mammoth-1 / elephant-1 — local AI model by Base Labs (Baseten)
Written by Jakub Rusinowski · Last updated
Base Labs' MIT-licensed decision models: full fine-tunes of Qwen3.5 on 6.21 million decision examples. mammoth-1 (27B, text) is third of 228 on the Jev Decision Index 0.4.1; elephant-1 (9B) keeps Qwen3.5's frozen vision encoder. Served by Base Labs' own CUDA runtime, not by llama.cpp or Ollama.
Variants
The smallest mammoth-1 / elephant-1 variant needs about 6 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| mammoth-1 27B → 27B | ~17.3 GB | 16K (16,384) | Base Labs runtime (OneJev + serve.patch) | |
| elephant-1 9B → 9B | ~6.5 GB | 16K (16,384) | Base Labs runtime (OneJev + serve.patch) |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run mammoth-1 / elephant-1 locally
Install Ollama, then pull the tag.
Served by Base Labs runtime (OneJev + serve.patch). This model does not run in Ollama.
Pick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: mammoth-1 27B, elephant-1 9BRecommended GPU
The cheapest catalogued GPU that runs mammoth-1 / elephant-1 locally (min 6 GB VRAM) is the Intel Arc B570 (10 GB).
mammoth-1 / elephant-1 — frequently asked questions
How much VRAM does mammoth-1 / elephant-1 need?
mammoth-1 / elephant-1 needs about 6 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: mammoth-1 27B (17 GB, Q4_K_M); elephant-1 9B (6 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run mammoth-1 / elephant-1 on an RTX 4090 (24 GB)?
Yes — mammoth-1 / elephant-1 runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
What quantization should I use for mammoth-1 / elephant-1?
Q4_K_M is the best balance of quality and VRAM for mammoth-1 / elephant-1 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run mammoth-1 / elephant-1 locally?
mammoth-1 / elephant-1 is a decision model: it is called over an HTTP endpoint (/v1/systemone), not chatted with. Where a variant is available in Ollama 0.35 or newer, pull it with `ollama pull` and send requests to the local server; the others ship their own server. Each variant page shows the exact commands for that model.