ace-step-1.5 vs gliner
Two of the top projects in Open models, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
ace-step-1.5
Text-to-music model with LM planner, DiT decoder, Gradio UI and REST API
gliner
Small zero-shot named entity recognition models with training and serving code
Verdict
ace-step-1.5 ranks higher, 73 to 67. Of the 4 score parts we measure for both, ace-step-1.5 leads on 2, gliner on 1 and 1 is even.
- Adoptionace-step-1.5 leads by 40 points, 79 to 39: a wide gap.
- FreshnessEven at 100.
- Maintenancegliner leads by 34 points, 83 to 49: a wide gap.
- Agent-readyace-step-1.5 leads by 45 points, 45 to 0: a wide gap.
Pick ace-step-1.5 if you are
Musicians and developers generating music locally, with LoRA fine-tuning
- Runs locally with under 4GB VRAM using the DiT-only turbo model; README claims under 10s per song on RTX 3090
- Covers text-to-music, cover, repaint, stem separation, and vocal-to-BGM in one model family
Pick gliner if you are
Engineers needing local entity extraction or PII redaction without an LLM
- Arbitrary entity labels at inference time, no labeled data needed
- Runs on CPU; FP16, INT8, torch.compile and ONNX export available
| What we compare | ace-step-1.5 | gliner |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 79, popular | 39, known |
| Freshness | 100, active | 100, active |
| Maintenance | 49, patchy | 83, healthy |
| Agent-ready | 45, minimal | 0, none |
| Facts from GitHub and the README | ||
| Stars | 15k | 4.3k |
| License | MIT (permissive) | Apache-2.0 (permissive) |
| Last commit | Aug 2026 | Nov 2026 |
| Last release | May 2026 | Nov 2026 |
| Language | Python | Python |
| Docker | Yes | Yes |
| GPU | Optional | Optional |
| arm64 or Apple Silicon | Mentioned | Not stated |
ace-step-1.5
ACE-Step 1.5 generates full songs from text prompts and lyrics using a Qwen3-based language model that plans the song and a diffusion transformer (2B or 4B XL) that renders audio. It also does cover generation, repainting, stem separation, vocal-to-accompaniment and LoRA training. It ships a Gradio UI on port 7860, a REST API on port 8001, a CLI and a Python API.
Who it is for: Musicians and developers generating music locally, with LoRA fine-tuning
Strengths
- Runs locally with under 4GB VRAM using the DiT-only turbo model; README claims under 10s per song on RTX 3090
- Covers text-to-music, cover, repaint, stem separation, and vocal-to-BGM in one model family
- LoRA training from a few songs: 8 songs in about 1 hour on a 12GB 3090, from the Gradio UI
- Several interfaces: Gradio UI, REST API, CLI, Python API, and a separate VST3 plugin
Weaknesses
- XL 4B models need at least 12GB VRAM with offload and quantization, 20GB for comfortable use
- Low-VRAM setups (6GB or less) run without the LM, losing planning and query rewriting
- Quality claims (between Suno v4.5 and v5) are the authors' own, with no numbers in the README
- Python 3.11-3.12 only; CUDA recommended, other backends (MPS, ROCm, XPU, CPU) are secondary
- GPU optional
- Docker + Compose
- Needs uv, Python 3.11-3.12, Gradio
- Models: ACE-Step DiT 2B (base, sft, turbo), ACE-Step XL DiT 4B (base, sft, turbo), acestep-5Hz-lm 0.6B/1.7B/4B (Qwen3-based), LoRA adapters
- port 7860
gliner
GLiNER is a Python framework for training and running small transformer models that extract entities of any user-supplied label from text, without task-specific training. Variants cover bi-encoder label scaling, joint entity and relation extraction, streaming NER and PII detection. It runs on CPU, supports FP16/INT8 quantization, torch.compile and ONNX export, and includes a Ray Serve HTTP server.
Who it is for: Engineers needing local entity extraction or PII redaction without an LLM
Strengths
- Arbitrary entity labels at inference time, no labeled data needed
- Runs on CPU; FP16, INT8, torch.compile and ONNX export available
- Several architectures: uni-encoder, bi-encoder, relation extraction, decoder, streaming
- Includes fine-tuning script and Ray Serve server with dynamic batching
Weaknesses
- Uni-encoder handles only about 50 entity types; more needs the bi-encoder
- INT8 accuracy requires models trained with quantization-aware training
- Classification, entity linking and schema extraction live in separate projects
- Serving layer depends on Ray Serve, an extra install
- GPU optional
- Compose
- Needs PyTorch, Ray Serve (optional, for serving)
- Models: GLiNER uni-encoder, GLiNER bi-encoder, GLiNER RelEx, GLiNER Decoder, StreamingSpan
- port 8000
More in Open models
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.