Compare LLM Models
Select 2-5 models to compare side by side.
Select models to compare
0 / 5 selected
Top downloads
Select 2-5 models
| Attribute |
ai21labs
|
|---|---|
| Company | ai21labs |
| Release Date | 2024-08-22 |
| Parameters (B) | 52 |
| Architecture | Hybrid Transformer-Mamba MoE |
| Context Window | 256000 |
| Input Modalities | text |
| Open Source | Yes |
| License | other |
| Score | 0 |
| VRAM (GB) | 160 |
| Compute | Minimum 2× NVIDIA A100 80GB GPUs for BF16 inference via vLLM or transformers; single A100 80GB with ExpertsInt8 quantization (vLLM ≥0.5.5); CUDA required for optimized Mamba kernels |
| Benchmark: ARC | 85.7 |
| Benchmark: BBH | 53.4 |
| Benchmark: GPQA | 32.3 |
| Benchmark: GSM8K | 75.8 |
| Benchmark: HumanEval | 62.8 |
| Benchmark: IFEval | 75.8 |
| Benchmark: MMLU | 69.7 |
| Benchmark: MMLU-Pro | 42.5 |
| Benchmark: TruthfulQA | 54.1 |
| Pricing (Input) | — |
| Pricing (Output) | — |