AI-Dynamo released Dynamo v1.4.1, a patch on top of v1.4.0, adding OpenAI-compatible /v1/classify and /v1/pooling endpoints to the Dynamo Frontend with vLLM worker support1. The release fixes a Dynamo Router bug where workers marked overloaded on the request path remained marked indefinitely, a logprob_token_ids passthrough rejection, and a vLLM-Omni NIXL object transfer failure caused by immutable bytes in torch.frombuffer calls. Pooling responses support vLLM-compatible float, base64, bytes, and bytes_only formats. Backend runtime versions remain unchanged from v1.4.0.
AI-Dynamo Releases Dynamo v1.4.1 With OpenAI-Compatible Classify and Pooling Endpoints
AI-Dynamo released Dynamo v1.4.1, adding OpenAI-compatible classify and pooling endpoints and fixing router overload recovery, logprob passthrough, and…