Model reference · open weights
Underdog-Woof-1.1 is an open-weight language model from ConwayResearch. Built for Apple silicon (MLX). It does not run on NVIDIA GPUs.
What it is
| Released by | ConwayResearch |
|---|---|
| Released | 2026-09-12 |
| Parameters | 4.2B |
| VRAM | 2.4 GB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory | |
|---|---|---|---|---|
| 8K each | 32K each | |||
| RTX 3060 12 GB | 22 | 7 | 248K | 11.6 GB |
| RTX 4060 Ti 16 GB | 32 | 10 | all 256K | 15.4 GB |
| RTX 3090 24 GB | 54 | 17 | all 256K | 23.4 GB |
| RTX 4090 24 GB | 54 | 17 | all 256K | 23.4 GB |
| RTX 5090 32 GB | 74 | 23 | all 256K | 31.0 GB |
| L40S 48 GB | 109 | 34 | all 256K | 44.0 GB |
| A100 80 GB | 201 | 63 | all 256K | 78.2 GB |
| H100 80 GB | 187 | 59 | all 256K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 229 | 72 | all 256K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 266 | 84 | all 256K | 107 GB |
| H200 141 GB | 348 | 110 | all 256K | 138 GB |
| B200 180 GB | 450 | 142 | all 256K | 176 GB |
| Requests at once | 8K tokens each | 32K tokens each |
|---|---|---|
| 1 | 3.5 GB | 4.3 GB |
| 5 | 5.0 GB | 9.1 GB |
| 8 | 6.1 GB | 12.6 GB |
| 16 | 9.1 GB | 22.0 GB |
| 32 | 15.1 GB | 40.8 GB |
| 64 | 26.9 GB | 78.5 GB |
One card, with vLLM's small-card settings.
From the model card
Woof is a small but mighty Underdog model.
Quantized to 4-bit MLX.
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.
Running it
Built for Apple silicon (MLX). It does not run on NVIDIA GPUs.