Model · mistral-7b-sft-beta

Mistral 7B SFT β

Developer: Hugging Face (H4 team)
Availability: available · checked 2026-09-24 · source

Raw record: /data/models/mistral-7b-sft-beta.json

Fields

id

mistral-7b-sft-beta

identifiers

huggingface

HuggingFaceH4/mistral-7b-sft-beta

developer

Hugging Face (H4 team)

release_date

2023-10-26partial · source

note:  HF repo creation date (2023-10-26). The Zephyr paper (arXiv 2310.16944) is dated 2023-10-25, one day earlier, so the weights' own release may predate this.

weights_status

open

availability

available · checked 2026-09-24 · source

license

MITrecorded · source

note:  Card metadata `license: mit` and card text 'License: MIT'. Licence file text not separately read.

architecture

family

decoder_only

note

family read from config 'architectures': ['MistralForCausalLM'].

n_layers

32recorded · source

hidden_size

4096recorded · source

n_heads

32recorded · source

vocab_size

32000recorded · source

context_length

32768recorded · source

n_kv_heads

8recorded · source

positional_encoding

rotary (RoPE)recorded · source

note:  Architecture unchanged from mistral-7b-v0-1 (fine-tune, not a structural change).

training_data

HuggingFaceH4/ultrachat_200k: a filtered and preprocessed subset (~200k examples) of UltraChat, 'synthetic dialogues generated by ChatGPT'. One epoch, SFT only.recorded · source

note:  Card: fine-tuned 'on a filtered and preprocessed of the UltraChat dataset, which contains a diverse range of synthetic dialogues generated by ChatGPT'. Zephyr paper (arXiv 2310.16944 sec. on datasets): the 1.47M-example UltraChat corpus was filtered to about 200k examples.

techniques

primary_sources

https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta

https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta/raw/d7b9d226fb91df9de570ec0ee119f46e007f6142/config.json

record_history

date:2026-09-24 · change:ingested as candidate from HF (HuggingFaceH4/mistral-7b-sft-beta@d7b9d226fb91df9de570ec0ee119f46e007f6142) · by:ingest_hf.py ·

date:2026-09-24 · change:preparer: developer, license, training data, positional encoding (propagated), 2 edges; SFT-stage-of-Zephyr note; sources: card, Zephyr paper · by:claude (preparer, Sonnet 5) ·

date:2026-09-24 · change:reviewed and promoted from staging (2 edge(s) accepted) · by:Wilson Pruitt ·

Parents

Weights descend

• fine_tuned_from → Mistral 7B (v0.1) declared source Card: 'This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the HuggingFaceH4/ultrachat_200k dataset.' Model description: 'Finetuned from model: mistralai/Mistral-7B-v0.1'. Uploader HuggingFaceH4 is the Zephyr developer.

Training data

• trained_on → UltraChat declared source Card: fine-tuned on the ultrachat_200k dataset (H4's filtered subset of UltraChat; the `ultrachat` record covers the parent corpus, so this maps a filtered subset to its source). The distilled step: UltraChat text was written by ChatGPT (see the ultrachat record's distilled_from_outputs edge).

Children

Weights descend

• ← fine_tuned_from Zephyr 7B β declared source Developer's card for the SFT model: 'It is the SFT model that was used to train Zephyr-7B-β with Direct Preference Optimization.' The Zephyr card's own 'fine-tuned version of Mistral-7B-v0.1' edge stays: this edge records the intermediate SFT stage, so Zephyr's path is Mistral-7B-v0.1 -> mistral-7b-sft-beta -> zephyr-7b-beta.

Read in

No station on the reading path has touched this record yet.