[SomeNobody21112]

Overview

This adds two classes Q1_0 + Q2_0 to gguf-py/gguf/quants.py, so now numpy Q1_0 / Q2_0 support in quants.py, so gguf.dequantize / quantize work for them. Closes #30626.

Additional information

• Tested the Python code against the C code using the inbuilt script (test_quants.py) and got an exact match 66/66 across 27 types. So the new ones are right and nothing broke.
• pytest seven passed
• Q1_0's scale is computed with np.cumsum not np.sum, since C adds the 128 values one by one and np.sum adds them pairwise and float rounding then differs. np.sum gave a different scale than C in 18 of 65,536 blocks and np.cumsum gave zero.
• setup: Windows 11, gcc 16, Python 3.11, numpy 2.4

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES, AI was used to write the code and test runs but I manually reviewed the codes and verified them.