Grand Quiz

🧠 Grand Quiz

1. 1. Qwen3.8-27B at Q4_K_M is roughly:

2. 2. Which loss curve shape describes quantization?

3. 3. In grouped quantization, the per-block scale lets the model:

4. 4. Calibrated quant methods (imatrix, AWQ, GPTQ) work by:

5. 5. Q5_K_M vs Q5_K_S — which is higher quality?

6. 6. Faced with legacy Q4_0 and Q4_K_M at similar sizes, pick:

7. 7. A 27B on a 16 GB GPU with room for context:

8. 8. imatrix and UD quants matter most when:

9. 9. AWQ means:

10. 10. nvfp4 achieves its speed benefit on:

11. 11. An abliterated model:

12. 12. A separate MTP file is used for:

13. Sanity-check math: a 27B model at 4.0 bpw ≈

14. 14. Which is a red flag on Hugging Face?

15. 15. The golden rule of local LLM choice:

16. 16. A 32 GB Mac should prefer: