RLAIF with Nectar rankings
According to the model card, Starling-7B trains by Reinforcement Learning from AI Feedback on the GPT-4-labeled Nectar ranking dataset with a dedicated reward and policy-tuning pipeline.
Open Source Model Profile · berkeley-nest
Starling-LM-7B-alpha
Starling-LM-7B-alpha is a 7.24B-parameter Mistral RLAIF fine-tune from berkeley-nest. Its model card documents Nectar-dataset training and a reported 8.09 MT-Bench score.
- Publisher
- berkeley-nest
- Task
- text-generation
- Model type
- mistral
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Starling-LM-7B-alpha is published by berkeley-nest as a Mistral-family text-generation model. The captured configuration identifies MistralForCausalLM, and Safetensors metadata reports 7,241,748,480 parameters (about 7.24B). According to the model card, it is an RLAIF fine-tune of Openchat 3.5 on the Nectar ranking dataset, reporting an 8.09 MT-Bench score.
Recorded capabilities
Reported 8.09 MT-Bench
The model card reports an 8.09 MT-Bench score judged by GPT-4, positioning it behind only OpenAI GPT-4 and GPT-4 Turbo at the time of writing.
OpenChat 3.5 chat usage
The card says the model follows the exact chat template and usage of Openchat 3.5 and documents single-turn, multi-turn, and coding conversation prompts.
Use cases in the source record
- Conversational text-generation workflows that follow the card's documented OpenChat 3.5 chat template for single-turn and multi-turn dialogue.
- Coding-assistance experiments using the card's documented code-conversation prompt format.
Limitations and unknowns
- No independent evaluation results were extracted; the 8.09 MT-Bench figure is a publisher-reported card value.
- License framing needs care: captured metadata records Apache-2.0 while the card text conditions use on non-competition with OpenAI and non-commercial research preview.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: berkeley-nest/Starling-LM-7B-alpha
Captured: Unknown. Processed: 2026-09-07T19:34:41.518604+00:00.
Starling-LM-7B-alpha Developed by: Banghua Zhu * , Evan Frick * , Tianhao Wu * , Hanlin Zhu and Jiantao Jiao. Model type: Language Model finetuned with RLHF / RLAIF License: Apache-2.0 license under the condition that the model is not used to compete with OpenAI Finetuned from model: Openchat 3.5 (based on Mistral-7B-v0.1 ) We introduce Starling-7B, an open large language model (LLM) trained by Reinforcement Learning from AI Feedback (RLAIF). The model harnesses the power of our new GPT-4 labeled ranking dataset, berkeley-nest/Nectar , and our new reward training and policy tuning pipeline. Starling-7B-alpha scores 8.09 in MT Bench…
F001F002F003F004F005F006F007F009F010F011F012F013F015F016F019F020