Feature Extraction
Transformers
Safetensors
hetero_moe_projspace
mixture-of-experts
heterogeneous
dimension-projection-space
ffn-fusion
custom_code
Instructions to use redmingwei/HeteroMoEProjSpace-v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use redmingwei/HeteroMoEProjSpace-v2 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="redmingwei/HeteroMoEProjSpace-v2", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("redmingwei/HeteroMoEProjSpace-v2", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
删除:v62.2 已被 v62.3 取代,且此目录内容与 v62.3 逐位相同(仅目录名错位)
Browse files- checkpoints/v62.2/report.json +0 -177
checkpoints/v62.2/report.json
DELETED
|
@@ -1,177 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"hist": [
|
| 3 |
-
{
|
| 4 |
-
"step": 0,
|
| 5 |
-
"ce": 1.4129877090454102,
|
| 6 |
-
"train_acc": 0.375,
|
| 7 |
-
"val_acc": 0.24825174825174826,
|
| 8 |
-
"bal": 1.8297117948532104,
|
| 9 |
-
"load": [
|
| 10 |
-
0.39306640625,
|
| 11 |
-
0.758544921875,
|
| 12 |
-
0.56884765625,
|
| 13 |
-
0.279541015625
|
| 14 |
-
],
|
| 15 |
-
"ent": 0.506121164887764,
|
| 16 |
-
"router_wnorm": 0.779486358165741,
|
| 17 |
-
"router_gnorm": 0.8539896011352539,
|
| 18 |
-
"slot_ent": 0.35127997398376465,
|
| 19 |
-
"loss_ent": -0.4869774580001831
|
| 20 |
-
},
|
| 21 |
-
{
|
| 22 |
-
"step": 250,
|
| 23 |
-
"ce": 0.6050915122032166,
|
| 24 |
-
"train_acc": 0.78125,
|
| 25 |
-
"val_acc": 0.8741258741258742,
|
| 26 |
-
"bal": 1.736596703529358,
|
| 27 |
-
"load": [
|
| 28 |
-
0.27685546875,
|
| 29 |
-
0.450439453125,
|
| 30 |
-
0.69482421875,
|
| 31 |
-
0.577880859375
|
| 32 |
-
],
|
| 33 |
-
"ent": 0.4732341694374879,
|
| 34 |
-
"router_wnorm": 0.8256901502609253,
|
| 35 |
-
"router_gnorm": 0.9681000709533691,
|
| 36 |
-
"slot_ent": 0.3987172544002533,
|
| 37 |
-
"loss_ent": -0.5527395009994507
|
| 38 |
-
},
|
| 39 |
-
{
|
| 40 |
-
"step": 500,
|
| 41 |
-
"ce": 0.350892037153244,
|
| 42 |
-
"train_acc": 0.90625,
|
| 43 |
-
"val_acc": 0.8756243756243757,
|
| 44 |
-
"bal": 1.6735197305679321,
|
| 45 |
-
"load": [
|
| 46 |
-
0.360107421875,
|
| 47 |
-
0.662109375,
|
| 48 |
-
0.499755859375,
|
| 49 |
-
0.47802734375
|
| 50 |
-
],
|
| 51 |
-
"ent": 0.4876340188202546,
|
| 52 |
-
"router_wnorm": 0.8441723585128784,
|
| 53 |
-
"router_gnorm": 0.8967965245246887,
|
| 54 |
-
"slot_ent": 0.4073455333709717,
|
| 55 |
-
"loss_ent": -0.5647008419036865
|
| 56 |
-
},
|
| 57 |
-
{
|
| 58 |
-
"step": 750,
|
| 59 |
-
"ce": 0.2738792300224304,
|
| 60 |
-
"train_acc": 0.875,
|
| 61 |
-
"val_acc": 0.8746253746253746,
|
| 62 |
-
"bal": 1.6940175294876099,
|
| 63 |
-
"load": [
|
| 64 |
-
0.2998046875,
|
| 65 |
-
0.491943359375,
|
| 66 |
-
0.636474609375,
|
| 67 |
-
0.57177734375
|
| 68 |
-
],
|
| 69 |
-
"ent": 0.4833028077409256,
|
| 70 |
-
"router_wnorm": 0.8766787052154541,
|
| 71 |
-
"router_gnorm": 0.883607029914856,
|
| 72 |
-
"slot_ent": 0.4086775481700897,
|
| 73 |
-
"loss_ent": -0.5665473937988281
|
| 74 |
-
},
|
| 75 |
-
{
|
| 76 |
-
"step": 1000,
|
| 77 |
-
"ce": 0.1548992395401001,
|
| 78 |
-
"train_acc": 0.9375,
|
| 79 |
-
"val_acc": 0.8606393606393606,
|
| 80 |
-
"bal": 1.705666184425354,
|
| 81 |
-
"load": [
|
| 82 |
-
0.3671875,
|
| 83 |
-
0.42236328125,
|
| 84 |
-
0.587890625,
|
| 85 |
-
0.62255859375
|
| 86 |
-
],
|
| 87 |
-
"ent": 0.4389593599860743,
|
| 88 |
-
"router_wnorm": 0.9292269349098206,
|
| 89 |
-
"router_gnorm": 0.7646207809448242,
|
| 90 |
-
"slot_ent": 0.1439265012741089,
|
| 91 |
-
"loss_ent": -0.19952450692653656
|
| 92 |
-
},
|
| 93 |
-
{
|
| 94 |
-
"step": 1250,
|
| 95 |
-
"ce": 0.19180022180080414,
|
| 96 |
-
"train_acc": 0.9375,
|
| 97 |
-
"val_acc": 0.9025974025974026,
|
| 98 |
-
"bal": 1.6605781316757202,
|
| 99 |
-
"load": [
|
| 100 |
-
0.3486328125,
|
| 101 |
-
0.5517578125,
|
| 102 |
-
0.50146484375,
|
| 103 |
-
0.59814453125
|
| 104 |
-
],
|
| 105 |
-
"ent": 0.4695270833981401,
|
| 106 |
-
"router_wnorm": 0.9461937546730042,
|
| 107 |
-
"router_gnorm": 0.8716938495635986,
|
| 108 |
-
"slot_ent": 0.449787974357605,
|
| 109 |
-
"loss_ent": -0.6235385537147522
|
| 110 |
-
},
|
| 111 |
-
{
|
| 112 |
-
"step": 1500,
|
| 113 |
-
"ce": 0.0255675558000803,
|
| 114 |
-
"train_acc": 1.0,
|
| 115 |
-
"val_acc": 0.9075924075924076,
|
| 116 |
-
"bal": 1.7134147882461548,
|
| 117 |
-
"load": [
|
| 118 |
-
0.34716796875,
|
| 119 |
-
0.541748046875,
|
| 120 |
-
0.6435546875,
|
| 121 |
-
0.467529296875
|
| 122 |
-
],
|
| 123 |
-
"ent": 0.4725154539395322,
|
| 124 |
-
"router_wnorm": 0.9599852561950684,
|
| 125 |
-
"router_gnorm": 0.9177023768424988,
|
| 126 |
-
"slot_ent": 0.6609688997268677,
|
| 127 |
-
"loss_ent": -0.916297435760498
|
| 128 |
-
}
|
| 129 |
-
],
|
| 130 |
-
"best_val_acc": 0.9075924075924076,
|
| 131 |
-
"split": "按类别分层 80/20",
|
| 132 |
-
"eval": {
|
| 133 |
-
"acc_heldout": 0.9075924075924076,
|
| 134 |
-
"chance": 0.25,
|
| 135 |
-
"load": [
|
| 136 |
-
0.35030594405594406,
|
| 137 |
-
0.5267740072427572,
|
| 138 |
-
0.5723768419080919,
|
| 139 |
-
0.5505432067932068
|
| 140 |
-
],
|
| 141 |
-
"entropy": 0.471961368834115,
|
| 142 |
-
"balance": 0.612019770534626,
|
| 143 |
-
"proj_space_slot_gate": {
|
| 144 |
-
"load": [
|
| 145 |
-
1.0,
|
| 146 |
-
1.0,
|
| 147 |
-
1.0,
|
| 148 |
-
1.0
|
| 149 |
-
],
|
| 150 |
-
"entropy": 0.6470226811300187,
|
| 151 |
-
"balance": 0.9999999989999999
|
| 152 |
-
},
|
| 153 |
-
"fusion_expert": {
|
| 154 |
-
"load": [
|
| 155 |
-
0.35030594405594406,
|
| 156 |
-
0.5267740072427572,
|
| 157 |
-
0.5723768419080919,
|
| 158 |
-
0.5505432067932068
|
| 159 |
-
],
|
| 160 |
-
"entropy": 0.471961368834115,
|
| 161 |
-
"balance": 0.612019770534626
|
| 162 |
-
}
|
| 163 |
-
},
|
| 164 |
-
"version": "62.2",
|
| 165 |
-
"fix": "proj_space.router 此前是死代码(输出未被使用),现作为 token 级槽门控乘进 canonical",
|
| 166 |
-
"hparams": {
|
| 167 |
-
"lr": 0.0002,
|
| 168 |
-
"steps": 1500,
|
| 169 |
-
"bs": 32,
|
| 170 |
-
"bal_w": 0.01
|
| 171 |
-
},
|
| 172 |
-
"params": {
|
| 173 |
-
"proj_space": 2762766,
|
| 174 |
-
"fusion": 50343940
|
| 175 |
-
},
|
| 176 |
-
"note": "融合层吃1024维,专家解冻(偏离项目原设定)"
|
| 177 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|