NullOr(CapabilityCalibrationConfig)),
|
|
1165
|
+
capable_tier: S.String,
|
|
1166
|
+
efficient_tier: S.String,
|
|
1167
|
+
max_output_tokens: S.optional(S.Number),
|
|
1168
|
+
response_format: S.optional(CapabilityClassifierConfigResponseFormat),
|
|
1169
|
+
threshold_step: S.optional(S.Number),
|
|
1170
|
+
})).annotate({
|
|
1171
|
+
identifier: "CapabilityClassifierConfig",
|
|
1172
|
+
});
|
|
1173
|
+
export const RequestComplexityRouterConfigClassificationMode = S.String;
|
|
1103
1174
|
export const RequestComplexityRouterConfigClassifierFallback = S.String;
|
|
1175
|
+
export const ClassifierLLMConfigReasoningEffort = S.String;
|
|
1176
|
+
export const ClassifierVisionConfig = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1177
|
+
enabled: S.optional(S.Boolean),
|
|
1178
|
+
max_images: S.optional(S.Number),
|
|
1179
|
+
})).annotate({
|
|
1180
|
+
identifier: "ClassifierVisionConfig",
|
|
1181
|
+
});
|
|
1104
1182
|
export const ClassifierLLMConfig = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1183
|
+
circuit_breaker_cooldown_seconds: S.optional(S.Number),
|
|
1184
|
+
circuit_breaker_enabled: S.optional(S.Boolean),
|
|
1105
1185
|
classification_rubric: S.optional(S.NullOr(ClassificationRubric)),
|
|
1106
1186
|
model: S.String,
|
|
1187
|
+
reasoning_effort: S.optional(S.NullOr(ClassifierLLMConfigReasoningEffort)),
|
|
1107
1188
|
system_prompt: S.optional(S.NullOr(S.String)),
|
|
1108
1189
|
timeout_ms: S.optional(S.Number),
|
|
1190
|
+
vision: S.optional(ClassifierVisionConfig),
|
|
1109
1191
|
})).annotate({
|
|
1110
1192
|
identifier: "ClassifierLLMConfig",
|
|
1111
1193
|
});
|
|
1112
1194
|
export const RequestComplexityRouterConfigClassifierType = S.String;
|
|
1113
1195
|
export const RequestComplexityRouterConfigCodeKeywordsList =
|
|
1114
1196
|
/*@__PURE__*/ S.Array(S.String);
|
|
1197
|
+
export const CustomDimensionKeywordsList = /*@__PURE__*/ S.Array(S.String);
|
|
1198
|
+
export const CustomDimensionPatternsList = /*@__PURE__*/ S.Array(S.String);
|
|
1199
|
+
export const CustomDimensionScoringMode = S.String;
|
|
1200
|
+
export const CustomDimension = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1201
|
+
keywords: S.optional(CustomDimensionKeywordsList),
|
|
1202
|
+
name: S.String,
|
|
1203
|
+
patterns: S.optional(CustomDimensionPatternsList),
|
|
1204
|
+
scoring_mode: S.optional(CustomDimensionScoringMode),
|
|
1205
|
+
weight: S.Number,
|
|
1206
|
+
})).annotate({
|
|
1207
|
+
identifier: "CustomDimension",
|
|
1208
|
+
});
|
|
1209
|
+
export const RequestComplexityRouterConfigCustomDimensionsList =
|
|
1210
|
+
/*@__PURE__*/ S.Array(CustomDimension);
|
|
1115
1211
|
export const RequestComplexityRouterConfigCustomTechnicalKeywordsList =
|
|
1116
1212
|
/*@__PURE__*/ S.Array(S.String);
|
|
1117
1213
|
export const RequestComplexityRouterConfigDimensionWeightsMap =
|
|
1118
1214
|
/*@__PURE__*/ S.Record(S.String, S.Number);
|
|
1119
1215
|
export const RequestComplexityRouterConfigEscalationKeywordsList =
|
|
1120
1216
|
/*@__PURE__*/ S.Array(S.String);
|
|
1217
|
+
export const TierCohortStatistic = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1218
|
+
cohort: S.String,
|
|
1219
|
+
observations: S.Number,
|
|
1220
|
+
successes: S.Number,
|
|
1221
|
+
tier: S.Number,
|
|
1222
|
+
})).annotate({
|
|
1223
|
+
identifier: "TierCohortStatistic",
|
|
1224
|
+
});
|
|
1225
|
+
export const TrainedTierArtifactCohortStatisticsList = /*@__PURE__*/ S.Array(TierCohortStatistic);
|
|
1226
|
+
export const TierDataset = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1227
|
+
license: S.String,
|
|
1228
|
+
name: S.String,
|
|
1229
|
+
rows: S.Number,
|
|
1230
|
+
success_definition: S.optional(S.String),
|
|
1231
|
+
url: S.String,
|
|
1232
|
+
})).annotate({ identifier: "TierDataset" });
|
|
1233
|
+
export const TrainedTierArtifactDatasetsList = /*@__PURE__*/ S.Array(TierDataset);
|
|
1234
|
+
export const RequestType = S.String;
|
|
1235
|
+
export const TierDomainStatistic = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1236
|
+
observations: S.Number,
|
|
1237
|
+
request_type: RequestType,
|
|
1238
|
+
successes: S.Number,
|
|
1239
|
+
tier: S.Number,
|
|
1240
|
+
})).annotate({
|
|
1241
|
+
identifier: "TierDomainStatistic",
|
|
1242
|
+
});
|
|
1243
|
+
export const TrainedTierArtifactDomainStatisticsList = /*@__PURE__*/ S.Array(TierDomainStatistic);
|
|
1244
|
+
export const TierGlobalStatistic = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1245
|
+
observations: S.Number,
|
|
1246
|
+
successes: S.Number,
|
|
1247
|
+
tier: S.Number,
|
|
1248
|
+
})).annotate({
|
|
1249
|
+
identifier: "TierGlobalStatistic",
|
|
1250
|
+
});
|
|
1251
|
+
export const TrainedTierArtifactGlobalStatisticsList = /*@__PURE__*/ S.Array(TierGlobalStatistic);
|
|
1252
|
+
export const TrainedTierArtifact = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1253
|
+
cohort_prior_mass: S.optional(S.Number),
|
|
1254
|
+
cohort_statistics: S.optional(TrainedTierArtifactCohortStatisticsList),
|
|
1255
|
+
datasets: S.optional(TrainedTierArtifactDatasetsList),
|
|
1256
|
+
domain_prior_mass: S.optional(S.Number),
|
|
1257
|
+
domain_statistics: S.optional(TrainedTierArtifactDomainStatisticsList),
|
|
1258
|
+
global_statistics: TrainedTierArtifactGlobalStatisticsList,
|
|
1259
|
+
routing_threshold: S.optional(S.Number),
|
|
1260
|
+
schema_version: S.optional(S.Number),
|
|
1261
|
+
split_method: S.optional(S.String),
|
|
1262
|
+
success_definition: S.optional(S.String),
|
|
1263
|
+
})).annotate({
|
|
1264
|
+
identifier: "TrainedTierArtifact",
|
|
1265
|
+
});
|
|
1266
|
+
export const RequestComplexityRouterConfigHeuristicV2Artifact = S.Unknown;
|
|
1121
1267
|
export const RequestComplexityRouterConfigHousekeepingPatternsList =
|
|
1122
1268
|
/*@__PURE__*/ S.Array(S.String);
|
|
1269
|
+
export const JevClassifierConfig = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1270
|
+
api_base: S.optional(S.NullOr(S.String)),
|
|
1271
|
+
api_key: S.optional(S.NullOr(S.String)),
|
|
1272
|
+
circuit_breaker_cooldown_seconds: S.optional(S.Number),
|
|
1273
|
+
circuit_breaker_enabled: S.optional(S.Boolean),
|
|
1274
|
+
instructions: S.optional(S.NullOr(S.String)),
|
|
1275
|
+
model: S.optional(S.String),
|
|
1276
|
+
timeout_ms: S.optional(S.Number),
|
|
1277
|
+
})).annotate({
|
|
1278
|
+
identifier: "JevClassifierConfig",
|
|
1279
|
+
});
|
|
1123
1280
|
export const KeywordTierRuleKeywordsList = /*@__PURE__*/ S.Array(S.String);
|
|
1124
1281
|
export const KeywordTierRule = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1125
1282
|
keywords: KeywordTierRuleKeywordsList,
|
|
@@ -1129,6 +1286,35 @@ export const KeywordTierRule = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
|
1129
1286
|
});
|
|
1130
1287
|
export const RequestComplexityRouterConfigKeywordTierRulesList =
|
|
1131
1288
|
/*@__PURE__*/ S.Array(KeywordTierRule);
|
|
1289
|
+
export const LLMV2ProbabilityCalibration = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1290
|
+
intercept: S.Number,
|
|
1291
|
+
slope: S.Number,
|
|
1292
|
+
})).annotate({
|
|
1293
|
+
identifier: "LLMV2ProbabilityCalibration",
|
|
1294
|
+
});
|
|
1295
|
+
export const LLMV2Calibration = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1296
|
+
capable: LLMV2ProbabilityCalibration,
|
|
1297
|
+
efficient: LLMV2ProbabilityCalibration,
|
|
1298
|
+
prompt_version: S.String,
|
|
1299
|
+
version: S.String,
|
|
1300
|
+
})).annotate({
|
|
1301
|
+
identifier: "LLMV2Calibration",
|
|
1302
|
+
});
|
|
1303
|
+
export const LLMV2ConfigResponseFormat = S.String;
|
|
1304
|
+
export const LLMV2Config = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1305
|
+
calibration: S.optional(S.NullOr(LLMV2Calibration)),
|
|
1306
|
+
capable_profile: S.optional(S.NullOr(S.String)),
|
|
1307
|
+
capable_profile_preset: S.optional(S.NullOr(S.String)),
|
|
1308
|
+
capable_tier: S.optional(S.String),
|
|
1309
|
+
efficient_profile: S.optional(S.NullOr(S.String)),
|
|
1310
|
+
efficient_profile_preset: S.optional(S.NullOr(S.String)),
|
|
1311
|
+
efficient_tier: S.optional(S.String),
|
|
1312
|
+
harness: S.optional(S.NullOr(S.String)),
|
|
1313
|
+
harness_preset: S.optional(S.NullOr(S.String)),
|
|
1314
|
+
max_output_tokens: S.optional(S.Number),
|
|
1315
|
+
max_quality_gap: S.Number,
|
|
1316
|
+
response_format: S.optional(LLMV2ConfigResponseFormat),
|
|
1317
|
+
})).annotate({ identifier: "LLMV2Config" });
|
|
1132
1318
|
export const RequestComplexityRouterConfigPlanModePatternsList =
|
|
1133
1319
|
/*@__PURE__*/ S.Array(S.String);
|
|
1134
1320
|
export const RequestComplexityRouterConfigReasoningKeywordsList =
|
|
@@ -1172,6 +1358,9 @@ export const RequestComplexityRouterConfig = /*@__PURE__*/ S.suspend(() => S.Str
|
|
|
1172
1358
|
adaptive: S.optional(S.Boolean),
|
|
1173
1359
|
adaptive_eligible: S.optional(RequestComplexityRouterConfigAdaptiveEligible),
|
|
1174
1360
|
adaptive_weights: S.optional(AdaptiveRouterWeights),
|
|
1361
|
+
capability_classifier_config: S.optional(S.NullOr(CapabilityClassifierConfig)),
|
|
1362
|
+
classification_examples: S.optional(S.NullOr(S.String)),
|
|
1363
|
+
classification_mode: S.optional(RequestComplexityRouterConfigClassificationMode),
|
|
1175
1364
|
classification_prompt: S.optional(S.NullOr(S.String)),
|
|
1176
1365
|
classifier_context_budget_chars: S.optional(S.Number),
|
|
1177
1366
|
classifier_context_include_assistant_turns: S.optional(S.Boolean),
|
|
@@ -1183,17 +1372,28 @@ export const RequestComplexityRouterConfig = /*@__PURE__*/ S.suspend(() => S.Str
|
|
|
1183
1372
|
classifier_plugin_timeout_ms: S.optional(S.Number),
|
|
1184
1373
|
classifier_type: S.optional(RequestComplexityRouterConfigClassifierType),
|
|
1185
1374
|
code_keywords: S.optional(S.NullOr(RequestComplexityRouterConfigCodeKeywordsList)),
|
|
1375
|
+
context_window_escalation_buffer: S.optional(S.Number),
|
|
1376
|
+
custom_dimensions: S.optional(RequestComplexityRouterConfigCustomDimensionsList),
|
|
1186
1377
|
custom_technical_keywords: S.optional(S.NullOr(RequestComplexityRouterConfigCustomTechnicalKeywordsList)),
|
|
1187
1378
|
default_model: S.optional(S.NullOr(S.String)),
|
|
1188
1379
|
deployment_affinity: S.optional(S.Boolean),
|
|
1189
1380
|
dimension_weights: S.optional(RequestComplexityRouterConfigDimensionWeightsMap),
|
|
1190
1381
|
embedding_model: S.optional(S.NullOr(S.String)),
|
|
1382
|
+
enable_context_window_escalation: S.optional(S.Boolean),
|
|
1383
|
+
enable_non_reasoning_tier: S.optional(S.Boolean),
|
|
1191
1384
|
escalation_keywords: S.optional(S.NullOr(RequestComplexityRouterConfigEscalationKeywordsList)),
|
|
1192
1385
|
fallback_tier: S.optional(S.NullOr(S.String)),
|
|
1193
1386
|
heuristic_first_max_tier: S.optional(S.NullOr(S.String)),
|
|
1387
|
+
heuristic_v2_artifact: S.optional(RequestComplexityRouterConfigHeuristicV2Artifact),
|
|
1194
1388
|
housekeeping_patterns: S.optional(S.NullOr(RequestComplexityRouterConfigHousekeepingPatternsList)),
|
|
1389
|
+
hybrid_boundary_margin: S.optional(S.NullOr(S.Number)),
|
|
1390
|
+
jev_classifier_config: S.optional(S.NullOr(JevClassifierConfig)),
|
|
1195
1391
|
keyword_tier_rules: S.optional(S.NullOr(RequestComplexityRouterConfigKeywordTierRulesList)),
|
|
1392
|
+
llm_v2_config: S.optional(S.NullOr(LLMV2Config)),
|
|
1196
1393
|
match_threshold: S.optional(S.Number),
|
|
1394
|
+
max_tokens_from_tier_model: S.optional(S.Boolean),
|
|
1395
|
+
modality_pin_override: S.optional(S.Boolean),
|
|
1396
|
+
modality_routing: S.optional(S.Boolean),
|
|
1197
1397
|
plan_mode_min_tier: S.optional(S.NullOr(S.String)),
|
|
1198
1398
|
plan_mode_patterns: S.optional(S.NullOr(RequestComplexityRouterConfigPlanModePatternsList)),
|
|
1199
1399
|
plugins: S.optional(S.NullOr(S.Unknown)),
|
|
@@ -1206,6 +1406,9 @@ export const RequestComplexityRouterConfig = /*@__PURE__*/ S.suspend(() => S.Str
|
|
|
1206
1406
|
session_affinity: S.optional(S.Boolean),
|
|
1207
1407
|
session_affinity_ttl_seconds: S.optional(S.Number),
|
|
1208
1408
|
simple_keywords: S.optional(S.NullOr(RequestComplexityRouterConfigSimpleKeywordsList)),
|
|
1409
|
+
stall_escalation_enabled: S.optional(S.Boolean),
|
|
1410
|
+
stall_escalation_repeat_threshold: S.optional(S.Number),
|
|
1411
|
+
stall_escalation_window: S.optional(S.Number),
|
|
1209
1412
|
technical_keywords: S.optional(S.NullOr(RequestComplexityRouterConfigTechnicalKeywordsList)),
|
|
1210
1413
|
tier_boundaries: S.optional(RequestComplexityRouterConfigTierBoundariesMap),
|
|
1211
1414
|
tier_definitions: S.optional(S.NullOr(RequestComplexityRouterConfigTierDefinitionsList)),
|
|
@@ -1244,6 +1447,18 @@ export const PostPreviewAutoRouterRoutingAutoRouterTestRoutingRequest =
|
|
|
1244
1447
|
identifier: "PostPreviewAutoRouterRoutingAutoRouterTestRoutingRequest",
|
|
1245
1448
|
});
|
|
1246
1449
|
export const StandardLoggingRoutingDecisionCause = S.String;
|
|
1450
|
+
export const StandardLoggingRoutingDecisionClassifierProbabilitiesMap =
|
|
1451
|
+
/*@__PURE__*/ S.Record(S.String, S.Number);
|
|
1452
|
+
export const StandardLoggingHeuristicV2ForecastProbabilitiesMap =
|
|
1453
|
+
/*@__PURE__*/ S.Record(S.String, S.Number);
|
|
1454
|
+
export const StandardLoggingHeuristicV2Forecast = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1455
|
+
predicted_tier: S.String,
|
|
1456
|
+
probabilities: StandardLoggingHeuristicV2ForecastProbabilitiesMap,
|
|
1457
|
+
request_type: S.String,
|
|
1458
|
+
threshold: S.Number,
|
|
1459
|
+
})).annotate({
|
|
1460
|
+
identifier: "StandardLoggingHeuristicV2Forecast",
|
|
1461
|
+
});
|
|
1247
1462
|
export const StandardLoggingRoutingDecisionRouterType = S.String;
|
|
1248
1463
|
export const StandardLoggingRoutingDecisionSignalsList = /*@__PURE__*/ S.Array(S.String);
|
|
1249
1464
|
export const StandardLoggingRoutingDecisionTierBoundaries =
|
|
@@ -1258,11 +1473,29 @@ export const StandardLoggingRoutingDecisionTierLitellmParamsMap =
|
|
|
1258
1473
|
/*@__PURE__*/ S.Record(S.String, S.Unknown);
|
|
1259
1474
|
export const StandardLoggingRoutingDecision = /*@__PURE__*/ S.suspend(() => S.Struct({
|
|
1260
1475
|
cause: S.optional(StandardLoggingRoutingDecisionCause),
|
|
1476
|
+
classifier_calibrated_capable_p_solve: S.optional(S.Number),
|
|
1477
|
+
classifier_calibrated_efficient_p_solve: S.optional(S.Number),
|
|
1478
|
+
classifier_calibrated_p_solve: S.optional(S.Number),
|
|
1479
|
+
classifier_calibration_version: S.optional(S.String),
|
|
1480
|
+
classifier_capability_boundary: S.optional(S.String),
|
|
1481
|
+
classifier_capable_p_solve: S.optional(S.Number),
|
|
1482
|
+
classifier_confidence: S.optional(S.Number),
|
|
1261
1483
|
classifier_cost: S.optional(S.Number),
|
|
1484
|
+
classifier_crux: S.optional(S.String),
|
|
1485
|
+
classifier_efficient_p_solve: S.optional(S.Number),
|
|
1486
|
+
classifier_max_quality_gap: S.optional(S.Number),
|
|
1262
1487
|
classifier_model: S.optional(S.String),
|
|
1488
|
+
classifier_p_solve: S.optional(S.Number),
|
|
1489
|
+
classifier_primary_rule: S.optional(S.String),
|
|
1490
|
+
classifier_probabilities: S.optional(StandardLoggingRoutingDecisionClassifierProbabilitiesMap),
|
|
1491
|
+
classifier_prompt_version: S.optional(S.String),
|
|
1492
|
+
classifier_threshold: S.optional(S.Number),
|
|
1493
|
+
context_escalated: S.optional(S.Boolean),
|
|
1494
|
+
context_escalation_original_tier: S.optional(S.String),
|
|
1263
1495
|
conversation_continuing: S.optional(S.Boolean),
|
|
1264
1496
|
escalated: S.optional(S.Boolean),
|
|
1265
1497
|
escalation_keyword: S.optional(S.String),
|
|
1498
|
+
heuristic_v2_forecast: S.optional(StandardLoggingHeuristicV2Forecast),
|
|
1266
1499
|
matched_keyword: S.optional(S.String),
|
|
1267
1500
|
reasoning_override_min_score: S.optional(S.Number),
|
|
1268
1501
|
request_type: S.optional(S.String),
|
|
@@ -1546,7 +1779,7 @@ export const getModelCostMapReloadStatusScheduleModelCostMapReloadStatusGet = /*
|
|
|
1546
1779
|
protocol: LitellmProtocol,
|
|
1547
1780
|
retry: Retry.Retry,
|
|
1548
1781
|
}));
|
|
1549
|
-
/** Get Model Cost Map Source ADMIN ONLY / MASTER KEY Only Endpoint Returns information about where the current model cost/pricing data was loaded from. Response fields: - source: "local" (bundled backup) or "remote" (fetched from URL) - url: the remote URL that was attempted (null when env-forced local) - is_env_forced: true if LITELLM_LOCAL_MODEL_COST_MAP=True forced local usage - fallback_reason: human-readable reason why remote failed (null on success) - model_count: number of models in the currently loaded cost map */
|
|
1782
|
+
/** Get Model Cost Map Source ADMIN ONLY / MASTER KEY Only Endpoint Returns information about where the current model cost/pricing data was loaded from. Response fields: - source: "local" (bundled backup) or "remote" (fetched from URL) - url: the remote URL that was attempted (null when env-forced local) - is_env_forced: true if LITELLM_LOCAL_MODEL_COST_MAP=True forced local usage - fallback_reason: human-readable reason why remote failed (null on success) - loaded_at: when this pod last loaded the map - source_revision: git blob id of the loaded file, what git rev-parse <commit>:<path> prints for it - etag: the ETag of the remote fetch (null for the bundled backup) - model_count: number of models in the currently loaded cost map */
|
|
1550
1783
|
export const getModelCostMapSourceModelCostMapSourceGet = /*@__PURE__*/ API.make(() => ({
|
|
1551
1784
|
input: GetModelCostMapSourceModelCostMapSourceGetRequest,
|
|
1552
1785
|
output: GetModelCostMapSourceModelCostMapSourceGetResponse,
|
|
@@ -1586,7 +1819,7 @@ export const getModelInfoModelsModelId = /*@__PURE__*/ API.make(() => ({
|
|
|
1586
1819
|
protocol: LitellmProtocol,
|
|
1587
1820
|
retry: Retry.Retry,
|
|
1588
1821
|
}));
|
|
1589
|
-
/** Model Info V1 Provides more info about each model in /models, including config.yaml descriptions (except api key and api base) Parameters: litellm_model_id: Optional[str] = None (this is the value of `x-litellm-model-id` returned in response headers) - When litellm_model_id is passed, it will return the info for that specific model - When litellm_model_id is not passed, it will return the info for all models - include_team_models: When true, filter to deployments the caller can use (same as /v2/model/info). - teamId: Filter to models accessible by the given team. - healthy_only: When true, hide models whose backing deployments are all marked unhealthy by background health checks, matching `/v1/models?healthy_only=true`. Set `general_settings.model_list_healthy_only: true` to apply this to every caller without the query parameter. Requires `background_health_checks: true`, plus either `model_list_healthy_only` or `enable_health_check_routing` to keep deployment health state cached; without health state the listing is returned unfiltered (fail open). Ignored when `litellm_model_id` is passed, since that is a direct lookup of one deployment rather than a listing. Hiding is presentation-only: a hidden model can still be called directly. Each model in the list response includes `model_info.access_via_team_ids` and `model_info.direct_access` when the proxy database is connected. Returns:
|
|
1822
|
+
/** Model Info V1 Provides more info about each model in /models, including config.yaml descriptions (except api key and api base) Parameters: litellm_model_id: Optional[str] = None (this is the value of `x-litellm-model-id` returned in response headers) - When litellm_model_id is passed, it will return the info for that specific model - When litellm_model_id is not passed, it will return the info for all models - include_team_models: When true, filter to deployments the caller can use (same as /v2/model/info). - teamId: Filter to models accessible by the given team. - healthy_only: When true, hide models whose backing deployments are all marked unhealthy by background health checks, matching `/v1/models?healthy_only=true`. Set `general_settings.model_list_healthy_only: true` to apply this to every caller without the query parameter. Requires `background_health_checks: true`, plus either `model_list_healthy_only` or `enable_health_check_routing` to keep deployment health state cached; without health state the listing is returned unfiltered (fail open). Ignored when `litellm_model_id` is passed, since that is a direct lookup of one deployment rather than a listing. Hiding is presentation-only: a hidden model can still be called directly. Each model in the list response includes `model_info.access_via_team_ids` and `model_info.direct_access` when the proxy database is connected. Returns: A JSON response whose `data` list holds one entry per model. Example Response: ```json { "data": [ { "model_name": "fake-openai-endpoint", "litellm_params": { "api_base": "https://exampleopenaiendpoint-production.up.railway.app/", "model": "openai/fake" }, "model_info": { "id": "112f74fab24a7a5245d2ced3536dd8f5f9192c57ee6e332af0f0512e08bed5af", "db_model": false } } ] } ``` */
|
|
1590
1823
|
export const getModelInfoV1ModelInfo = /*@__PURE__*/ API.make(() => ({
|
|
1591
1824
|
input: GetModelInfoV1ModelInfoRequest,
|
|
1592
1825
|
output: GetModelInfoV1ModelInfoResponse,
|
|
@@ -1602,7 +1835,7 @@ export const getModelInfoV1ModelsModelId = /*@__PURE__*/ API.make(() => ({
|
|
|
1602
1835
|
protocol: LitellmProtocol,
|
|
1603
1836
|
retry: Retry.Retry,
|
|
1604
1837
|
}));
|
|
1605
|
-
/** Model Info V1 Provides more info about each model in /models, including config.yaml descriptions (except api key and api base) Parameters: litellm_model_id: Optional[str] = None (this is the value of `x-litellm-model-id` returned in response headers) - When litellm_model_id is passed, it will return the info for that specific model - When litellm_model_id is not passed, it will return the info for all models - include_team_models: When true, filter to deployments the caller can use (same as /v2/model/info). - teamId: Filter to models accessible by the given team. - healthy_only: When true, hide models whose backing deployments are all marked unhealthy by background health checks, matching `/v1/models?healthy_only=true`. Set `general_settings.model_list_healthy_only: true` to apply this to every caller without the query parameter. Requires `background_health_checks: true`, plus either `model_list_healthy_only` or `enable_health_check_routing` to keep deployment health state cached; without health state the listing is returned unfiltered (fail open). Ignored when `litellm_model_id` is passed, since that is a direct lookup of one deployment rather than a listing. Hiding is presentation-only: a hidden model can still be called directly. Each model in the list response includes `model_info.access_via_team_ids` and `model_info.direct_access` when the proxy database is connected. Returns:
|
|
1838
|
+
/** Model Info V1 Provides more info about each model in /models, including config.yaml descriptions (except api key and api base) Parameters: litellm_model_id: Optional[str] = None (this is the value of `x-litellm-model-id` returned in response headers) - When litellm_model_id is passed, it will return the info for that specific model - When litellm_model_id is not passed, it will return the info for all models - include_team_models: When true, filter to deployments the caller can use (same as /v2/model/info). - teamId: Filter to models accessible by the given team. - healthy_only: When true, hide models whose backing deployments are all marked unhealthy by background health checks, matching `/v1/models?healthy_only=true`. Set `general_settings.model_list_healthy_only: true` to apply this to every caller without the query parameter. Requires `background_health_checks: true`, plus either `model_list_healthy_only` or `enable_health_check_routing` to keep deployment health state cached; without health state the listing is returned unfiltered (fail open). Ignored when `litellm_model_id` is passed, since that is a direct lookup of one deployment rather than a listing. Hiding is presentation-only: a hidden model can still be called directly. Each model in the list response includes `model_info.access_via_team_ids` and `model_info.direct_access` when the proxy database is connected. Returns: A JSON response whose `data` list holds one entry per model. Example Response: ```json { "data": [ { "model_name": "fake-openai-endpoint", "litellm_params": { "api_base": "https://exampleopenaiendpoint-production.up.railway.app/", "model": "openai/fake" }, "model_info": { "id": "112f74fab24a7a5245d2ced3536dd8f5f9192c57ee6e332af0f0512e08bed5af", "db_model": false } } ] } ``` */
|
|
1606
1839
|
export const getModelInfoV1V1ModelInfo = /*@__PURE__*/ API.make(() => ({
|
|
1607
1840
|
input: GetModelInfoV1V1ModelInfoRequest,
|
|
1608
1841
|
output: GetModelInfoV1V1ModelInfoResponse,
|
|
@@ -1610,7 +1843,7 @@ export const getModelInfoV1V1ModelInfo = /*@__PURE__*/ API.make(() => ({
|
|
|
1610
1843
|
protocol: LitellmProtocol,
|
|
1611
1844
|
retry: Retry.Retry,
|
|
1612
1845
|
}));
|
|
1613
|
-
/** Model Info V2 Paginated model metadata for proxy deployments (pricing, provider, team access). Returns configured router deployments with enriched `model_info` (costs, provider, context window, etc.). Sensitive fields such as API keys and api_base are omitted. Query parameters: model: Filter to a single public `model_name`. user_models_only: When true, only return models created by the calling user. include_team_models: When true, populate `access_via_team_ids` and `direct_access` on each model and filter to deployments the caller can use. page / size: Pagination controls (defaults: page=1, size=50). search: Case-insensitive partial match on model name or team public name. modelId: Return a single deployment by LiteLLM model id. teamId: Filter to models with direct access or team membership for this team id. sortBy / sortOrder: Sort by model_name, created_at, updated_at, costs, or status. Example request: ``` curl -X GET 'http://localhost:4000/v2/model/info?include_team_models=true&page=1&size=50' \ --header 'Authorization: Bearer sk-1234' ``` Example response: ```json { "data": [ { "model_name": "gpt-4", "litellm_params": {"model": "openai/gpt-4.1"}, "model_info": { "id": "abc123", "litellm_provider": "openai", "access_via_team_ids": ["team-1"], "direct_access": true } } ], "total_count": 1, "current_page": 1, "total_pages": 1, "size": 50 } ``` */
|
|
1846
|
+
/** Model Info V2 Paginated model metadata for proxy deployments (pricing, provider, team access). Returns configured router deployments with enriched `model_info` (costs, provider, context window, etc.). Sensitive fields such as API keys and api_base are omitted. Query parameters: model: Filter to a single public `model_name`. user_models_only: When true, only return models created by the calling user. include_team_models: When true, populate `access_via_team_ids` and `direct_access` on each model and filter to deployments the caller can use. page / size: Pagination controls (defaults: page=1, size=50). search: Case-insensitive partial match on model name or team public name. modelId: Return a single deployment by LiteLLM model id. teamId: Filter to models with direct access or team membership for this team id. sortBy / sortOrder: Sort by model_name, created_at, updated_at, costs, or status. access_group: Only return deployments in this model access group. wildcard_only: Only return deployments whose `model_name` contains `*`. Example request: ``` curl -X GET 'http://localhost:4000/v2/model/info?include_team_models=true&page=1&size=50' \ --header 'Authorization: Bearer sk-1234' ``` Example response: ```json { "data": [ { "model_name": "gpt-4", "litellm_params": {"model": "openai/gpt-4.1"}, "model_info": { "id": "abc123", "litellm_provider": "openai", "access_via_team_ids": ["team-1"], "direct_access": true } } ], "total_count": 1, "current_page": 1, "total_pages": 1, "size": 50 } ``` */
|
|
1614
1847
|
export const getModelInfoV2V2ModelInfo = /*@__PURE__*/ API.make(() => ({
|
|
1615
1848
|
input: GetModelInfoV2V2ModelInfoRequest,
|
|
1616
1849
|
output: GetModelInfoV2V2ModelInfoResponse,
|
|
@@ -1794,7 +2027,7 @@ export const updateUsefulLinksModelHubUpdateUsefulLinksPost = /*@__PURE__*/ API.
|
|
|
1794
2027
|
protocol: LitellmProtocol,
|
|
1795
2028
|
retry: Retry.Retry,
|
|
1796
2029
|
}));
|
|
1797
|
-
/** Validate Complexity Router Config Validate a complexity-router config without saving it. Runs the same check every write path runs (the router's own pydantic model), so a form can show the backend's exact verdict while the operator is still editing rather than after a rejected save.
|
|
2030
|
+
/** Validate Complexity Router Config Validate a complexity-router config without saving it. Runs the same check every write path runs (the router's own pydantic model), so a form can show the backend's exact verdict while the operator is still editing rather than after a rejected save. Uses the same team opt-in and model-access checks as configuration writes for members. Nothing is created, routed, or billed. */
|
|
1798
2031
|
export const validateComplexityRouterConfigAutoRouterValidateComplexityRouterConfigPost = /*@__PURE__*/ API.make(() => ({
|
|
1799
2032
|
input: ValidateComplexityRouterConfigAutoRouterValidateComplexityRouterConfigPostRequest,
|
|
1800
2033
|
output: ComplexityRouterConfigValidationResponse,
|