Model · gpt4all-j
GPT4All-J
Developer: Nomic AI
Availability: available · checked 2026-09-24 · source
Raw record: /data/models/gpt4all-j.json
Fields
- id
- gpt4all-j
- identifiers
- huggingface
- nomic-ai/gpt4all-j
- developer
- Nomic AI
- release_date
- 2023-04-11partial · sourcenote: HF repo creation date; the technical report carries no date and no dated developer post was found this session.
- weights_status
- open
- availability
- available · checked 2026-09-24 · source
- license
- apache-2.0recorded · sourcenote: Card: 'An Apache-2 licensed chatbot'; the technical report is titled 'GPT4All-J: An Apache-2 Licensed Assistant-Style Chatbot'. Uploader nomic-ai is the developer.
- architecture
- family
- decoder_only
- note
- family read from config 'architectures': ['GPTJForCausalLM'].
- n_layers
- 28recorded · source
- hidden_size
- 4096recorded · source
- n_heads
- 16recorded · source
- vocab_size
- 50400recorded · source
- context_length
- 2048recorded · source
- positional_encoding
- rotary (RoPE), partial: 64 of 256 dims per headrecorded · sourcenote: Propagated from gpt-j-6b: the report says the weights are derived from GPT-J and the config matches (28 layers, hidden 4096, 16 heads).
- training_data
- The GPT4All-J dataset: prompts from LAION OIG subsets, Stack Overflow questions, a Bigscience/P3 subsample and custom creative-writing prompts; ~800k points, 'a superset of the original 400k GPT4All examples'. The initial-release model was trained on 437,605 post-processed examples (four epochs, LoRA); the full fine-tune of GPT-J for one epoch. Default HF revision is v1.0; later revisions (v1.1-breezy, v1.2-jazzy, v1.3-groovy) filter or extend the data.recorded · sourcenote: Report: 'The model associated with our initial public release is trained with LoRA on the 437,605 post-processed examples for four epochs while the finetuned GPT-J was trained for one epoch.' Which of those two checkpoints is nomic-ai/gpt4all-j (full GPTJForCausalLM weights) is not stated on the card.
- techniques
- primary_sources
- record_history
- date:2026-09-24 · change:ingested as candidate from HF (nomic-ai/gpt4all-j@5000faf803b3edcbeff9bdb6fdbbabd1a42addd0) · by:ingest_hf.py ·date:2026-09-24 · change:preparer: developer, license, positional encoding (propagated from gpt-j-6b), training data, 3 edges prepared; sources: card, GPT4All-J technical report · by:claude (preparer, Sonnet 5) ·date:2026-09-24 · change:reviewed and promoted from staging (3 edge(s) accepted) · by:Wilson Pruitt ·
Parents
Weights descend
- fine_tuned_from → GPT-J 6B declared source Report abstract: 'deriving its weights from the Apache-licensed GPT-J model rather than the GPL-licensed of LLaMA'; card: 'Finetuned From: GPT-J'.
Training data
- trained_on → GPT4All-J prompt generations declared source Card metadata `datasets: nomic-ai/gpt4all-j-prompt-generations`, and the report names the GPT4All-J dataset as its training set. Nomic is both uploader and developer, so declared. Default revision v1.0.
- trained_on → GPT4All prompt generations declared source Report: 'we curated the GPT4All-J dataset by augmenting the original 400k GPT4All examples' and the 800k set is 'a superset of the original 400k points GPT4All dataset'. This is the distillation path: the GPT4All examples were generated with GPT-3.5-Turbo (see the `gpt4all` dataset record).
Children
No edges recorded.
Read in
No station on the reading path has touched this record yet.