# jzhang38/TinyLlama

The TinyLlama project is an open endeavor to pretrain a 1.1B Llama model on 3 trillion tokens.

Repository: https://github.com/jzhang38/TinyLlama
Canonical: https://ross.abutalabs.com/products/tinyllama
Language: Python
License: Apache-2.0
License Family: permissive
Archived: true
Last push: 2024-05-03T20:21:20+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 78
- inputs: {"age_days": 1096, "days_push": 852, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 9019, forks 630 (observed 2026-08-28T04:10:26.789576+00:00)

## What it is
TinyLlama is an open-source project that pretrained a compact 1.1B-parameter Llama 2 architecture model on 3 trillion tokens, releasing checkpoints and training details. It includes evaluation results, fine-tuning scripts, and examples such as speculative decoding with llama.cpp.

## Use cases
- pretrain a small Llama-compatible language model
- run a compact LLM on limited compute or memory
- fine-tune a small base model for chat
- use a Llama 2 drop-in replacement with fewer parameters
- study intermediate checkpoints of an open LLM pretraining run
- speed up inference with speculative decoding

## When to choose
- you need a small, Llama 2-compatible model for resource-constrained environments
- you want an open base model to fine-tune cheaply
- you want to study or reproduce an open pretraining effort

## When to avoid
- you need state-of-the-art reasoning or broad knowledge from a larger model
- you want a ready-made production assistant rather than a base model
- you cannot host or fine-tune models yourself

## Facets
- artifact type: learning-resource
- maturity: stable
- function: llm-training, machine-learning, deep-learning
- domain: large-language-models, deep-learning, machine-learning
- platform: python
- tags: tinyllama, llama-2, pretraining, small-language-model, open-source-model, checkpoints, speculative-decoding, gpu, linux

## Member repositories
- jzhang38/TinyLlama (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:26.789576+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:24:20.262615+00:00, confidence not recorded.
  - readme: https://github.com/jzhang38/TinyLlama (fetched 2026-08-28T04:10:26.789576+00:00, sha 95dddbd5f0c7)
- Data as of 2026-08-30T08:39:29.467469+00:00.