{"openapi":"3.0.3","info":{"title":"vllm-medgemma-4b API","description":"MedGemma 4B BF16 abliterated medical vision inference via vLLM.\n\nGemma3ForConditionalGeneration with SigLIP medical vision encoder.\nSpecialized for: radiology, dermatology, histopathology, ophthalmology.\nNull-space abliteration preserves vision while removing refusals.\n\n128K context. No reasoning parser, no tool calling.\nUses --enforce-eager for correct Gemma3 bidirectional image attention.\n","version":"1.0.0","contact":{"name":"Haiven Infrastructure"}},"servers":[{"url":"https://medgemma-4b.haiven.site","description":"Direct access (via Traefik)"},{"url":"https://llm.haiven.site/v1","description":"Via LiteLLM gateway (recommended)"}],"paths":{"/health":{"get":{"summary":"Health check","operationId":"healthCheck","responses":{"200":{"description":"Service healthy"}}}},"/v1/models":{"get":{"summary":"List models","operationId":"listModels","responses":{"200":{"description":"Available models"}}}},"/v1/chat/completions":{"post":{"summary":"Chat completions","operationId":"chatCompletions","description":"OpenAI-compatible chat completions endpoint.\nSupports text and medical image input (radiology, dermatology, histopathology, ophthalmology).\nModel name: medgemma-4b\n","requestBody":{"required":true,"content":{"application/json":{"schema":{"type":"object","required":["model","messages"],"properties":{"model":{"type":"string","example":"medgemma-4b"},"messages":{"type":"array","items":{"type":"object"}},"temperature":{"type":"number","default":1.0},"max_tokens":{"type":"integer"},"stream":{"type":"boolean","default":false}}}}}},"responses":{"200":{"description":"Chat completion response"}}}},"/metrics":{"get":{"summary":"Prometheus metrics","operationId":"getMetrics","responses":{"200":{"description":"vLLM metrics in Prometheus format"}}}}}}