Topic hub pagination
Continue browsing this topic cluster with SEO-safe static pagination.
LiteLLM dashboard UI completely crashes when rendering wildcard model lists; infinite recursion in frontend model list expansion function; blocks admin management of proxy models Includes evidence for LiteLLM troubleshooting demand.
[Bug]: UI Crash: RangeError: Maximum call stack size exceeded in unfurlWildcardModelsInList Fix 504 gateway timeout on LiteLLM proxy when using non-streaming mode for long-running LLM responses Includes evidence for LiteLLM troubleshooting demand.
504 Gateway Time-out — The server didn't respond in time (LiteLLM proxy, non-streaming mode only) Diagnose and fix spurious BudgetExceededError in LiteLLM proxy caused by leaked Redis spending counters persisting from previous instances after upgrading versions Includes evidence for LiteLLM troubleshooting demand.
reserve_budget_for_request() leaks Redis spend counters — phantom BudgetExceededError after upgrade to LiteLLM v1.84.0 Fix LiteLLM proxy bug where standard HTTP Retry-After headers from upstream providers are not passed through to the client, breaking proper rate limit handling Includes evidence for LiteLLM troubleshooting demand.
upstream provider "retry-after" header not forwarded to client — only "llm_provider-retry-after" is returned Debug LiteLLM proxy when it fails to properly surface upstream provider rate limits or times out while routing to multiple LLM providers Includes evidence for LiteLLM troubleshooting demand.
LiteLLM rate_limit_exceeded / timeout when proxying requests through litellm.Router to upstream providers Fix regression in LiteLLM v1.85.0 where Bedrock AWS pass-through through the /converse endpoint returns 200 status but body is empty, causing silent data loss for clients consuming AI responses through LiteLLM proxy Includes evidence for LiteLLM troubleshooting demand.
Bedrock pass-through /converse returns HTTP 200 OK with empty response body on v1.85.0 — callers receive no content despite successful status code Fix spend logging gap where Ollama Cloud requests via LiteLLM proxy never appear in spend logs/Request Log UI due to non-unique sequential ID collisions with existing rows Includes evidence for LiteLLM troubleshooting demand.
LiteLLM_SpendLogs INSERT fails silently on PK constraint when Ollama Cloud returns short sequential IDs like chatcmpl-213 that collide across users 开发者通过 LiteLLM 代理调用 Modal Endpoint 时持续收到 404，但直接使用 Modal Client 和 Postman 正常，需要修复代理配置或排查兼容性问题 Includes evidence for LiteLLM troubleshooting demand.
404 Error with Modal Endpoint — works via Modal Client and PostMan but fails through LiteLLM proxy 使用 LiteLLM Proxy 的用户发现 stream_timeout 参数在首 chunk 未生效，导致慢启动 provider（如 Bedrock）请求被错误终止 Includes evidence for LiteLLM troubleshooting demand.
stream_timeout not enforced on first chunk; hanging providers cause completions to fail Fix 401 authentication failure when routing Azure OpenAI through LiteLLM proxy; authenticate properly using Azure AD tokens instead of api_key Includes evidence for LiteLLM / Azure OpenAI troubleshooting demand.
Authentication Error 401 — Connecting Azure OpenAI via LiteLLM (v1.63.7) with DSPy framework fails with 401 when using Azure AD Token Refresh method Prevent LiteLLM library from flooding LLM endpoints with unbounded concurrent requests causing 429 errors; fix cascading KeyError crash when retries exhaust Includes evidence for LiteLLM / OpenAI API troubleshooting demand.
KeyError: 'toc_detected' caused by unthrottled concurrent asyncio.gather calls triggering HTTP 429 Too Many Requests from LLM API — empty JSON parse returns {} then direct dict access crashes Developer using LiteLLM proxy server encounters model routing errors when connecting to third-party APIs — models specified in config.yaml not available or incorrectly mapped, causing production inference failures and API billing waste Includes evidence for LiteLLM troubleshooting demand.
litellm.exceptions.BadRequestError: xxx: (invalid_request_error) Model: 'xxx' doesn't exist Fix LiteLLM proxy that incorrectly returns HTTP 429 BudgetExceededError for users who haven't exceeded their budget due to Redis counter accumulation bug Includes evidence for LiteLLM troubleshooting demand.
reserve_budget_for_request() leaks Redis spend counters — phantom BudgetExceededError despite actual database spend being near $0 Fix phantom BudgetExceededError caused by leaking Redis spend counters in LiteLLM proxy — real database spend near $0 but users blocked at budget limit Includes evidence for LiteLLM troubleshooting demand.
Budget has been exceeded! EndUser=<id> Current cost: $52.19, Max budget: $50.0 — DB spend shows $0.001 while Redis counter reports $52 (BudgetExceededError HTTP 429) Fix multi-turn chat failures when using DeepSeek V4 Pro through LiteLLM proxy where assistant messages lose reasoning content on sequential turns Includes evidence for LiteLLM troubleshooting demand.
Assistant message missing reasoning_content — LiteLLM strips it when serializing response from turn 1 Fix LiteLLM proxy leaking internal _litellm_* parameters through virtual key rate limiting boundaries, breaking per-key rate limit enforcement Includes evidence for LiteLLM troubleshooting demand.
Virtual keys with tpm/rpm limits leak _litellm_* params — rate limit enforcement broken at proxy layer Fix TPM rate limiting that bypasses multi-replica deployments; each pod enforces its own TPM limit independently doubling/quadrupling effective throughput beyond intended cap Includes evidence for LiteLLM troubleshooting demand.
Deployment-level TPM enforcement is per-pod not cross-pod — effective limit becomes tpm_limit × N_replica allowing uncontrolled token consumption Fix phantom BudgetExceededError in LiteLLM proxy caused by Redis spend counter leaks; end users randomly blocked despite being within budget limits Includes evidence for LiteLLM troubleshooting demand.
BudgetExceededError (HTTP 429): "Budget has been exceeded! EndUser=<id> Current cost: <inflated>, Max budget: 50.0" — DB spend shows $0.001 while Redis counter reports $50+ after ~4 min of normal traffic Fix streaming fallback mechanism in LiteLLM router that drops mid-stream errors instead of gracefully switching to backup model Includes evidence for LiteLLM troubleshooting demand.
Router.aresponses streaming bypasses mid-stream fallback — MidStreamFallbackError not handled, breaking streaming responses when primary model fails mid-stream Developer follows LiteLLM docs for Cloudflare Workers AI setup but encounters KeyError because response structure changed; needs fix or updated example Includes evidence for LiteLLM troubleshooting demand.
KeyError: 'response' — Cloudflare Workers AI example code fails when using LiteLLM provider