TinyLlama

weights

open(Apache-2.0)

base

TinyLlama-1.1B(open, 3T-token pretrain documented)

post-training-data

open(SFT on UltraChat + DPO on UltraFeedback, both public datasets)

code

open(pretraining project public)

license

Apache-2.0(OSI)

TinyLlama-Chat builds on an open 1.1B base with a documented 3-trillion-token pretraining run, and its post-training uses named public datasets, UltraChat for supervised fine-tuning and UltraFeedback for DPO alignment. Weights, base, code and post-training data are all released under Apache-2.0.

• https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0 recorded 2026-08-13

  `license:apache-2.0` in the repo tags and `"gated":false`; base TinyLlama-1.1B; SFT on HuggingFaceH4/ultrachat_200k and DPO on HuggingFaceH4/ultrafeedback_binarized, both public; the TinyLlama pretraining project linked from the card

Adoption is measured as Hugging Face downloads for the single declared TinyLlama-Chat checkpoint. It is a workhorse tiny-chat model used for edge deployment, testing, CI pipelines and teaching.

• https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0 recorded 2026-08-13

  2,415,028 downloads in the trailing 30 days

TinyLlama-Chat is a 1.1B model on the 2023 Llama 2 architecture, and its model card publishes no benchmark table. Its base checkpoint scores 26.04 on MMLU (5-shot), so the model sits far below current frontier chat models.

• https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0/raw/main/README.md recorded 2026-09-29

  Chat card: 1.1B parameters, "exactly the same architecture and tokenizer as Llama 2", pretrained on 3 trillion tokens; the card publishes no benchmark table (searched the body for MMLU, benchmark, leaderboard: none).

• https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T/raw/main/README.md recorded 2026-09-29

  Base card for "TinyLlama-1.1B-intermediate-step-1431k-3T", the checkpoint the chat model is tuned from: Open LLM Leaderboard rows "MMLU (5-Shot)" 26.04 and "GSM8k (5-shot)" 1.44.