| """HPC-Quantize V2: factored Q2_K quantization in Python. |
| |
| The package keeps the full HPC decision stack (batched GGML seeds, WLS |
| superblock solve, factored candidate field, sequential Sieve, 36-state |
| Viterbi, spectral/DC shaping, local-safety floors) while replacing the |
| original factored solver with a monotone, multi-start, fp16-exact version. |
| """ |
|
|
| from .analog import ( |
| build_block_analogs, |
| subblock_isomoment_analog, |
| subscale_lattice_analog, |
| vesica_gram_analog, |
| ) |
| from .config import HPCConfig |
| from .cuda_backend import cuda_device_name, cuda_is_available, quantize_cuda |
| from .factored import ( |
| FactorResult, |
| factor_1d, |
| factor_candidates, |
| factor_only_solve, |
| factor_q2k_block, |
| ) |
| from .gguf import ( |
| GGUFReader, |
| expand_importance, |
| is_q1_0_eligible, |
| is_q2k_eligible, |
| is_q3_k_eligible, |
| is_q4_0_eligible, |
| lookup_imatrix, |
| read_imatrix, |
| target_quant_type, |
| ) |
| from .pipeline import Q2KQuantizer, QuantizedTensor, TensorStats |
| from .q1_0 import Q1_0_BLOCK_BYTES, QK1_0, dequantize_q1_0_bytes, quantize_tensor_q1_0 |
| from .q2k import Q2KBlock, QK_K, dequantize_params, pack_codes, unpack_codes |
| from .q3_k import Q3_K_BLOCK_BYTES, QK3_K, dequantize_q3_k_bytes, quantize_tensor_q3_k |
| from .q4_0 import QK4_0, dequantize_q4_0_bytes, quantize_tensor_q4_0 |
|
|
| __all__ = [ |
| "HPCConfig", |
| "cuda_device_name", |
| "cuda_is_available", |
| "quantize_cuda", |
| "FactorResult", |
| "Q2KBlock", |
| "Q2KQuantizer", |
| "QuantizedTensor", |
| "TensorStats", |
| "QK_K", |
| "build_block_analogs", |
| "subblock_isomoment_analog", |
| "subscale_lattice_analog", |
| "vesica_gram_analog", |
| "dequantize_params", |
| "factor_1d", |
| "factor_candidates", |
| "factor_only_solve", |
| "factor_q2k_block", |
| "GGUFReader", |
| "expand_importance", |
| "is_q1_0_eligible", |
| "is_q2k_eligible", |
| "is_q3_k_eligible", |
| "is_q4_0_eligible", |
| "lookup_imatrix", |
| "read_imatrix", |
| "target_quant_type", |
| "QK1_0", |
| "Q1_0_BLOCK_BYTES", |
| "dequantize_q1_0_bytes", |
| "quantize_tensor_q1_0", |
| "QK3_K", |
| "Q3_K_BLOCK_BYTES", |
| "dequantize_q3_k_bytes", |
| "quantize_tensor_q3_k", |
| "QK4_0", |
| "dequantize_q4_0_bytes", |
| "quantize_tensor_q4_0", |
| "pack_codes", |
| "unpack_codes", |
| ] |
|
|
