# yangjianxin1/Firefly

Firefly: 大模型训练工具，支持训练Qwen2.5、Qwen2、Yi1.5、Phi-3、Llama3、Gemma、MiniCPM、Yi、Deepseek、Orion、Xverse、Mixtral-8x7B、Zephyr、Mistral、Baichuan2、Llma2、Llama、Qwen、Baichuan、ChatGLM2、InternLM、Ziya2、Vicuna、Bloom等大模型

Repository: https://github.com/yangjianxin1/Firefly
Canonical: https://ross.abutalabs.com/products/firefly
Language: Python
License Family: other
Topics: gpt, alpaca, llm, baichuan, llama, lora, qlora, peft, llama2, internlm, chatglm, qwen, aquila, mistral, mixtral, zephyr, minicpm, gemma, llama3, qwen2
Last push: 2024-10-24T02:27:42+00:00

## Health v2 (maintenance only)
Score: 21/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 0, release rhythm 8, longevity 89
- inputs: {"age_days": 1249, "days_push": 679, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 6653, forks 582 (observed 2026-08-28T04:09:47.365820+00:00)

## What it is
Firefly is an open-source one-stop training tool for large language models, supporting pretraining, instruction fine-tuning (SFT), and DPO with full-parameter, LoRA, and QLoRA methods. It supports many mainstream open-source LLMs such as Qwen2, Llama3, Gemma, Mistral, Mixtral, ChatGLM, and InternLM, with chat-template alignment and Unsloth acceleration.

## Use cases
- fine-tune llama3 with lora on a single gpu
- qlora instruction tuning for qwen2.5
- train a chat model with dpo
- pretrain a large language model on custom data
- fine-tune chatglm2 or internlm with sft
- reduce gpu memory usage when training llms with unsloth
- align training templates with official chat models

## When to choose
- you want to fine-tune or DPO-train mainstream open-source LLMs with LoRA/QLoRA on limited GPU resources
- you need config-file-driven training that matches each model's official chat template
- you want memory- and time-efficient training via Unsloth integration

## When to avoid
- you need a managed no-code training platform rather than Python scripts and config files
- you only want inference or serving of models rather than training
- you require a permissively licensed project - the repository has no declared license

## Facets
- artifact type: framework
- maturity: active
- function: llm-training, machine-learning, deep-learning
- domain: large-language-models, machine-learning, deep-learning
- platform: python
- tags: fine-tuning, lora, qlora, sft, dpo, pretraining, unsloth, peft, chinese-llm, gpu, linux

## Member repositories
- yangjianxin1/Firefly (main) score 21

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:47.365820+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 2026-08-29T17:43:07.284106+00:00, confidence not recorded.
  - readme: https://github.com/yangjianxin1/Firefly (fetched 2026-08-28T04:09:47.365820+00:00, sha baba99829c20)
- Data as of 2026-08-30T08:39:29.467469+00:00.