Model · gpt-4
GPT-4 (Mar. 2023)
Developer: OpenAI
Availability: never_released · checked 2026-09-24 · source
Weights never released; API access status not tracked by this field.
Raw record: /data/models/gpt-4.json
Fields
- id
- gpt-4
- developer
- OpenAI
- release_date
- 2023-03-15partial · sourcenote: arXiv v1 date of the technical report.
- weights_status
- closed
- availability
- never_released · checked 2026-09-24 · sourcenote: Weights never released; API access status not tracked by this field.
- license
- Proprietary; no weights released.partial · source
- architecture
- family
- decoder_only
- note
- Stub specimen. Report: 'a Transformer-based model pre-trained to predict the next token in a document' and 'contains no further details about the architecture (including model size), hardware, training compute, dataset construction, training method, or similar.' Which GPT-4 snapshot ranked UltraFeedback is not_recorded.
- n_layers
- nullnot_recorded
- hidden_size
- nullnot_recorded
- n_heads
- nullnot_recorded
- vocab_size
- nullnot_recorded
- positional_encoding
- nullnot_recorded
- training_data
- nullnot_recorded
- techniques
- rlhf
- primary_sources
- record_history
- date:2026-09-24 · change:created as a stub specimen to resolve zephyr-7b-beta feedback_from edge · by:wilson-pruitt + claude ·date:2026-09-24 · change:availability checked and recorded · by:wilson-pruitt + claude ·
Parents
No edges recorded.
Children
Influence without weights
- ← feedback_from UltraFeedback declared source Builder's own paper: GPT-4 employed 'to offer detailed feedback in both numerical and textual forms.'
- ← distilled_from_outputs OpenOrca declared source Card: '~1M GPT-4 completions'. Builder's own card.
- ← distilled_from_outputs OpenHermes 2.5 declared source Builder's card tags the set 'GPT-4' and 'Distillation' and several sources are marked GPT-4-only. The card does not give a per-source split, so the share generated by GPT-4 is not_recorded; other closed-model outputs may also be present.
- ← feedback_from Nectar declared source Card: 'generated through GPT-4-based ranking'. The candidate answers were written by several models; only GPT-4's judgments are asserted here.
- ← distilled_from_outputs Tulu V2 SFT mixture declared source Card names GPT4-Alpaca as 'distilled GPT-4 data' and 30,000 Open-Orca samples 'generated by GPT-4'. Only these subsets are asserted; ShareGPT and WizardLM subsets also hold other models' outputs, not linked here.
- ← distilled_from_outputs Orca 1 explanation-tuning data (FLAN-5M / FLAN-1M) declared source Builder's own paper: '... and GPT-4 responses to FLAN-1M' (1M queries sampled from the 5M).
- ← distilled_from_outputs Orca 2 dataset (~817K) declared source Orca 2 paper section 4.2: 'training on ~1.8 million GPT-4 data' = Orca 1's 1M GPT-4 data plus Orca 2's 817K. The paper names GPT-4 for the doctor-patient conversations explicitly; for the other three sources it gives no separate teacher line, so this edge rests on the section 4.2 labelling. Tag `declared` is the paper's aggregate label ('~1.8 million GPT-4 data'), not a per-source statement: only the doctor-patient portion is attributed to GPT-4 by name, so for the rest GPT-4 authorship is the paper's own summary, not separately sourced (Wilson's ruling, 2026-09-24: keep declared, note the limit).
Read in
No station on the reading path has touched this record yet.