Instructions to use Jakevin/d3-nano-ternary-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Jakevin/d3-nano-ternary-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download Jakevin/d3-nano-ternary-mlx --local-dir d3-nano-ternary-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download allocation.json from Jakevin/d3-nano-ternary-mlx: direct link, hf CLI and curl.
- Browser
- Download file 37.1 kB
-
https://huggingface.co/Jakevin/d3-nano-ternary-mlx/resolve/main/allocation.json
- Command line
-
hf download hf://Jakevin/d3-nano-ternary-mlx/allocation.json
-
curl -L -o allocation.json https://huggingface.co/Jakevin/d3-nano-ternary-mlx/resolve/main/allocation.json
37.1 kB
{
"average_bits": 3.4985438747254847,
"embed_bits": 8,
"budget": {
"low_avg": 3.4,
"high_avg": 3.5
},
"sum_kl": 2.977956267718296,
"dp_sum_kl": 2.540285839100619,
"joint_refine": {
"calibration_n": 20,
"calibration_ids": [
"mnli/validation_matched/8418/relation",
"banking77/test/1454/intent",
"boolq/validation/1209/answer",
"composition/v3-composition-dev/negation/1/relevant/a/decision",
"amazon/test/605/stars",
"dbpedia14/test/44634/category",
"contrastive/age_eligibility/v3-control-dev-age_eligibility-0013/b/decision",
"yelp/test/10368/rating",
"agnews/test/4286/topic",
"sst5/test/445/sentiment",
"trec/test/455/answer_type",
"imdb/test/3003/positive",
"readme-urgency",
"mnli/validation_matched/3117/relation",
"dbpedia14/test/48873/category",
"composition/v3-composition-dev/conditional/2/irrelevant/b/decision",
"mnli/validation_matched/3000/relation",
"boolq/validation/376/answer",
"trec/test/280/none_absent/answer_type",
"mnli/validation_matched/8418/permuted/relation"
],
"joint_kl_before": 0.359920485658611,
"joint_kl_after": 0.15394268493215005,
"swaps": 12
},
"assignment": [
{
"path": "language_model.model.layers.0.linear_attn.in_proj_qkv",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.036167649794940304
},
{
"path": "language_model.model.layers.0.linear_attn.in_proj_z",
"layer": 0,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.024077155345270138
},
{
"path": "language_model.model.layers.0.linear_attn.in_proj_a",
"layer": 0,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.004719737666264188
},
{
"path": "language_model.model.layers.0.linear_attn.in_proj_b",
"layer": 0,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.009882406261872978
},
{
"path": "language_model.model.layers.0.linear_attn.out_proj",
"layer": 0,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.19649393687983674
},
{
"path": "language_model.model.layers.0.mlp.gate_proj",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.014610158877154451
},
{
"path": "language_model.model.layers.0.mlp.up_proj",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.019142927850698384
},
{
"path": "language_model.model.layers.0.mlp.down_proj",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.018993703064068047
},
{
"path": "language_model.model.layers.1.linear_attn.in_proj_qkv",
"layer": 1,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.02159494921496259
},
{
"path": "language_model.model.layers.1.linear_attn.in_proj_z",
"layer": 1,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.009249335523625427
},
{
"path": "language_model.model.layers.1.linear_attn.in_proj_a",
"layer": 1,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.009942285073314044
},
{
"path": "language_model.model.layers.1.linear_attn.in_proj_b",
"layer": 1,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.011165700598795746
},
{
"path": "language_model.model.layers.1.linear_attn.out_proj",
"layer": 1,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.02533359810132539
},
{
"path": "language_model.model.layers.1.mlp.gate_proj",
"layer": 1,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.020001911429270484
},
{
"path": "language_model.model.layers.1.mlp.up_proj",
"layer": 1,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.018694505846325777
},
{
"path": "language_model.model.layers.1.mlp.down_proj",
"layer": 1,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.045032779148968485
},
{
"path": "language_model.model.layers.2.linear_attn.in_proj_qkv",
"layer": 2,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.028996766755013885
},
{
"path": "language_model.model.layers.2.linear_attn.in_proj_z",
"layer": 2,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.018902605744829708
},
{
"path": "language_model.model.layers.2.linear_attn.in_proj_a",
"layer": 2,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.006896526199611989
},
{
"path": "language_model.model.layers.2.linear_attn.in_proj_b",
"layer": 2,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.009745471732003758
},
{
"path": "language_model.model.layers.2.linear_attn.out_proj",
"layer": 2,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.037437193703247655
},
{
"path": "language_model.model.layers.2.mlp.gate_proj",
"layer": 2,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.020557114209522904
},
{
"path": "language_model.model.layers.2.mlp.up_proj",
"layer": 2,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.031465961257551245
},
{
"path": "language_model.model.layers.2.mlp.down_proj",
"layer": 2,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.032831266240288565
},
{
"path": "language_model.model.layers.3.self_attn.q_proj",
"layer": 3,
"nparams": 8388608,
"bits": 3,
"kind": "3",
"kl": 0.060484579321996676
},
{
"path": "language_model.model.layers.3.self_attn.k_proj",
"layer": 3,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.01184541686985306
},
{
"path": "language_model.model.layers.3.self_attn.v_proj",
"layer": 3,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.04672758597267979
},
{
"path": "language_model.model.layers.3.self_attn.o_proj",
"layer": 3,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.03858699061895039
},
{
"path": "language_model.model.layers.3.mlp.gate_proj",
"layer": 3,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.015935746183335358
},
{
"path": "language_model.model.layers.3.mlp.up_proj",
"layer": 3,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.04384401309329482
},
{
"path": "language_model.model.layers.3.mlp.down_proj",
"layer": 3,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.02399046514312047
},
{
"path": "language_model.model.layers.4.linear_attn.in_proj_qkv",
"layer": 4,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.03380978350370204
},
{
"path": "language_model.model.layers.4.linear_attn.in_proj_z",
"layer": 4,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.015917916860611572
},
{
"path": "language_model.model.layers.4.linear_attn.in_proj_a",
"layer": 4,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.0071051056101440815
},
{
"path": "language_model.model.layers.4.linear_attn.in_proj_b",
"layer": 4,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.011950433791683937
},
{
"path": "language_model.model.layers.4.linear_attn.out_proj",
"layer": 4,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.029419899497863965
},
{
"path": "language_model.model.layers.4.mlp.gate_proj",
"layer": 4,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.02037560206293295
},
{
"path": "language_model.model.layers.4.mlp.up_proj",
"layer": 4,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.026954146641484288
},
{
"path": "language_model.model.layers.4.mlp.down_proj",
"layer": 4,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.031966774444906774
},
{
"path": "language_model.model.layers.5.linear_attn.in_proj_qkv",
"layer": 5,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.024604880627422072
},
{
"path": "language_model.model.layers.5.linear_attn.in_proj_z",
"layer": 5,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.01946826237510793
},
{
"path": "language_model.model.layers.5.linear_attn.in_proj_a",
"layer": 5,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.006358107711251463
},
{
"path": "language_model.model.layers.5.linear_attn.in_proj_b",
"layer": 5,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.007007401410325599
},
{
"path": "language_model.model.layers.5.linear_attn.out_proj",
"layer": 5,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.04259233745882887
},
{
"path": "language_model.model.layers.5.mlp.gate_proj",
"layer": 5,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.021353867373001368
},
{
"path": "language_model.model.layers.5.mlp.up_proj",
"layer": 5,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.03279443582076645
},
{
"path": "language_model.model.layers.5.mlp.down_proj",
"layer": 5,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.030314831495057707
},
{
"path": "language_model.model.layers.6.linear_attn.in_proj_qkv",
"layer": 6,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.03271812705592554
},
{
"path": "language_model.model.layers.6.linear_attn.in_proj_z",
"layer": 6,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.01999809052483925
},
{
"path": "language_model.model.layers.6.linear_attn.in_proj_a",
"layer": 6,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.006959841064670462
},
{
"path": "language_model.model.layers.6.linear_attn.in_proj_b",
"layer": 6,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.008939020186841463
},
{
"path": "language_model.model.layers.6.linear_attn.out_proj",
"layer": 6,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.02744619635510964
},
{
"path": "language_model.model.layers.6.mlp.gate_proj",
"layer": 6,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.025605077184504846
},
{
"path": "language_model.model.layers.6.mlp.up_proj",
"layer": 6,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.04094105855935139
},
{
"path": "language_model.model.layers.6.mlp.down_proj",
"layer": 6,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.03757126084623266
},
{
"path": "language_model.model.layers.7.self_attn.q_proj",
"layer": 7,
"nparams": 8388608,
"bits": 3,
"kind": "3",
"kl": 0.12697585014100587
},
{
"path": "language_model.model.layers.7.self_attn.k_proj",
"layer": 7,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.011872071865418204
},
{
"path": "language_model.model.layers.7.self_attn.v_proj",
"layer": 7,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.03341289663710259
},
{
"path": "language_model.model.layers.7.self_attn.o_proj",
"layer": 7,
"nparams": 4194304,
"bits": 3,
"kind": "3",
"kl": 0.2620323669243249
},
{
"path": "language_model.model.layers.7.mlp.gate_proj",
"layer": 7,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.020175184201294494
},
{
"path": "language_model.model.layers.7.mlp.up_proj",
"layer": 7,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.028832777507502485
},
{
"path": "language_model.model.layers.7.mlp.down_proj",
"layer": 7,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.03271163825399574
},
{
"path": "language_model.model.layers.8.linear_attn.in_proj_qkv",
"layer": 8,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.017208185022725524
},
{
"path": "language_model.model.layers.8.linear_attn.in_proj_z",
"layer": 8,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.008539868907538503
},
{
"path": "language_model.model.layers.8.linear_attn.in_proj_a",
"layer": 8,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.005789555197582355
},
{
"path": "language_model.model.layers.8.linear_attn.in_proj_b",
"layer": 8,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.004701556493609789
},
{
"path": "language_model.model.layers.8.linear_attn.out_proj",
"layer": 8,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.017634184957528064
},
{
"path": "language_model.model.layers.8.mlp.gate_proj",
"layer": 8,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.0123514499375841
},
{
"path": "language_model.model.layers.8.mlp.up_proj",
"layer": 8,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.04801137650896942
},
{
"path": "language_model.model.layers.8.mlp.down_proj",
"layer": 8,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.024051073026259938
},
{
"path": "language_model.model.layers.9.linear_attn.in_proj_qkv",
"layer": 9,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.033309427679144944
},
{
"path": "language_model.model.layers.9.linear_attn.in_proj_z",
"layer": 9,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.008170563685431696
},
{
"path": "language_model.model.layers.9.linear_attn.in_proj_a",
"layer": 9,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.004981094336314765
},
{
"path": "language_model.model.layers.9.linear_attn.in_proj_b",
"layer": 9,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.004953332458841868
},
{
"path": "language_model.model.layers.9.linear_attn.out_proj",
"layer": 9,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.02677805449801729
},
{
"path": "language_model.model.layers.9.mlp.gate_proj",
"layer": 9,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.013528162657920756
},
{
"path": "language_model.model.layers.9.mlp.up_proj",
"layer": 9,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.022252418419872137
},
{
"path": "language_model.model.layers.9.mlp.down_proj",
"layer": 9,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.009998477832443698
},
{
"path": "language_model.model.layers.10.linear_attn.in_proj_qkv",
"layer": 10,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.021357037951699957
},
{
"path": "language_model.model.layers.10.linear_attn.in_proj_z",
"layer": 10,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.00935590656026332
},
{
"path": "language_model.model.layers.10.linear_attn.in_proj_a",
"layer": 10,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.003710759392787249
},
{
"path": "language_model.model.layers.10.linear_attn.in_proj_b",
"layer": 10,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.004896769494430089
},
{
"path": "language_model.model.layers.10.linear_attn.out_proj",
"layer": 10,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.014627255957999335
},
{
"path": "language_model.model.layers.10.mlp.gate_proj",
"layer": 10,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.01346382014639341
},
{
"path": "language_model.model.layers.10.mlp.up_proj",
"layer": 10,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.02852888862376233
},
{
"path": "language_model.model.layers.10.mlp.down_proj",
"layer": 10,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.01685266069767212
},
{
"path": "language_model.model.layers.11.self_attn.q_proj",
"layer": 11,
"nparams": 8388608,
"bits": 4,
"kind": "4",
"kl": 0.02689443748441111
},
{
"path": "language_model.model.layers.11.self_attn.k_proj",
"layer": 11,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.015188959495439124
},
{
"path": "language_model.model.layers.11.self_attn.v_proj",
"layer": 11,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.022758453695708634
},
{
"path": "language_model.model.layers.11.self_attn.o_proj",
"layer": 11,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.027215593169970794
},
{
"path": "language_model.model.layers.11.mlp.gate_proj",
"layer": 11,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.011182316594605536
},
{
"path": "language_model.model.layers.11.mlp.up_proj",
"layer": 11,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.01113242412163202
},
{
"path": "language_model.model.layers.11.mlp.down_proj",
"layer": 11,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.009885245854565492
},
{
"path": "language_model.model.layers.12.linear_attn.in_proj_qkv",
"layer": 12,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.018947691650334015
},
{
"path": "language_model.model.layers.12.linear_attn.in_proj_z",
"layer": 12,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.006582978430832682
},
{
"path": "language_model.model.layers.12.linear_attn.in_proj_a",
"layer": 12,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.004025139540875921
},
{
"path": "language_model.model.layers.12.linear_attn.in_proj_b",
"layer": 12,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.0050430632419457035
},
{
"path": "language_model.model.layers.12.linear_attn.out_proj",
"layer": 12,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.006496160800308889
},
{
"path": "language_model.model.layers.12.mlp.gate_proj",
"layer": 12,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.007851327559779145
},
{
"path": "language_model.model.layers.12.mlp.up_proj",
"layer": 12,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.011228132923593922
},
{
"path": "language_model.model.layers.12.mlp.down_proj",
"layer": 12,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.004278566865995799
},
{
"path": "language_model.model.layers.13.linear_attn.in_proj_qkv",
"layer": 13,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.01233742495220853
},
{
"path": "language_model.model.layers.13.linear_attn.in_proj_z",
"layer": 13,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.004191599132337177
},
{
"path": "language_model.model.layers.13.linear_attn.in_proj_a",
"layer": 13,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.0026647038361747354
},
{
"path": "language_model.model.layers.13.linear_attn.in_proj_b",
"layer": 13,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.002115649949411989
},
{
"path": "language_model.model.layers.13.linear_attn.out_proj",
"layer": 13,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.005793977031598178
},
{
"path": "language_model.model.layers.13.mlp.gate_proj",
"layer": 13,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.017964901610867313
},
{
"path": "language_model.model.layers.13.mlp.up_proj",
"layer": 13,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.008408326668711358
},
{
"path": "language_model.model.layers.13.mlp.down_proj",
"layer": 13,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.009545881598574094
},
{
"path": "language_model.model.layers.14.linear_attn.in_proj_qkv",
"layer": 14,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.009219910411077833
},
{
"path": "language_model.model.layers.14.linear_attn.in_proj_z",
"layer": 14,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.004161417746713759
},
{
"path": "language_model.model.layers.14.linear_attn.in_proj_a",
"layer": 14,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.003323217458649921
},
{
"path": "language_model.model.layers.14.linear_attn.in_proj_b",
"layer": 14,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.0021712184461907596
},
{
"path": "language_model.model.layers.14.linear_attn.out_proj",
"layer": 14,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.004262398622704975
},
{
"path": "language_model.model.layers.14.mlp.gate_proj",
"layer": 14,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.028299536252745942
},
{
"path": "language_model.model.layers.14.mlp.up_proj",
"layer": 14,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.0029855243357640063
},
{
"path": "language_model.model.layers.14.mlp.down_proj",
"layer": 14,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.006347102378594772
},
{
"path": "language_model.model.layers.15.self_attn.q_proj",
"layer": 15,
"nparams": 8388608,
"bits": 4,
"kind": "4",
"kl": 0.003643140858535345
},
{
"path": "language_model.model.layers.15.self_attn.k_proj",
"layer": 15,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.0052556378501606195
},
{
"path": "language_model.model.layers.15.self_attn.v_proj",
"layer": 15,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.006322858785280738
},
{
"path": "language_model.model.layers.15.self_attn.o_proj",
"layer": 15,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.005030957821139618
},
{
"path": "language_model.model.layers.15.mlp.gate_proj",
"layer": 15,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0031369600735519213
},
{
"path": "language_model.model.layers.15.mlp.up_proj",
"layer": 15,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.011699879395203576
},
{
"path": "language_model.model.layers.15.mlp.down_proj",
"layer": 15,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0065643853607818565
},
{
"path": "language_model.model.layers.16.linear_attn.in_proj_qkv",
"layer": 16,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.00652818737017804
},
{
"path": "language_model.model.layers.16.linear_attn.in_proj_z",
"layer": 16,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.002564514868387881
},
{
"path": "language_model.model.layers.16.linear_attn.in_proj_a",
"layer": 16,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.001746105079547036
},
{
"path": "language_model.model.layers.16.linear_attn.in_proj_b",
"layer": 16,
"nparams": 32768,
"bits": 2,
"kind": "2",
"kl": 0.0024505408935464497
},
{
"path": "language_model.model.layers.16.linear_attn.out_proj",
"layer": 16,
"nparams": 4194304,
"bits": 3,
"kind": "3",
"kl": 0.0035994804530835194
},
{
"path": "language_model.model.layers.16.mlp.gate_proj",
"layer": 16,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.005310726191144668
},
{
"path": "language_model.model.layers.16.mlp.up_proj",
"layer": 16,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.003419146426726522
},
{
"path": "language_model.model.layers.16.mlp.down_proj",
"layer": 16,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.004005140748420486
},
{
"path": "language_model.model.layers.17.linear_attn.in_proj_qkv",
"layer": 17,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0037804420468492837
},
{
"path": "language_model.model.layers.17.linear_attn.in_proj_z",
"layer": 17,
"nparams": 4194304,
"bits": 2,
"kind": "2",
"kl": 0.0035376995942987717
},
{
"path": "language_model.model.layers.17.linear_attn.in_proj_a",
"layer": 17,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.001846708406487311
},
{
"path": "language_model.model.layers.17.linear_attn.in_proj_b",
"layer": 17,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.0016845755580104484
},
{
"path": "language_model.model.layers.17.linear_attn.out_proj",
"layer": 17,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.001894043444011438
},
{
"path": "language_model.model.layers.17.mlp.gate_proj",
"layer": 17,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0032691978276667404
},
{
"path": "language_model.model.layers.17.mlp.up_proj",
"layer": 17,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.007563979548310841
},
{
"path": "language_model.model.layers.17.mlp.down_proj",
"layer": 17,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.01315814833027159
},
{
"path": "language_model.model.layers.18.linear_attn.in_proj_qkv",
"layer": 18,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.0052731935273633256
},
{
"path": "language_model.model.layers.18.linear_attn.in_proj_z",
"layer": 18,
"nparams": 4194304,
"bits": 2,
"kind": "2",
"kl": 0.0031266101628979383
},
{
"path": "language_model.model.layers.18.linear_attn.in_proj_a",
"layer": 18,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.0019552802619046527
},
{
"path": "language_model.model.layers.18.linear_attn.in_proj_b",
"layer": 18,
"nparams": 32768,
"bits": 2,
"kind": "2",
"kl": 0.0018198732860189083
},
{
"path": "language_model.model.layers.18.linear_attn.out_proj",
"layer": 18,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.0028847887347183973
},
{
"path": "language_model.model.layers.18.mlp.gate_proj",
"layer": 18,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.01898859203658337
},
{
"path": "language_model.model.layers.18.mlp.up_proj",
"layer": 18,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0031293535751949733
},
{
"path": "language_model.model.layers.18.mlp.down_proj",
"layer": 18,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.0037580042827521183
},
{
"path": "language_model.model.layers.19.self_attn.q_proj",
"layer": 19,
"nparams": 8388608,
"bits": 4,
"kind": "4",
"kl": 0.0033964504416273814
},
{
"path": "language_model.model.layers.19.self_attn.k_proj",
"layer": 19,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.0048760714135432336
},
{
"path": "language_model.model.layers.19.self_attn.v_proj",
"layer": 19,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.003869429662332861
},
{
"path": "language_model.model.layers.19.self_attn.o_proj",
"layer": 19,
"nparams": 4194304,
"bits": 3,
"kind": "3",
"kl": 0.018552193105078302
},
{
"path": "language_model.model.layers.19.mlp.gate_proj",
"layer": 19,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0020007518508758588
},
{
"path": "language_model.model.layers.19.mlp.up_proj",
"layer": 19,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0020792066361257477
},
{
"path": "language_model.model.layers.19.mlp.down_proj",
"layer": 19,
"nparams": 12582912,
"bits": 2,
"kind": "ternary",
"kl": 0.006833145492931654
},
{
"path": "language_model.model.layers.20.linear_attn.in_proj_qkv",
"layer": 20,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0025472445161104753
},
{
"path": "language_model.model.layers.20.linear_attn.in_proj_z",
"layer": 20,
"nparams": 4194304,
"bits": 2,
"kind": "2",
"kl": 0.002592213691815188
},
{
"path": "language_model.model.layers.20.linear_attn.in_proj_a",
"layer": 20,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.0009051710490314626
},
{
"path": "language_model.model.layers.20.linear_attn.in_proj_b",
"layer": 20,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.0011463968076723563
},
{
"path": "language_model.model.layers.20.linear_attn.out_proj",
"layer": 20,
"nparams": 4194304,
"bits": 2,
"kind": "2",
"kl": 0.005356271293440156
},
{
"path": "language_model.model.layers.20.mlp.gate_proj",
"layer": 20,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.0021228656968947774
},
{
"path": "language_model.model.layers.20.mlp.up_proj",
"layer": 20,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0024053604971531754
},
{
"path": "language_model.model.layers.20.mlp.down_proj",
"layer": 20,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.004477630598479986
},
{
"path": "language_model.model.layers.21.linear_attn.in_proj_qkv",
"layer": 21,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.004200707271791614
},
{
"path": "language_model.model.layers.21.linear_attn.in_proj_z",
"layer": 21,
"nparams": 4194304,
"bits": 2,
"kind": "2",
"kl": 0.002413397581031617
},
{
"path": "language_model.model.layers.21.linear_attn.in_proj_a",
"layer": 21,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.0005669125558536048
},
{
"path": "language_model.model.layers.21.linear_attn.in_proj_b",
"layer": 21,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.0006892036110613707
},
{
"path": "language_model.model.layers.21.linear_attn.out_proj",
"layer": 21,
"nparams": 4194304,
"bits": 3,
"kind": "3",
"kl": 0.002135502337464641
},
{
"path": "language_model.model.layers.21.mlp.gate_proj",
"layer": 21,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.003934425296862159
},
{
"path": "language_model.model.layers.21.mlp.up_proj",
"layer": 21,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0022890209206124167
},
{
"path": "language_model.model.layers.21.mlp.down_proj",
"layer": 21,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.009127860100761054
},
{
"path": "language_model.model.layers.22.linear_attn.in_proj_qkv",
"layer": 22,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.0053570328330360085
},
{
"path": "language_model.model.layers.22.linear_attn.in_proj_z",
"layer": 22,
"nparams": 4194304,
"bits": 3,
"kind": "3",
"kl": 0.0030685547854956505
},
{
"path": "language_model.model.layers.22.linear_attn.in_proj_a",
"layer": 22,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.0006597311378955131
},
{
"path": "language_model.model.layers.22.linear_attn.in_proj_b",
"layer": 22,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.001038058904486026
},
{
"path": "language_model.model.layers.22.linear_attn.out_proj",
"layer": 22,
"nparams": 4194304,
"bits": 3,
"kind": "3",
"kl": 0.0029542600866622193
},
{
"path": "language_model.model.layers.22.mlp.gate_proj",
"layer": 22,
"nparams": 12582912,
"bits": 2,
"kind": "2",
"kl": 0.005691486545448434
},
{
"path": "language_model.model.layers.22.mlp.up_proj",
"layer": 22,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.0028654545434373922
},
{
"path": "language_model.model.layers.22.mlp.down_proj",
"layer": 22,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.004033058114196539
},
{
"path": "language_model.model.layers.23.self_attn.q_proj",
"layer": 23,
"nparams": 8388608,
"bits": 4,
"kind": "4",
"kl": 0.0034863865396836083
},
{
"path": "language_model.model.layers.23.self_attn.k_proj",
"layer": 23,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.003449246781242146
},
{
"path": "language_model.model.layers.23.self_attn.v_proj",
"layer": 23,
"nparams": 1048576,
"bits": 4,
"kind": "4",
"kl": 0.004781225921661813
},
{
"path": "language_model.model.layers.23.self_attn.o_proj",
"layer": 23,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.004980091902814513
},
{
"path": "language_model.model.layers.23.mlp.gate_proj",
"layer": 23,
"nparams": 12582912,
"bits": 3,
"kind": "3",
"kl": 0.005546539586926521
},
{
"path": "language_model.model.layers.23.mlp.up_proj",
"layer": 23,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.0036412261524133302
},
{
"path": "language_model.model.layers.23.mlp.down_proj",
"layer": 23,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.009847478632996235
}
]
}
{
"average_bits": 3.4985438747254847,
"embed_bits": 8,
"budget": {
"low_avg": 3.4,
"high_avg": 3.5
},
"sum_kl": 2.977956267718296,
"dp_sum_kl": 2.540285839100619,
"joint_refine": {
"calibration_n": 20,
"calibration_ids": [
"mnli/validation_matched/8418/relation",
"banking77/test/1454/intent",
"boolq/validation/1209/answer",
"composition/v3-composition-dev/negation/1/relevant/a/decision",
"amazon/test/605/stars",
"dbpedia14/test/44634/category",
"contrastive/age_eligibility/v3-control-dev-age_eligibility-0013/b/decision",
"yelp/test/10368/rating",
"agnews/test/4286/topic",
"sst5/test/445/sentiment",
"trec/test/455/answer_type",
"imdb/test/3003/positive",
"readme-urgency",
"mnli/validation_matched/3117/relation",
"dbpedia14/test/48873/category",
"composition/v3-composition-dev/conditional/2/irrelevant/b/decision",
"mnli/validation_matched/3000/relation",
"boolq/validation/376/answer",
"trec/test/280/none_absent/answer_type",
"mnli/validation_matched/8418/permuted/relation"
],
"joint_kl_before": 0.359920485658611,
"joint_kl_after": 0.15394268493215005,
"swaps": 12
},
"assignment": [
{
"path": "language_model.model.layers.0.linear_attn.in_proj_qkv",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.036167649794940304
},
{
"path": "language_model.model.layers.0.linear_attn.in_proj_z",
"layer": 0,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.024077155345270138
},
{
"path": "language_model.model.layers.0.linear_attn.in_proj_a",
"layer": 0,
"nparams": 32768,
"bits": 3,
"kind": "3",
"kl": 0.004719737666264188
},
{
"path": "language_model.model.layers.0.linear_attn.in_proj_b",
"layer": 0,
"nparams": 32768,
"bits": 4,
"kind": "4",
"kl": 0.009882406261872978
},
{
"path": "language_model.model.layers.0.linear_attn.out_proj",
"layer": 0,
"nparams": 4194304,
"bits": 4,
"kind": "4",
"kl": 0.19649393687983674
},
{
"path": "language_model.model.layers.0.mlp.gate_proj",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.014610158877154451
},
{
"path": "language_model.model.layers.0.mlp.up_proj",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",
"kl": 0.019142927850698384
},
{
"path": "language_model.model.layers.0.mlp.down_proj",
"layer": 0,
"nparams": 12582912,
"bits": 4,
"kind": "4",