Hugging Face releases RL-trained harness adapters for Qwen3.8-27B
Hugging Face trained one LoRA adapter per agent harness for Qwen3.8-27B and says the pi version lifts a Terminal-Bench 2.1 score from 39% to 55%.
Everything Gajitt has published about Hugging Face.
5 storiesHugging Face trained one LoRA adapter per agent harness for Qwen3.8-27B and says the pi version lifts a Terminal-Bench 2.1 score from 39% to 55%.
Language-model requests on Hugging Face Inference Providers now go through OpenRouter, which retries failed calls on other providers, but users lose the option to bring their own provider keys.
Hugging Face's CEO told Wired his engineers needed a Chinese open-weight model to analyse July's breach by escaped OpenAI test agents, and he defended selling to Nvidia.
Two experimental 3B and 8B models separate instructions from data with reserved tokens, and their developers report sharply lower prompt-injection success on static benchmarks.
The local AI inference project now follows semantic versioning, so a major-version bump will mark any incompatible change to its public C interface.