Model comparison
SmolVLM-256M-Instruct vs embeddinggemma-300m
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: embeddinggemma-300m's KV cache at 32K is 4.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| SmolVLM-256M-Instruct | embeddinggemma-300m | |
|---|---|---|
| Parameters | 256M | 303M |
| Architecture | llama | gemma-embedding |
| Layers | 30 | 24 |
| Native context | 8,192 | 2,048 |
| Mixture of experts | no | no |
| Quantizations published | 2 | 10 |
| Smallest quantization | 0.16 GiB | 0.26 GiB |
| Q4_K_M | — | — |
| Licence | apache-2.0 | — |
KV cache by context
the term that decides long-context viability
| Context | SmolVLM-256M-Instruct | embeddinggemma-300m | Ratio |
|---|---|---|---|
| 4,096 | 0.09 GiB | 0.04 GiB | 2.50× |
| 8,192 | 0.18 GiB | 0.05 GiB | 3.46× |
| 16,384 | 0.35 GiB | 0.08 GiB | 4.29× |
| 32,768 | 0.70 GiB | 0.14 GiB | 4.86× |
| 65,536 | 1.41 GiB | 0.27 GiB | 5.22× |
| 131,072 | 2.81 GiB | 0.52 GiB | 5.41× |