diff --git a/providers/inferx/models.toml b/providers/inferx/models.toml new file mode 100644 index 000000000..4ba158ef1 --- /dev/null +++ b/providers/inferx/models.toml @@ -0,0 +1,19 @@ +[[models]] +id = "qwen3.6-27b-fp8" +name = "Qwen/Qwen3.6-27B-FP8" + +[[models]] +id = "qwen3.6-27b-fp8-1m" +name = "Qwen/Qwen3.6-27B-FP8-1M" + +[[models]] +id = "qwen3.6-35b-a3b-fp8" +name = "Qwen/Qwen3.6-35B-A3B-FP8" + +[[models]] +id = "qwen3.6-35b-a3b-fp8-1m" +name = "Qwen/Qwen3.6-35B-A3B-FP8-1M" + +[[models]] +id = "gemma-4-31b-it-fp8" +name = "google/gemma-4-31b-it-fp8" diff --git a/providers/inferx/provider.toml b/providers/inferx/provider.toml new file mode 100644 index 000000000..2e8c744df --- /dev/null +++ b/providers/inferx/provider.toml @@ -0,0 +1,8 @@ +id = "inferx" +name = "InferX" +type = "openai-compatible" +docs = "https://inferx.net" + +[api] +base_url = "https://model.inferx.net/funccall/{tenant}/endpoints/{endpoint}/v1" +auth_type = "bearer"