Model comparison
internlm3-8b-instruct vs Qwen3-8B
At Q4_K_M, Qwen3-8B is the smaller download — 5,027,783,488 bytes against 5,358,623,936. At long context the gap widens: internlm3-8b-instruct's KV cache at 32K is 3.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| internlm3-8b-instruct | Qwen3-8B | |
|---|---|---|
| Parameters | 8.8B | 8.2B |
| Architecture | llama | qwen3 |
| Layers | 48 | 36 |
| Native context | 32,768 | 40,960 |
| Mixture of experts | no | no |
| Quantizations published | 35 | 51 |
| Smallest quantization | 2.98 GiB | 2.12 GiB |
| Q4_K_M | 4.99 GiB | 4.68 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | internlm3-8b-instruct | Qwen3-8B | Ratio |
|---|---|---|---|
| 4,096 | 0.19 GiB | 0.56 GiB | 3.00× |
| 8,192 | 0.38 GiB | 1.13 GiB | 3.00× |
| 16,384 | 0.75 GiB | 2.25 GiB | 3.00× |
| 32,768 | 1.50 GiB | 4.50 GiB | 3.00× |
| 65,536 | 3.00 GiB | 9.00 GiB | 3.00× |
| 131,072 | 6.00 GiB | 18.00 GiB | 3.00× |