Model comparison
GLM-OCR 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 13.8× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| GLM-OCR | embeddinggemma-300m | |
|---|---|---|
| Parameters | 1.3B | 303M |
| Architecture | glm4 | gemma-embedding |
| Layers | 16 | 24 |
| Native context | 131,072 | 2,048 |
| Mixture of experts | no | no |
| Quantizations published | 30 | 10 |
| Smallest quantization | 0.34 GiB | 0.26 GiB |
| Q4_K_M | 0.51 GiB | — |
| Licence | — | — |
KV cache by context
the term that decides long-context viability
| Context | GLM-OCR | embeddinggemma-300m | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | 0.04 GiB | 7.11× |
| 8,192 | 0.50 GiB | 0.05 GiB | 9.85× |
| 16,384 | 1.00 GiB | 0.08 GiB | 12.19× |
| 32,768 | 2.00 GiB | 0.14 GiB | 13.84× |
| 65,536 | 4.00 GiB | 0.27 GiB | 14.84× |
| 131,072 | 8.00 GiB | 0.52 GiB | 15.40× |