Listwise reranking behavior
According to the model card, RankZephyr is fine-tuned to act as a listwise reranking agent that takes a query plus documents and returns reordered document identifiers.
Open Source Model Profile · castorini
rank_zephyr_7b_v1_full
rank_zephyr_7b_v1_full is a 7.24B-parameter Mistral text-generation fine-tune from castorini. Its model card describes it as a listwise reranking assistant built on Zephyr-7B-beta.
- Publisher
- castorini
- Task
- text-generation
- Model type
- mistral
- License
- mit
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
rank_zephyr_7b_v1_full is published by castorini as a Mistral-based text-generation model. The captured configuration identifies MistralForCausalLM and Safetensors metadata reports 7,241,732,096 parameters. According to the model card, RankZephyr is trained to act as a reranking assistant built on the Zephyr-7B-beta model, with the Full variant further fine-tuned on RankGPT-4 reorderings.
Recorded capabilities
RankGPT-4 second-stage tuning
According to the model card, RankZephyr Full is further fine-tuned on RankGPT-4 reorderings of OpenAI Ada2 orderings for 5K queries, after single-stage RankGPT-3.5 tuning.
Zephyr-7B-beta lineage
According to the model card, the model is built on Zephyr-7B-beta and fine-tuned from HuggingFaceH4/zephyr-7b-beta, with hub tags recording mistralai/Mistral-7B-v0.1 as base.
MIT license record
Card data records mit, and the model card states MIT licensing.
Use cases in the source record
- Listwise document reranking where the model receives a query and documents and returns a reordered list of document identifiers, used with the RankLLM repository.
Limitations and unknowns
- No evaluation results were extracted in structured form; publisher effectiveness claims about DL19/20/21/22, TREC-COVID, and TREC-News are model-card claims, not independently verified measurements.
- According to the model card, the model is trained on monolingual English data, so effectiveness on multilingual sets is not guaranteed.
- According to the model card, the Zephyr base was not safety-aligned with RLHF and can produce problematic outputs when prompted; base corpus size and composition are unknown.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: castorini/rank_zephyr_7b_v1_full
Captured: Unknown. Processed: 2026-09-07T19:34:42.126208+00:00.
Model Card for RankZephyr 7B V1 - Full RankZephyr is a series of language models trained to act as helpful reranking assistants built on the Zephyr-7B-β model. RankZephyr Base is the model that follows single-stage fine-tuning on the RankGPT-3.5 model, while RankZephyr Full is the model that is further fine-tuned on RankGPT-4 reorderings of OpenAI's Ada2 orderings for 5K queries. Model description Model type: A 7B parameter GPT-like model initially fine-tuned on a mix of publicly available, synthetic datasets, followed by task-specific listwise reranking data. Language(s) (NLP): Primarily English License: MIT Fine-tuned from model:…
F001F002F003F004F005F006F007F009F010F011F012F014F015F016F017F018