4-bit Autotrain compatible Base model:moonshotai/moonligh... Base model:quantized:moonshota... Conversational Custom code Deepseek v3 Endpoints compatible Instruct Mlx Region:us Safetensors Sharded Tensorflow
Model Card on HF 🤗: https://huggingface.co/mlx-community/Moonlight-16B-A3B-Instruct-4-bit
Moonlight 16B A3B Instruct 4 Bit Benchmarks
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
What is the LLM Explorer Rank (Score)
Moonlight 16B A3B Instruct 4 Bit Parameters and Internals
| LLM Name | Moonlight 16B A3B Instruct 4 Bit |
| Repository 🤗 | https://huggingface.co/mlx-community/Moonlight-16B-A3B-Instruct-4-bit |
| Base Model(s) | |
| Model Size | 16b |
| Required VRAM | 9 GB |
| Updated | 2025-07-15 |
| Maintainer | mlx-community |
| Model Type | deepseek_v3 |
| Instruction-Based | Yes |
| Model Files | |
| Model Architecture | DeepseekV3ForCausalLM |
| License | mit |
| Context Length | 8192 |
| Model Max Length | 8192 |
| Transformers Version | 4.46.3 |
| Tokenizer Class | TikTokenTokenizer |
| Padding Token | [PAD] |
| Vocabulary Size | 163840 |
| Torch Data Type | bfloat16 |
Best Alternatives to Moonlight 16B A3B Instruct 4 Bit
Best Alternatives |
Context / RAM |
Downloads |
Likes |
|---|---|---|---|
| Moonlight 16B A3B Instruct | 8K / 32.6 GB | 9147 | 168 |
| ...3B Instruct Abliterated Pruned | 8K / 17.5 GB | 9 | 5 |
| ...t 16B A3B Instruct Abliterated | 8K / 31.9 GB | 9 | 8 |
Note: green Score (e.g. "73.2") means that the model is better than mlx-community/Moonlight-16B-A3B-Instruct-4-bit .
Rank the Moonlight 16B A3B Instruct 4 Bit Capabilities
🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟
What open-source LLMs or SLMs are you in search of? 49755 in total.
Original data from HuggingFace, OpenCompass and various public git repos.
Release v20241124