qwen36-twla-beam-lambda0p2

Local code references:

• TWLA/branch_bound_search.py
• TWLA/build_twla_expert_level_bank.py
• TWLA/datautils.py
• TWLA/inference.py
• TWLA/optimize_twla_hierarchical_nll.py
• TWLA/prepare_bipea_v3_data.py
• TWLA/quantize/E2M_ATQ.py
• TWLA/quantize/E2M_ATQ_10level.py
• TWLA/quantize/E2M_ATQ_11level.py
• TWLA/quantize/E2M_ATQ_4level.py
• TWLA/quantize/E2M_ATQ_5level.py
• TWLA/quantize/E2M_ATQ_6level.py
• TWLA/quantize/E2M_ATQ_7level.py
• TWLA/quantize/E2M_ATQ_8level.py
• TWLA/quantize/E2M_ATQ_9level.py
• TWLA/quantize/E2M_ATQ_bidirectional.py
• TWLA/quantize/E2M_ATQ_groupwise.py
• TWLA/quantize/E2M_ATQ_mixed_precision.py
• TWLA/quantize/__init__.py
• TWLA/quantize/const.py
• TWLA/quantize/e2m_atq_fwrd.py
• TWLA/quantize/e2m_atq_fwrd_multigpu.py
• TWLA/quantize/k_preprocessor_ternary.py
• TWLA/quantize/moe_round_worker.py
• TWLA/quantize/quantizer.py
• TWLA/quantize/ternary_kernel.py
• TWLA/quantize/ternary_pack.py
• TWLA/quantize_qwen36_experts_5level.py
• TWLA/requirements.txt
• TWLA/run_benchmark_qwen36_multilevel.py
• TWLA/taylor_fisher_proxy.py