Cosmos-Predict2-2B-Text2Image
nvidia/Cosmos-Predict2-2B-Text2ImageCosmos-Predict2-2B-Text2Image at Q4_K_M is exactly 16,049,921,632 bytes (14.95 GiB / 16.05 GB) — an effective 9.001 bits per weight, not the nominal 4.
Shipped quantizations
| Quant | Size● | Exact bytes● | Effective bpw● | Tensors● | Publisher |
|---|---|---|---|---|---|
| Q3_K_M8 shards | 11.52 GiB | 12,366,188,640 | 6.935 | — | calcuis |
| Q4_K_M8 shards | 14.95 GiB | 16,049,921,632 | 9.001 | — | calcuis |
| Q5_K_M8 shards | 18.18 GiB | 19,516,964,448 | 10.945 | — | calcuis |
| Q2_K12 shards | 32.34 GiB | 34,729,181,536 | 19.476 | — | calcuis |
| Q4_110 shards | 47.99 GiB | 51,533,681,440 | 28.900 | — | calcuis |
| Q5_110 shards | 57.37 GiB | 61,595,874,080 | — | — | calcuis |
| Q4_017 shards | 65.28 GiB | 70,090,148,192 | — | — | calcuis |
| Q6_K12 shards | 65.66 GiB | 70,505,718,624 | — | — | calcuis |
| Q5_016 shards | 76.30 GiB | 81,925,395,520 | — | — | calcuis |
| Q8_018 shards | 120.91 GiB | 129,830,502,016 | — | — | calcuis |
Pipeline components
| Component | Size | Share | Can live on the CPU? |
|---|---|---|---|
| text encoder | 9.06 GiB | 69% | yes — the usual first VRAM saving |
| vae | 0.47 GiB | 4% | no, must be resident |
| denoiser | 3.64 GiB | 28% | no, must be resident |
| Full pipeline | 13.18 GiB | resident if nothing is offloaded | |
The parameter count published for a diffusion model describes the denoiser alone. Running it also requires its text encoder and VAE, and the text encoder is often nearly as large as the denoiser — which is why offloading it is the standard first move when you run out of memory.
We publish component sizes here, not throughput. Community-submitted image-generation rates do exist for many GPUs and we show them on the hardware pages, but they aggregate runs at different resolutions, step counts and settings, so they cannot be attributed to one model. Peak memory during sampling is unmeasured by any public source, and we do not estimate it.
Compare with
Will it run on your card?
Why other calculators give a different number
A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 7.47 GiB. The real file is 14.95 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.
Architecture
Architecture unavailable — this repository is gated and no ungated mirror was found. Exact file sizes above are still authoritative; only the KV math needs the config.
Questions people ask
- How much VRAM does Cosmos-Predict2-2B-Text2Image need?
- Q4_K_M is exactly 16,049,921,632 bytes (14.95 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
- Which quantization of Cosmos-Predict2-2B-Text2Image should I use?
- Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.