Llamafactory
App in the BluixApps catalog
What it is
LLaMA-Factory is a visual web UI for LLM training — config-builder for SFT, DPO, ORPO, PPO, GRPO, KTO across 100+ models (Llama, Mistral, Qwen, ChatGLM, Phi, Gemma, etc.). No-code fine-tuning made accessible, with integrated dataset conversion, training monitoring, and export tools.
The easiest entry to LLM fine-tuning — Axolotl's UI counterpart.
What it's for
- Visual training configuration — no YAML editing
- Multi-stage training — SFT, DPO, ORPO, PPO, GRPO, KTO
- Dataset conversion — built-in format adapters
- Training monitoring — loss curves + eval metrics live
- Model export — merged weights or adapter files
- 100+ models supported — broadest coverage in OSS training space
Who it's for
- Non-engineers learning LLM fine-tuning
- AI agency teams offering managed fine-tuning to clients
- Educators teaching LLM training fundamentals
- Researchers needing rapid iteration
- Hosting providers offering visual fine-tuning tier
Why teams pick LLaMA-Factory over alternatives
- Apache 2.0 — fully open
- Easiest UX for LLM training in OSS space
- Broadest model coverage — 100+ models supported out-of-box
- All major training stages — SFT + alignment (DPO/ORPO) + RL (PPO/GRPO)
- Live monitoring — loss curves, eval scores
- Visual everything — model selection, dataset preview, hyperparameter tuning, export
- Active community — 30k+ GitHub stars
Integrations
- HuggingFace Transformers + Datasets + PEFT + TRL underneath
- DeepSpeed for multi-GPU
- bitsandbytes for quantization
- WandB / TensorBoard for monitoring
- Pair with: vLLM/TGI to serve fine-tuned model
Notable users & community
- 38k+ GitHub stars (one of most popular LLM training tools)
- hiyouga + extensive contributor base
- Used widely in academic + commercial fine-tuning
- Multiple tutorials + courses
- Active Discord + GitHub community
Tips & operations
- VRAM:
- 7B QLoRA: 16 GB
- 7B LoRA: 24 GB
- 13B QLoRA: 24 GB
- 70B QLoRA: 80+ GB
- Training stages:
- SFT (Supervised Fine-Tuning) — most common
- DPO / ORPO / SimPO / CPO — preference alignment
- PPO / GRPO — reinforcement learning (RLHF / R1-style)
- KTO — Kahneman-Tversky Optimization
- Pre-training — continued pretraining on new domain
- Dataset preview: built-in viewer for sharegpt / alpaca / openai format
- Multi-GPU: enable in advanced tab; DeepSpeed ZeRO 2 or 3
- Export: merged model OR LoRA adapter
- vs Axolotl: LLaMA-Factory = visual UI; Axolotl = config-driven YAML
- vs Unsloth: LLaMA-Factory = UI; Unsloth = code library (faster but no UI)
What we ship in BluixApps
- Cloned
hiyouga/LLaMA-Factoryrepo - pytorch CUDA 12.4 devel base + git pre-installed
- Pip install with
[torch,metrics]extras + gradio llamafactory-cli webuilauncher (Gradio server on port 7860)- Persistent volumes: repo, data (training datasets), saves (output), cache (HF)
- Port 7881 mapped
- HF_TOKEN environment variable for gated models
- Install report at
/root/bluixapps/llamafactory.txt - Training stage guidance
- Quick-start workflow documentation
- LLaMA-Factory vs Axolotl vs Unsloth comparison
- Pairing notes (vLLM/TGI for serving)
- GPU pre-flight check via
bluixapps_ensure_nvidia_runtime - Backup hook covers data + saves
Get this app — pick a BluixApps plan
Same catalog. Scaling tenant isolation, white-label and support tier.
| Tier | Tenants | Catalog | Support | White-label | Monthly | |
|---|---|---|---|---|---|---|
| Stacks | 1 | 19 curated stacks | Standard | — | $19/mo | DetailDeploy |
| Starter | 10 | Full catalog | Standard | +$15–25/mo | $49/mo | DetailDeploy |
| Pro | 25 | Full catalog | Priority bugfix | +$15–25/mo | $149/mo | DetailDeploy |
| Growth | 100 | Full catalog | Priority bugfix | +$15–25/mo | $349/mo | DetailDeploy |
| Scale | 500 | Full catalog | 7-day window | +$15–25/mo | $799/mo | DetailDeploy |
| Enterprise | Unlimited | Full catalog | Priority 7-day | Bundled | $1,499/mo | DetailDeploy |
