{ "schema": "dynquant_allocation_v1", "dynquant_core": "0.5.3", "model": "runs/ft/model", "stats": "runs/ft/stats", "allocator": "sensitivity", "group_size": 128, "map_key": "4.25", "map": { "target_label": "4.25 avg bits", "average_bits": 4.247889911339871, "nbytes": 898002432, "group_size": 128, "pricing": { "measured_modules": 133, "proxied_modules": 72, "measured_params": 331743232, "proxied_params": 1358954496, "proxied_share": 0.803783239010776, "scale": 1.1908850493541702e-17 }, "histogram": { "2": 9, "3": 27, "4": 119, "8": 50 }, "violations": [], "bits": { "model.embed_tokens": 8, "model.layers.0.conv.in_proj": 8, "model.layers.0.conv.out_proj": 8, "model.layers.0.moe.sw1": 4, "model.layers.0.moe.sw2": 8, "model.layers.0.moe.sw3": 8, "model.layers.0.moe.w1": 4, "model.layers.0.moe.w2": 2, "model.layers.0.moe.w3": 4, "model.layers.1.conv.in_proj": 4, "model.layers.1.conv.out_proj": 4, "model.layers.1.moe.sw1": 4, "model.layers.1.moe.sw2": 4, "model.layers.1.moe.sw3": 4, "model.layers.1.moe.w1": 4, "model.layers.1.moe.w2": 2, "model.layers.1.moe.w3": 2, "model.layers.10.conv.in_proj": 4, "model.layers.10.conv.out_proj": 4, "model.layers.10.moe.sw1": 4, "model.layers.10.moe.sw2": 8, "model.layers.10.moe.sw3": 8, "model.layers.10.moe.w1": 4, "model.layers.10.moe.w2": 4, "model.layers.10.moe.w3": 3, "model.layers.11.moe.sw1": 4, "model.layers.11.moe.sw2": 4, "model.layers.11.moe.sw3": 4, "model.layers.11.moe.w1": 4, "model.layers.11.moe.w2": 4, "model.layers.11.moe.w3": 4, "model.layers.11.self_attn.k_proj": 8, "model.layers.11.self_attn.o_proj": 8, "model.layers.11.self_attn.q_proj": 8, "model.layers.11.self_attn.v_proj": 8, "model.layers.12.conv.in_proj": 4, "model.layers.12.conv.out_proj": 4, "model.layers.12.moe.sw1": 4, "model.layers.12.moe.sw2": 4, "model.layers.12.moe.sw3": 4, "model.layers.12.moe.w1": 4, "model.layers.12.moe.w2": 3, "model.layers.12.moe.w3": 3, "model.layers.13.conv.in_proj": 4, "model.layers.13.conv.out_proj": 8, "model.layers.13.moe.sw1": 4, "model.layers.13.moe.sw2": 4, "model.layers.13.moe.sw3": 4, "model.layers.13.moe.w1": 4, "model.layers.13.moe.w2": 4, "model.layers.13.moe.w3": 4, "model.layers.14.conv.in_proj": 8, "model.layers.14.conv.out_proj": 8, "model.layers.14.moe.sw1": 4, "model.layers.14.moe.sw2": 8, "model.layers.14.moe.sw3": 8, "model.layers.14.moe.w1": 4, "model.layers.14.moe.w2": 3, "model.layers.14.moe.w3": 4, "model.layers.15.moe.sw1": 4, "model.layers.15.moe.sw2": 4, "model.layers.15.moe.sw3": 4, "model.layers.15.moe.w1": 4, "model.layers.15.moe.w2": 4, "model.layers.15.moe.w3": 3, "model.layers.15.self_attn.k_proj": 8, "model.layers.15.self_attn.o_proj": 8, "model.layers.15.self_attn.q_proj": 8, "model.layers.15.self_attn.v_proj": 8, "model.layers.16.conv.in_proj": 8, "model.layers.16.conv.out_proj": 8, "model.layers.16.moe.sw1": 4, "model.layers.16.moe.sw2": 4, "model.layers.16.moe.sw3": 4, "model.layers.16.moe.w1": 4, "model.layers.16.moe.w2": 3, "model.layers.16.moe.w3": 3, "model.layers.17.conv.in_proj": 4, "model.layers.17.conv.out_proj": 4, "model.layers.17.moe.sw1": 4, "model.layers.17.moe.sw2": 4, "model.layers.17.moe.sw3": 4, "model.layers.17.moe.w1": 4, "model.layers.17.moe.w2": 3, "model.layers.17.moe.w3": 3, "model.layers.18.conv.in_proj": 8, "model.layers.18.conv.out_proj": 8, "model.layers.18.moe.sw1": 4, "model.layers.18.moe.sw2": 8, "model.layers.18.moe.sw3": 8, "model.layers.18.moe.w1": 4, "model.layers.18.moe.w2": 3, "model.layers.18.moe.w3": 3, "model.layers.19.moe.sw1": 4, "model.layers.19.moe.sw2": 8, "model.layers.19.moe.sw3": 4, "model.layers.19.moe.w1": 4, "model.layers.19.moe.w2": 3, "model.layers.19.moe.w3": 3, "model.layers.19.self_attn.k_proj": 8, "model.layers.19.self_attn.o_proj": 8, "model.layers.19.self_attn.q_proj": 8, "model.layers.19.self_attn.v_proj": 8, "model.layers.2.conv.in_proj": 4, "model.layers.2.conv.out_proj": 4, "model.layers.2.moe.sw1": 4, "model.layers.2.moe.sw2": 4, "model.layers.2.moe.sw3": 4, "model.layers.2.moe.w1": 4, "model.layers.2.moe.w2": 2, "model.layers.2.moe.w3": 2, "model.layers.20.conv.in_proj": 4, "model.layers.20.conv.out_proj": 4, "model.layers.20.moe.sw1": 4, "model.layers.20.moe.sw2": 4, "model.layers.20.moe.sw3": 4, "model.layers.20.moe.w1": 4, "model.layers.20.moe.w2": 2, "model.layers.20.moe.w3": 2, "model.layers.21.conv.in_proj": 4, "model.layers.21.conv.out_proj": 4, "model.layers.21.moe.sw1": 4, "model.layers.21.moe.sw2": 4, "model.layers.21.moe.sw3": 4, "model.layers.21.moe.w1": 4, "model.layers.21.moe.w2": 3, "model.layers.21.moe.w3": 3, "model.layers.22.conv.in_proj": 4, "model.layers.22.conv.out_proj": 4, "model.layers.22.moe.sw1": 4, "model.layers.22.moe.sw2": 8, "model.layers.22.moe.sw3": 4, "model.layers.22.moe.w1": 4, "model.layers.22.moe.w2": 3, "model.layers.22.moe.w3": 3, "model.layers.23.moe.sw1": 8, "model.layers.23.moe.sw2": 8, "model.layers.23.moe.sw3": 8, "model.layers.23.moe.w1": 4, "model.layers.23.moe.w2": 3, "model.layers.23.moe.w3": 3, "model.layers.23.self_attn.k_proj": 8, "model.layers.23.self_attn.o_proj": 4, "model.layers.23.self_attn.q_proj": 8, "model.layers.23.self_attn.v_proj": 8, "model.layers.3.moe.sw1": 4, "model.layers.3.moe.sw2": 4, "model.layers.3.moe.sw3": 4, "model.layers.3.moe.w1": 4, "model.layers.3.moe.w2": 3, "model.layers.3.moe.w3": 2, "model.layers.3.self_attn.k_proj": 8, "model.layers.3.self_attn.o_proj": 8, "model.layers.3.self_attn.q_proj": 8, "model.layers.3.self_attn.v_proj": 8, "model.layers.4.conv.in_proj": 4, "model.layers.4.conv.out_proj": 4, "model.layers.4.moe.sw1": 4, "model.layers.4.moe.sw2": 4, "model.layers.4.moe.sw3": 4, "model.layers.4.moe.w1": 4, "model.layers.4.moe.w2": 3, "model.layers.4.moe.w3": 3, "model.layers.5.conv.in_proj": 4, "model.layers.5.conv.out_proj": 4, "model.layers.5.moe.sw1": 4, "model.layers.5.moe.sw2": 4, "model.layers.5.moe.sw3": 4, "model.layers.5.moe.w1": 4, "model.layers.5.moe.w2": 3, "model.layers.5.moe.w3": 3, "model.layers.6.conv.in_proj": 8, "model.layers.6.conv.out_proj": 8, "model.layers.6.moe.sw1": 4, "model.layers.6.moe.sw2": 8, "model.layers.6.moe.sw3": 4, "model.layers.6.moe.w1": 4, "model.layers.6.moe.w2": 2, "model.layers.6.moe.w3": 3, "model.layers.7.moe.sw1": 4, "model.layers.7.moe.sw2": 8, "model.layers.7.moe.sw3": 4, "model.layers.7.moe.w1": 4, "model.layers.7.moe.w2": 3, "model.layers.7.moe.w3": 3, "model.layers.7.self_attn.k_proj": 8, "model.layers.7.self_attn.o_proj": 8, "model.layers.7.self_attn.q_proj": 8, "model.layers.7.self_attn.v_proj": 8, "model.layers.8.conv.in_proj": 4, "model.layers.8.conv.out_proj": 4, "model.layers.8.moe.sw1": 4, "model.layers.8.moe.sw2": 4, "model.layers.8.moe.sw3": 4, "model.layers.8.moe.w1": 4, "model.layers.8.moe.w2": 4, "model.layers.8.moe.w3": 4, "model.layers.9.conv.in_proj": 4, "model.layers.9.conv.out_proj": 4, "model.layers.9.moe.sw1": 4, "model.layers.9.moe.sw2": 4, "model.layers.9.moe.sw3": 4, "model.layers.9.moe.w1": 4, "model.layers.9.moe.w2": 4, "model.layers.9.moe.w3": 4 } } }