{ "average_bits": 3.4999670749374423, "embed_bits": 8, "embed_uses_tied_head": false, "embed_reason": "The decision readout is last_hidden @ readout.T and does not use the tied lm_head. The embedding is input-only and is stored as 8-bit affine, group 64.", "embed_sum_kl": { "4": 0.055970916987464966, "8": 0.010904520401237319 }, "embed_bytes": { "4": 143032320, "8": 270172160 }, "joint_refine": { "calibration_n": 0, "calibration_ids": [], "joint_kl_before": null, "joint_kl_after": null, "swaps": 0 }, "assignment": [ { "path": "language_model.model.layers.0.linear_attn.in_proj_qkv", "layer": 0, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.04734761182665103 }, { "path": "language_model.model.layers.0.linear_attn.in_proj_z", "layer": 0, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.1388501915724426 }, { "path": "language_model.model.layers.0.linear_attn.in_proj_b", "layer": 0, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.01794408019266606 }, { "path": "language_model.model.layers.0.linear_attn.in_proj_a", "layer": 0, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.011710990232290872 }, { "path": "language_model.model.layers.0.linear_attn.out_proj", "layer": 0, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.4239300562384638 }, { "path": "language_model.model.layers.0.mlp.gate_proj", "layer": 0, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.06092327188142901 }, { "path": "language_model.model.layers.0.mlp.down_proj", "layer": 0, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.02344921252365501 }, { "path": "language_model.model.layers.0.mlp.up_proj", "layer": 0, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.027068575012547745 }, { "path": "language_model.model.layers.1.linear_attn.in_proj_qkv", "layer": 1, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.023629120959874177 }, { "path": "language_model.model.layers.1.linear_attn.in_proj_z", "layer": 1, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.025880121687977955 }, { "path": "language_model.model.layers.1.linear_attn.in_proj_b", "layer": 1, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.01083456115500153 }, { "path": "language_model.model.layers.1.linear_attn.in_proj_a", "layer": 1, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.01273320237789333 }, { "path": "language_model.model.layers.1.linear_attn.out_proj", "layer": 1, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.0927767372959222 }, { "path": "language_model.model.layers.1.mlp.gate_proj", "layer": 1, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.021329556389151568 }, { "path": "language_model.model.layers.1.mlp.down_proj", "layer": 1, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.02737624581982894 }, { "path": "language_model.model.layers.1.mlp.up_proj", "layer": 1, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.037710730514948596 }, { "path": "language_model.model.layers.2.linear_attn.in_proj_qkv", "layer": 2, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.05043157109883073 }, { "path": "language_model.model.layers.2.linear_attn.in_proj_z", "layer": 2, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.03697225452164565 }, { "path": "language_model.model.layers.2.linear_attn.in_proj_b", "layer": 2, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.013773699583866855 }, { "path": "language_model.model.layers.2.linear_attn.in_proj_a", "layer": 2, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.011592477724978621 }, { "path": "language_model.model.layers.2.linear_attn.out_proj", "layer": 2, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.09072199036986346 }, { "path": "language_model.model.layers.2.mlp.gate_proj", "layer": 2, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.04075199032161302 }, { "path": "language_model.model.layers.2.mlp.down_proj", "layer": 2, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.057830560223890416 }, { "path": "language_model.model.layers.2.mlp.up_proj", "layer": 2, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.03367332028470472 }, { "path": "language_model.model.layers.3.self_attn.q_proj", "layer": 3, "layer_type": "full_attention", "nparams": 4194304, "bits": 4, "kind": "4", "kl": 0.0363731313499352 }, { "path": "language_model.model.layers.3.self_attn.k_proj", "layer": 3, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.017387628331155 }, { "path": "language_model.model.layers.3.self_attn.v_proj", "layer": 3, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.03509479391362726 }, { "path": "language_model.model.layers.3.self_attn.o_proj", "layer": 3, "layer_type": "full_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.047125915910451774 }, { "path": "language_model.model.layers.3.mlp.gate_proj", "layer": 3, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.019515000915455945 }, { "path": "language_model.model.layers.3.mlp.down_proj", "layer": 3, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.061700401195738386 }, { "path": "language_model.model.layers.3.mlp.up_proj", "layer": 3, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.042194126852922675 }, { "path": "language_model.model.layers.4.linear_attn.in_proj_qkv", "layer": 4, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.02397921018909746 }, { "path": "language_model.model.layers.4.linear_attn.in_proj_z", "layer": 4, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.023035051569588204 }, { "path": "language_model.model.layers.4.linear_attn.in_proj_b", "layer": 4, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.01564397704726879 }, { "path": "language_model.model.layers.4.linear_attn.in_proj_a", "layer": 4, "layer_type": "linear_attention", "nparams": 16384, "bits": 2, "kind": "ternary", "kl": 0.013407841441155961 }, { "path": "language_model.model.layers.4.linear_attn.out_proj", "layer": 4, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.023875757661911464 }, { "path": "language_model.model.layers.4.mlp.gate_proj", "layer": 4, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.031587638719582915 }, { "path": "language_model.model.layers.4.mlp.down_proj", "layer": 4, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.136789628508474 }, { "path": "language_model.model.layers.4.mlp.up_proj", "layer": 4, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.03427242462143647 }, { "path": "language_model.model.layers.5.linear_attn.in_proj_qkv", "layer": 5, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.03755131517139272 }, { "path": "language_model.model.layers.5.linear_attn.in_proj_z", "layer": 5, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.02201721690262345 }, { "path": "language_model.model.layers.5.linear_attn.in_proj_b", "layer": 5, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.014129301163061457 }, { "path": "language_model.model.layers.5.linear_attn.in_proj_a", "layer": 5, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.00921678707939411 }, { "path": "language_model.model.layers.5.linear_attn.out_proj", "layer": 5, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.05138253773226259 }, { "path": "language_model.model.layers.5.mlp.gate_proj", "layer": 5, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.032154125558556325 }, { "path": "language_model.model.layers.5.mlp.down_proj", "layer": 5, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.043968335455040024 }, { "path": "language_model.model.layers.5.mlp.up_proj", "layer": 5, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.062463296864341974 }, { "path": "language_model.model.layers.6.linear_attn.in_proj_qkv", "layer": 6, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.051413092635799004 }, { "path": "language_model.model.layers.6.linear_attn.in_proj_z", "layer": 6, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.02528388093525213 }, { "path": "language_model.model.layers.6.linear_attn.in_proj_b", "layer": 6, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.010495090312282684 }, { "path": "language_model.model.layers.6.linear_attn.in_proj_a", "layer": 6, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.010463668713146171 }, { "path": "language_model.model.layers.6.linear_attn.out_proj", "layer": 6, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.03687874271655201 }, { "path": "language_model.model.layers.6.mlp.gate_proj", "layer": 6, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.041433529224225885 }, { "path": "language_model.model.layers.6.mlp.down_proj", "layer": 6, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.04199239106005209 }, { "path": "language_model.model.layers.6.mlp.up_proj", "layer": 6, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.09370311534083604 }, { "path": "language_model.model.layers.7.self_attn.q_proj", "layer": 7, "layer_type": "full_attention", "nparams": 4194304, "bits": 4, "kind": "4", "kl": 0.0639473471482219 }, { "path": "language_model.model.layers.7.self_attn.k_proj", "layer": 7, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.01995626164111499 }, { "path": "language_model.model.layers.7.self_attn.v_proj", "layer": 7, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.03879161901214778 }, { "path": "language_model.model.layers.7.self_attn.o_proj", "layer": 7, "layer_type": "full_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.07775133380157301 }, { "path": "language_model.model.layers.7.mlp.gate_proj", "layer": 7, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.01902822349416639 }, { "path": "language_model.model.layers.7.mlp.down_proj", "layer": 7, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.02074294201563859 }, { "path": "language_model.model.layers.7.mlp.up_proj", "layer": 7, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.03005311131768014 }, { "path": "language_model.model.layers.8.linear_attn.in_proj_qkv", "layer": 8, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.026959423507237353 }, { "path": "language_model.model.layers.8.linear_attn.in_proj_z", "layer": 8, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.010542849449285302 }, { "path": "language_model.model.layers.8.linear_attn.in_proj_b", "layer": 8, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.006259536017413378 }, { "path": "language_model.model.layers.8.linear_attn.in_proj_a", "layer": 8, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.005066103889982274 }, { "path": "language_model.model.layers.8.linear_attn.out_proj", "layer": 8, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.04619141676349868 }, { "path": "language_model.model.layers.8.mlp.gate_proj", "layer": 8, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.019728178176647466 }, { "path": "language_model.model.layers.8.mlp.down_proj", "layer": 8, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.026803769680060083 }, { "path": "language_model.model.layers.8.mlp.up_proj", "layer": 8, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.022680045334707168 }, { "path": "language_model.model.layers.9.linear_attn.in_proj_qkv", "layer": 9, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.03976326973341923 }, { "path": "language_model.model.layers.9.linear_attn.in_proj_z", "layer": 9, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.018086428763740822 }, { "path": "language_model.model.layers.9.linear_attn.in_proj_b", "layer": 9, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.008665403679117296 }, { "path": "language_model.model.layers.9.linear_attn.in_proj_a", "layer": 9, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.005111378159239233 }, { "path": "language_model.model.layers.9.linear_attn.out_proj", "layer": 9, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.024027441586703807 }, { "path": "language_model.model.layers.9.mlp.gate_proj", "layer": 9, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.017517147359535766 }, { "path": "language_model.model.layers.9.mlp.down_proj", "layer": 9, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.018829855820785585 }, { "path": "language_model.model.layers.9.mlp.up_proj", "layer": 9, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.021916257691088493 }, { "path": "language_model.model.layers.10.linear_attn.in_proj_qkv", "layer": 10, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.04024679641347246 }, { "path": "language_model.model.layers.10.linear_attn.in_proj_z", "layer": 10, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.011116799937381346 }, { "path": "language_model.model.layers.10.linear_attn.in_proj_b", "layer": 10, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0043433013822391404 }, { "path": "language_model.model.layers.10.linear_attn.in_proj_a", "layer": 10, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.004665172839471169 }, { "path": "language_model.model.layers.10.linear_attn.out_proj", "layer": 10, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.02778384028510572 }, { "path": "language_model.model.layers.10.mlp.gate_proj", "layer": 10, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.007522035321253345 }, { "path": "language_model.model.layers.10.mlp.down_proj", "layer": 10, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.015915166493531592 }, { "path": "language_model.model.layers.10.mlp.up_proj", "layer": 10, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.017182175372423135 }, { "path": "language_model.model.layers.11.self_attn.q_proj", "layer": 11, "layer_type": "full_attention", "nparams": 4194304, "bits": 4, "kind": "4", "kl": 0.01918938407635554 }, { "path": "language_model.model.layers.11.self_attn.k_proj", "layer": 11, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.02101637260234458 }, { "path": "language_model.model.layers.11.self_attn.v_proj", "layer": 11, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.03523801972554967 }, { "path": "language_model.model.layers.11.self_attn.o_proj", "layer": 11, "layer_type": "full_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.044128938397831746 }, { "path": "language_model.model.layers.11.mlp.gate_proj", "layer": 11, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.006564071190068745 }, { "path": "language_model.model.layers.11.mlp.down_proj", "layer": 11, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.009954753161174175 }, { "path": "language_model.model.layers.11.mlp.up_proj", "layer": 11, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.016408592796186942 }, { "path": "language_model.model.layers.12.linear_attn.in_proj_qkv", "layer": 12, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.012744699638467517 }, { "path": "language_model.model.layers.12.linear_attn.in_proj_z", "layer": 12, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.004262488847540524 }, { "path": "language_model.model.layers.12.linear_attn.in_proj_b", "layer": 12, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.003491112939298953 }, { "path": "language_model.model.layers.12.linear_attn.in_proj_a", "layer": 12, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0023025657529113326 }, { "path": "language_model.model.layers.12.linear_attn.out_proj", "layer": 12, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.006398780878283918 }, { "path": "language_model.model.layers.12.mlp.gate_proj", "layer": 12, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.005122177783788138 }, { "path": "language_model.model.layers.12.mlp.down_proj", "layer": 12, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.00533927442938409 }, { "path": "language_model.model.layers.12.mlp.up_proj", "layer": 12, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.0061679559347479485 }, { "path": "language_model.model.layers.13.linear_attn.in_proj_qkv", "layer": 13, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.011533852265464987 }, { "path": "language_model.model.layers.13.linear_attn.in_proj_z", "layer": 13, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.0028721824605509763 }, { "path": "language_model.model.layers.13.linear_attn.in_proj_b", "layer": 13, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0023588539987408805 }, { "path": "language_model.model.layers.13.linear_attn.in_proj_a", "layer": 13, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.00199120487849583 }, { "path": "language_model.model.layers.13.linear_attn.out_proj", "layer": 13, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.004224189863711925 }, { "path": "language_model.model.layers.13.mlp.gate_proj", "layer": 13, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.012466730321177164 }, { "path": "language_model.model.layers.13.mlp.down_proj", "layer": 13, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.002558384258649209 }, { "path": "language_model.model.layers.13.mlp.up_proj", "layer": 13, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.0029687005528531204 }, { "path": "language_model.model.layers.14.linear_attn.in_proj_qkv", "layer": 14, "layer_type": "linear_attention", "nparams": 6291456, "bits": 4, "kind": "4", "kl": 0.007997483313712273 }, { "path": "language_model.model.layers.14.linear_attn.in_proj_z", "layer": 14, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.0052743370907232245 }, { "path": "language_model.model.layers.14.linear_attn.in_proj_b", "layer": 14, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0024342861764566695 }, { "path": "language_model.model.layers.14.linear_attn.in_proj_a", "layer": 14, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0016676004025067079 }, { "path": "language_model.model.layers.14.linear_attn.out_proj", "layer": 14, "layer_type": "linear_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.004083662991277806 }, { "path": "language_model.model.layers.14.mlp.gate_proj", "layer": 14, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.008108828858961267 }, { "path": "language_model.model.layers.14.mlp.down_proj", "layer": 14, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.004123490271222968 }, { "path": "language_model.model.layers.14.mlp.up_proj", "layer": 14, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.007036047053022412 }, { "path": "language_model.model.layers.15.self_attn.q_proj", "layer": 15, "layer_type": "full_attention", "nparams": 4194304, "bits": 4, "kind": "4", "kl": 0.004153576767074394 }, { "path": "language_model.model.layers.15.self_attn.k_proj", "layer": 15, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.004112759303064329 }, { "path": "language_model.model.layers.15.self_attn.v_proj", "layer": 15, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.0045544497693278275 }, { "path": "language_model.model.layers.15.self_attn.o_proj", "layer": 15, "layer_type": "full_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.003484462206984148 }, { "path": "language_model.model.layers.15.mlp.gate_proj", "layer": 15, "layer_type": "full_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.002868431546808957 }, { "path": "language_model.model.layers.15.mlp.down_proj", "layer": 15, "layer_type": "full_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.006095650806291535 }, { "path": "language_model.model.layers.15.mlp.up_proj", "layer": 15, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.004815552077497851 }, { "path": "language_model.model.layers.16.linear_attn.in_proj_qkv", "layer": 16, "layer_type": "linear_attention", "nparams": 6291456, "bits": 3, "kind": "3", "kl": 0.00292960581312596 }, { "path": "language_model.model.layers.16.linear_attn.in_proj_z", "layer": 16, "layer_type": "linear_attention", "nparams": 2097152, "bits": 2, "kind": "2", "kl": 0.0034485573524671727 }, { "path": "language_model.model.layers.16.linear_attn.in_proj_b", "layer": 16, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.0012654082102243791 }, { "path": "language_model.model.layers.16.linear_attn.in_proj_a", "layer": 16, "layer_type": "linear_attention", "nparams": 16384, "bits": 2, "kind": "2", "kl": 0.0012813463656705713 }, { "path": "language_model.model.layers.16.linear_attn.out_proj", "layer": 16, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.005193882758604845 }, { "path": "language_model.model.layers.16.mlp.gate_proj", "layer": 16, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.002351040174577789 }, { "path": "language_model.model.layers.16.mlp.down_proj", "layer": 16, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.00440584063789979 }, { "path": "language_model.model.layers.16.mlp.up_proj", "layer": 16, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.007104415338209336 }, { "path": "language_model.model.layers.17.linear_attn.in_proj_qkv", "layer": 17, "layer_type": "linear_attention", "nparams": 6291456, "bits": 2, "kind": "ternary", "kl": 0.0021934894088310236 }, { "path": "language_model.model.layers.17.linear_attn.in_proj_z", "layer": 17, "layer_type": "linear_attention", "nparams": 2097152, "bits": 2, "kind": "2", "kl": 0.0032236417576331686 }, { "path": "language_model.model.layers.17.linear_attn.in_proj_b", "layer": 17, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.000886947803901282 }, { "path": "language_model.model.layers.17.linear_attn.in_proj_a", "layer": 17, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.000976995525569164 }, { "path": "language_model.model.layers.17.linear_attn.out_proj", "layer": 17, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.002114814895788902 }, { "path": "language_model.model.layers.17.mlp.gate_proj", "layer": 17, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.004665009833481794 }, { "path": "language_model.model.layers.17.mlp.down_proj", "layer": 17, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.004484325279917158 }, { "path": "language_model.model.layers.17.mlp.up_proj", "layer": 17, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "ternary", "kl": 0.005121177837445393 }, { "path": "language_model.model.layers.18.linear_attn.in_proj_qkv", "layer": 18, "layer_type": "linear_attention", "nparams": 6291456, "bits": 2, "kind": "2", "kl": 0.006290895747047132 }, { "path": "language_model.model.layers.18.linear_attn.in_proj_z", "layer": 18, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.002068792589064126 }, { "path": "language_model.model.layers.18.linear_attn.in_proj_b", "layer": 18, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.0008810069546544934 }, { "path": "language_model.model.layers.18.linear_attn.in_proj_a", "layer": 18, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.000987608700835434 }, { "path": "language_model.model.layers.18.linear_attn.out_proj", "layer": 18, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.002887347314595934 }, { "path": "language_model.model.layers.18.mlp.gate_proj", "layer": 18, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.0018628583013199987 }, { "path": "language_model.model.layers.18.mlp.down_proj", "layer": 18, "layer_type": "linear_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.0023940537857527703 }, { "path": "language_model.model.layers.18.mlp.up_proj", "layer": 18, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.006746079837461991 }, { "path": "language_model.model.layers.19.self_attn.q_proj", "layer": 19, "layer_type": "full_attention", "nparams": 4194304, "bits": 3, "kind": "3", "kl": 0.006032338261324889 }, { "path": "language_model.model.layers.19.self_attn.k_proj", "layer": 19, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.0021031284974654106 }, { "path": "language_model.model.layers.19.self_attn.v_proj", "layer": 19, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.004348253480209944 }, { "path": "language_model.model.layers.19.self_attn.o_proj", "layer": 19, "layer_type": "full_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.004557360657928573 }, { "path": "language_model.model.layers.19.mlp.gate_proj", "layer": 19, "layer_type": "full_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.0029232948535193653 }, { "path": "language_model.model.layers.19.mlp.down_proj", "layer": 19, "layer_type": "full_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.0058071099995981815 }, { "path": "language_model.model.layers.19.mlp.up_proj", "layer": 19, "layer_type": "full_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.0023828522802972463 }, { "path": "language_model.model.layers.20.linear_attn.in_proj_qkv", "layer": 20, "layer_type": "linear_attention", "nparams": 6291456, "bits": 2, "kind": "2", "kl": 0.008125929608638258 }, { "path": "language_model.model.layers.20.linear_attn.in_proj_z", "layer": 20, "layer_type": "linear_attention", "nparams": 2097152, "bits": 2, "kind": "2", "kl": 0.0017258157022901578 }, { "path": "language_model.model.layers.20.linear_attn.in_proj_b", "layer": 20, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0005154021203862869 }, { "path": "language_model.model.layers.20.linear_attn.in_proj_a", "layer": 20, "layer_type": "linear_attention", "nparams": 16384, "bits": 2, "kind": "ternary", "kl": 0.0005305295310921597 }, { "path": "language_model.model.layers.20.linear_attn.out_proj", "layer": 20, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.0030152998807947977 }, { "path": "language_model.model.layers.20.mlp.gate_proj", "layer": 20, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.0022819108116763583 }, { "path": "language_model.model.layers.20.mlp.down_proj", "layer": 20, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.0015101953736824964 }, { "path": "language_model.model.layers.20.mlp.up_proj", "layer": 20, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "ternary", "kl": 0.004383521037158472 }, { "path": "language_model.model.layers.21.linear_attn.in_proj_qkv", "layer": 21, "layer_type": "linear_attention", "nparams": 6291456, "bits": 2, "kind": "2", "kl": 0.0035359140742138107 }, { "path": "language_model.model.layers.21.linear_attn.in_proj_z", "layer": 21, "layer_type": "linear_attention", "nparams": 2097152, "bits": 2, "kind": "2", "kl": 0.0033209959668630907 }, { "path": "language_model.model.layers.21.linear_attn.in_proj_b", "layer": 21, "layer_type": "linear_attention", "nparams": 16384, "bits": 2, "kind": "2", "kl": 0.0005307516921509054 }, { "path": "language_model.model.layers.21.linear_attn.in_proj_a", "layer": 21, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0002769620310376814 }, { "path": "language_model.model.layers.21.linear_attn.out_proj", "layer": 21, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.0017829395230716451 }, { "path": "language_model.model.layers.21.mlp.gate_proj", "layer": 21, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "ternary", "kl": 0.004034291358501529 }, { "path": "language_model.model.layers.21.mlp.down_proj", "layer": 21, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.0008362982179293819 }, { "path": "language_model.model.layers.21.mlp.up_proj", "layer": 21, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.003467660124694521 }, { "path": "language_model.model.layers.22.linear_attn.in_proj_qkv", "layer": 22, "layer_type": "linear_attention", "nparams": 6291456, "bits": 3, "kind": "3", "kl": 0.006269973941838318 }, { "path": "language_model.model.layers.22.linear_attn.in_proj_z", "layer": 22, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.0034496159643663885 }, { "path": "language_model.model.layers.22.linear_attn.in_proj_b", "layer": 22, "layer_type": "linear_attention", "nparams": 16384, "bits": 4, "kind": "4", "kl": 0.0005182467336934553 }, { "path": "language_model.model.layers.22.linear_attn.in_proj_a", "layer": 22, "layer_type": "linear_attention", "nparams": 16384, "bits": 3, "kind": "3", "kl": 0.0003643164638947492 }, { "path": "language_model.model.layers.22.linear_attn.out_proj", "layer": 22, "layer_type": "linear_attention", "nparams": 2097152, "bits": 3, "kind": "3", "kl": 0.003477276947365615 }, { "path": "language_model.model.layers.22.mlp.gate_proj", "layer": 22, "layer_type": "linear_attention", "nparams": 3670016, "bits": 2, "kind": "2", "kl": 0.005100013411379256 }, { "path": "language_model.model.layers.22.mlp.down_proj", "layer": 22, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.0033566558855619606 }, { "path": "language_model.model.layers.22.mlp.up_proj", "layer": 22, "layer_type": "linear_attention", "nparams": 3670016, "bits": 3, "kind": "3", "kl": 0.00313410047627708 }, { "path": "language_model.model.layers.23.self_attn.q_proj", "layer": 23, "layer_type": "full_attention", "nparams": 4194304, "bits": 3, "kind": "3", "kl": 0.004782156469623039 }, { "path": "language_model.model.layers.23.self_attn.k_proj", "layer": 23, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.006514890193217124 }, { "path": "language_model.model.layers.23.self_attn.v_proj", "layer": 23, "layer_type": "full_attention", "nparams": 524288, "bits": 4, "kind": "4", "kl": 0.002427561433381982 }, { "path": "language_model.model.layers.23.self_attn.o_proj", "layer": 23, "layer_type": "full_attention", "nparams": 2097152, "bits": 4, "kind": "4", "kl": 0.002634655396783468 }, { "path": "language_model.model.layers.23.mlp.gate_proj", "layer": 23, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.001242476476743588 }, { "path": "language_model.model.layers.23.mlp.down_proj", "layer": 23, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.029779743010173505 }, { "path": "language_model.model.layers.23.mlp.up_proj", "layer": 23, "layer_type": "full_attention", "nparams": 3670016, "bits": 4, "kind": "4", "kl": 0.006386583194459594 } ] }