Model · llama-2-13b
Llama-2-13b-hf
Developer: Meta
Availability: gated · checked 2026-09-24 · source
HF gate: manual.
Raw record: /data/models/llama-2-13b.json
Fields
- id
- llama-2-13b
- identifiers
- huggingface
- meta-llama/Llama-2-13b-hf
- developer
- Meta
- release_date
- 2023-07-18recorded · source
- weights_status
- open
- availability
- gated · checked 2026-09-24 · sourcenote: HF gate: manual.
- license
- Llama 2 Community License Agreement (research and commercial use; licensees with >700M monthly active users on the release date must request a separate license from Meta).recorded · source
- architecture
- family
- unknown
- n_layers
- nullnot_recorded
- hidden_size
- nullnot_recorded
- n_heads
- nullnot_recorded
- vocab_size
- 32000recorded · sourcenote: Paper: 'The total vocabulary size is 32k tokens.'
- context_length
- 4096recorded · sourcenote: Table 1: '4k'.
- positional_encoding
- nullnot_recorded
- note
- Meta's HF config is gated (not fetched this session). Paper does not publish per-size layer/hidden/head counts for Llama 2 (unlike Llama 1's paper); same situation as the already-recorded llama-2-7b.
- training_data
- 2.0T tokens, 'a new mix of data from publicly available sources' (paper does not disclose sources). Same recipe as the 7B model.recorded · source
- techniques
- primary_sources
- record_history
- date:2026-09-24 · change:ingested as candidate from HF (meta-llama/Llama-2-13b-hf@5c31dfb671ce7cfe2d7bb7c04375e44c55e815b1) · by:ingest_hf.py ·date:2026-09-24 · change:preparer: developer, license, release date, vocab/context from paper, successor_in_series to llama-7b added · by:claude (preparer, Sonnet 5) ·date:2026-09-24 · change:preparer cleanup: removed 2 stale flag(s), filled training_data · by:claude (preparer, Sonnet 5) ·date:2026-09-24 · change:reviewed and promoted from staging (1 edge(s) accepted) · by:Wilson Pruitt ·
Parents
Design
- successor_in_series → LLaMA 7B declared source Paper: 'Llama 2, an updated version of Llama 1... trained on a new mix of publicly available data', not initialized from Llama 1 weights. Same edge as llama-2-7b.
Children
Weights descend
- ← fine_tuned_from vicuna-13b-v1.5 declared source Same as vicuna-7b-v1-5, 13B size.
- ← fine_tuned_from WizardLM-13B-V1.2 declared_by_uploader source Card: 'this model is trained from Llama-2 13b'.
- ← fine_tuned_from OpenOrcaxOpenChat-Preview2-13B declared source Card: 'We have used our own OpenOrca dataset to fine-tune Llama2-13B using OpenChat packing.'
- ← fine_tuned_from Platypus2-13B declared source Card: 'Platypus-13B is an instruction fine-tuned model based on the LLaMA2-13B transformer architecture'; 'instruction fine-tuned using LoRA on 1 A100 80GB'. Uploader garage-bAInd is the developer (card: trained by Cole Hunter & Ariel Lee). 'Based on ... architecture' alone would not show weight descent; the LoRA fine-tuning statement plus the Llama 2 tokenizer/config values do. The paper's abstract also describes 'fine-tuning and merging LoRA modules'; this card does not say whether a merge step produced this checkpoint.
Read in
No station on the reading path has touched this record yet.