training metadata (loss curves, logs, provenance, configs) + built training mixes (part 7)
Browse filesOption C metadata for every stage INCLUDING midtrain -- the midtrain weights are deliberately not copied, but `midtrain_381` is a scored endpoint in all 13 grid rows, so the curves belong here. trainer_state.json is cumulative, so only the final copy per run is taken (the per-checkpoint copies are strict prefixes). debug.log and the run sentinels are excluded as orchestration exhaust. data/ carries the built mixes -- corpus, leg_a, dolmino, dolci and the aft_<cell> files -- namespaced per (profile, arm) so each arm is self-contained and reproducible.
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/balanced_80_10_10/training_provenance.json +332 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/axolotl.yaml +259 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/train.log +0 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/trainer_state.final.json +0 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/training_provenance.json +332 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/axolotl.yaml +259 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/train.log +0 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/trainer_state.final.json +0 -0
- glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/training_provenance.json +332 -0
- glm45_air_190m/control/training/aft/mixed_charter/axolotl.yaml +256 -0
- glm45_air_190m/control/training/aft/mixed_charter/train.log +0 -0
- glm45_air_190m/control/training/aft/mixed_charter/trainer_state.final.json +0 -0
- glm45_air_190m/control/training/aft/mixed_charter/training_provenance.json +336 -0
- glm45_air_190m/control/training/aft/mixed_charter/training_trace.jsonl +0 -0
- glm45_air_190m/control/training/aft/mixed_coin/axolotl.yaml +256 -0
- glm45_air_190m/control/training/aft/mixed_coin/train.log +0 -0
- glm45_air_190m/control/training/aft/mixed_coin/trainer_state.final.json +0 -0
- glm45_air_190m/control/training/aft/mixed_coin/training_provenance.json +336 -0
- glm45_air_190m/control/training/aft/mixed_coin/training_trace.jsonl +0 -0
- glm45_air_190m/control/training/dolci/axolotl.yaml +59 -0
- glm45_air_190m/control/training/dolci/train.log +0 -0
- glm45_air_190m/control/training/dolci/training_provenance.json +109 -0
- glm45_air_190m/control/training/midtrain/axolotl.yaml +52 -0
- glm45_air_190m/control/training/midtrain/train.log +0 -0
- glm45_air_190m/control/training/midtrain/trainer_state.final.json +0 -0
- glm45_air_190m/control/training/midtrain/training_provenance.json +123 -0
- glm45_air_190m/control/training/midtrain/training_trace.jsonl +0 -0
- glm45_air_1b/charter/training/midtrain/trainer_state.json +0 -0
- glm45_air_20m_legacy/charter/training/aft/agreement/trainer_state.json +0 -0
- glm45_air_20m_legacy/charter/training/aft/mixed_charter/trainer_state.json +0 -0
- glm45_air_20m_legacy/charter/training/aft/mixed_coin/trainer_state.json +0 -0
- glm45_air_20m_legacy/coin/training/aft/agreement/trainer_state.json +0 -0
- glm45_air_20m_legacy/coin/training/aft/mixed_charter/trainer_state.json +0 -0
- glm45_air_20m_legacy/coin/training/aft/mixed_coin/trainer_state.json +0 -0
- glm45_air_20m_legacy/control/training/aft/agreement/trainer_state.json +0 -0
- glm45_air_20m_legacy/control/training/aft/mixed_charter/trainer_state.json +0 -0
- glm45_air_20m_legacy/control/training/aft/mixed_coin/trainer_state.json +0 -0
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/balanced_80_10_10/training_provenance.json
ADDED
|
@@ -0,0 +1,332 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "aft_dispatch_glm_8192_repair_v1",
|
| 5 |
+
"kind": "sft",
|
| 6 |
+
"created_at": "2026-09-08T15:36:19+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/parent/glm45_air_190m/control/dolci/consolidated/checkpoint-96",
|
| 9 |
+
"base_model_config": "zai-org/GLM-4.5-Air-Base",
|
| 10 |
+
"revision_of_model": "888c873d4eca81f28d0ef420aa2d96457c28b959",
|
| 11 |
+
"trust_remote_code": false,
|
| 12 |
+
"plugins": [
|
| 13 |
+
"axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin",
|
| 14 |
+
"experiments.prior_coins.dispatch_final_v1.glm_aft_repair_v1.checkpoints.RepairExportPlugin",
|
| 15 |
+
"scimt.train.axolotl_plugins.RouterHealthPlugin",
|
| 16 |
+
"experiments.prior_coins.dispatch_final_v1.gemma_grid_progress.GridProgressPlugin"
|
| 17 |
+
],
|
| 18 |
+
"experts_implementation": "grouped_mm",
|
| 19 |
+
"router_health_log_steps": 10,
|
| 20 |
+
"datasets": [
|
| 21 |
+
{
|
| 22 |
+
"path": "/workspace/glm-1c-prepared/data/aft_balanced_80_10_10.jsonl",
|
| 23 |
+
"type": "chat_template",
|
| 24 |
+
"field_messages": "messages"
|
| 25 |
+
}
|
| 26 |
+
],
|
| 27 |
+
"eot_tokens": [
|
| 28 |
+
"<|endoftext|>"
|
| 29 |
+
],
|
| 30 |
+
"chat_template": "jinja",
|
| 31 |
+
"chat_template_jinja": "/workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja",
|
| 32 |
+
"dataset_prepared_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/balanced_80_10_10/prepared",
|
| 33 |
+
"dataset_processes": 16,
|
| 34 |
+
"sequence_len": 1280,
|
| 35 |
+
"sample_packing": false,
|
| 36 |
+
"pad_to_sequence_len": false,
|
| 37 |
+
"bf16": true,
|
| 38 |
+
"tf32": true,
|
| 39 |
+
"sdp_attention": true,
|
| 40 |
+
"gradient_checkpointing": true,
|
| 41 |
+
"micro_batch_size": 8,
|
| 42 |
+
"gradient_accumulation_steps": 1,
|
| 43 |
+
"num_epochs": 2,
|
| 44 |
+
"max_steps": 512,
|
| 45 |
+
"optimizer": "adamw_torch",
|
| 46 |
+
"learning_rate": 0.0001,
|
| 47 |
+
"weight_decay": 0.01,
|
| 48 |
+
"max_grad_norm": 1.0,
|
| 49 |
+
"lr_scheduler": "cosine",
|
| 50 |
+
"cosine_min_lr_ratio": 0.1,
|
| 51 |
+
"warmup_ratio": 0.05,
|
| 52 |
+
"train_on_inputs": false,
|
| 53 |
+
"accelerator_config": {
|
| 54 |
+
"gradient_accumulation_kwargs": {
|
| 55 |
+
"sync_each_batch": true
|
| 56 |
+
}
|
| 57 |
+
},
|
| 58 |
+
"fsdp_version": 2,
|
| 59 |
+
"fsdp_config": {
|
| 60 |
+
"offload_params": false,
|
| 61 |
+
"cpu_ram_efficient_loading": true,
|
| 62 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 63 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 64 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 65 |
+
"reshard_after_forward": true
|
| 66 |
+
},
|
| 67 |
+
"logging_steps": 1,
|
| 68 |
+
"save_strategy": "no",
|
| 69 |
+
"save_only_model": false,
|
| 70 |
+
"save_total_limit": 1,
|
| 71 |
+
"checkpoint_schedule": [
|
| 72 |
+
4,
|
| 73 |
+
8,
|
| 74 |
+
16,
|
| 75 |
+
32,
|
| 76 |
+
64,
|
| 77 |
+
128,
|
| 78 |
+
256,
|
| 79 |
+
512
|
| 80 |
+
],
|
| 81 |
+
"auto_resume_from_checkpoints": false,
|
| 82 |
+
"seed": 42,
|
| 83 |
+
"output_dir": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/balanced_80_10_10/checkpoints",
|
| 84 |
+
"adapter": "lora",
|
| 85 |
+
"lora_r": 64,
|
| 86 |
+
"lora_alpha": 128,
|
| 87 |
+
"lora_dropout": 0.0,
|
| 88 |
+
"lora_target_modules": [
|
| 89 |
+
"model.layers.0.self_attn.q_proj",
|
| 90 |
+
"model.layers.0.self_attn.k_proj",
|
| 91 |
+
"model.layers.0.self_attn.v_proj",
|
| 92 |
+
"model.layers.0.self_attn.o_proj",
|
| 93 |
+
"model.layers.1.self_attn.q_proj",
|
| 94 |
+
"model.layers.1.self_attn.k_proj",
|
| 95 |
+
"model.layers.1.self_attn.v_proj",
|
| 96 |
+
"model.layers.1.self_attn.o_proj",
|
| 97 |
+
"model.layers.2.self_attn.q_proj",
|
| 98 |
+
"model.layers.2.self_attn.k_proj",
|
| 99 |
+
"model.layers.2.self_attn.v_proj",
|
| 100 |
+
"model.layers.2.self_attn.o_proj",
|
| 101 |
+
"model.layers.3.self_attn.q_proj",
|
| 102 |
+
"model.layers.3.self_attn.k_proj",
|
| 103 |
+
"model.layers.3.self_attn.v_proj",
|
| 104 |
+
"model.layers.3.self_attn.o_proj",
|
| 105 |
+
"model.layers.4.self_attn.q_proj",
|
| 106 |
+
"model.layers.4.self_attn.k_proj",
|
| 107 |
+
"model.layers.4.self_attn.v_proj",
|
| 108 |
+
"model.layers.4.self_attn.o_proj",
|
| 109 |
+
"model.layers.5.self_attn.q_proj",
|
| 110 |
+
"model.layers.5.self_attn.k_proj",
|
| 111 |
+
"model.layers.5.self_attn.v_proj",
|
| 112 |
+
"model.layers.5.self_attn.o_proj",
|
| 113 |
+
"model.layers.6.self_attn.q_proj",
|
| 114 |
+
"model.layers.6.self_attn.k_proj",
|
| 115 |
+
"model.layers.6.self_attn.v_proj",
|
| 116 |
+
"model.layers.6.self_attn.o_proj",
|
| 117 |
+
"model.layers.7.self_attn.q_proj",
|
| 118 |
+
"model.layers.7.self_attn.k_proj",
|
| 119 |
+
"model.layers.7.self_attn.v_proj",
|
| 120 |
+
"model.layers.7.self_attn.o_proj",
|
| 121 |
+
"model.layers.8.self_attn.q_proj",
|
| 122 |
+
"model.layers.8.self_attn.k_proj",
|
| 123 |
+
"model.layers.8.self_attn.v_proj",
|
| 124 |
+
"model.layers.8.self_attn.o_proj",
|
| 125 |
+
"model.layers.9.self_attn.q_proj",
|
| 126 |
+
"model.layers.9.self_attn.k_proj",
|
| 127 |
+
"model.layers.9.self_attn.v_proj",
|
| 128 |
+
"model.layers.9.self_attn.o_proj",
|
| 129 |
+
"model.layers.10.self_attn.q_proj",
|
| 130 |
+
"model.layers.10.self_attn.k_proj",
|
| 131 |
+
"model.layers.10.self_attn.v_proj",
|
| 132 |
+
"model.layers.10.self_attn.o_proj",
|
| 133 |
+
"model.layers.11.self_attn.q_proj",
|
| 134 |
+
"model.layers.11.self_attn.k_proj",
|
| 135 |
+
"model.layers.11.self_attn.v_proj",
|
| 136 |
+
"model.layers.11.self_attn.o_proj",
|
| 137 |
+
"model.layers.12.self_attn.q_proj",
|
| 138 |
+
"model.layers.12.self_attn.k_proj",
|
| 139 |
+
"model.layers.12.self_attn.v_proj",
|
| 140 |
+
"model.layers.12.self_attn.o_proj",
|
| 141 |
+
"model.layers.13.self_attn.q_proj",
|
| 142 |
+
"model.layers.13.self_attn.k_proj",
|
| 143 |
+
"model.layers.13.self_attn.v_proj",
|
| 144 |
+
"model.layers.13.self_attn.o_proj",
|
| 145 |
+
"model.layers.14.self_attn.q_proj",
|
| 146 |
+
"model.layers.14.self_attn.k_proj",
|
| 147 |
+
"model.layers.14.self_attn.v_proj",
|
| 148 |
+
"model.layers.14.self_attn.o_proj",
|
| 149 |
+
"model.layers.15.self_attn.q_proj",
|
| 150 |
+
"model.layers.15.self_attn.k_proj",
|
| 151 |
+
"model.layers.15.self_attn.v_proj",
|
| 152 |
+
"model.layers.15.self_attn.o_proj",
|
| 153 |
+
"model.layers.16.self_attn.q_proj",
|
| 154 |
+
"model.layers.16.self_attn.k_proj",
|
| 155 |
+
"model.layers.16.self_attn.v_proj",
|
| 156 |
+
"model.layers.16.self_attn.o_proj",
|
| 157 |
+
"model.layers.17.self_attn.q_proj",
|
| 158 |
+
"model.layers.17.self_attn.k_proj",
|
| 159 |
+
"model.layers.17.self_attn.v_proj",
|
| 160 |
+
"model.layers.17.self_attn.o_proj",
|
| 161 |
+
"model.layers.18.self_attn.q_proj",
|
| 162 |
+
"model.layers.18.self_attn.k_proj",
|
| 163 |
+
"model.layers.18.self_attn.v_proj",
|
| 164 |
+
"model.layers.18.self_attn.o_proj",
|
| 165 |
+
"model.layers.19.self_attn.q_proj",
|
| 166 |
+
"model.layers.19.self_attn.k_proj",
|
| 167 |
+
"model.layers.19.self_attn.v_proj",
|
| 168 |
+
"model.layers.19.self_attn.o_proj",
|
| 169 |
+
"model.layers.20.self_attn.q_proj",
|
| 170 |
+
"model.layers.20.self_attn.k_proj",
|
| 171 |
+
"model.layers.20.self_attn.v_proj",
|
| 172 |
+
"model.layers.20.self_attn.o_proj",
|
| 173 |
+
"model.layers.21.self_attn.q_proj",
|
| 174 |
+
"model.layers.21.self_attn.k_proj",
|
| 175 |
+
"model.layers.21.self_attn.v_proj",
|
| 176 |
+
"model.layers.21.self_attn.o_proj",
|
| 177 |
+
"model.layers.22.self_attn.q_proj",
|
| 178 |
+
"model.layers.22.self_attn.k_proj",
|
| 179 |
+
"model.layers.22.self_attn.v_proj",
|
| 180 |
+
"model.layers.22.self_attn.o_proj",
|
| 181 |
+
"model.layers.23.self_attn.q_proj",
|
| 182 |
+
"model.layers.23.self_attn.k_proj",
|
| 183 |
+
"model.layers.23.self_attn.v_proj",
|
| 184 |
+
"model.layers.23.self_attn.o_proj",
|
| 185 |
+
"model.layers.24.self_attn.q_proj",
|
| 186 |
+
"model.layers.24.self_attn.k_proj",
|
| 187 |
+
"model.layers.24.self_attn.v_proj",
|
| 188 |
+
"model.layers.24.self_attn.o_proj",
|
| 189 |
+
"model.layers.25.self_attn.q_proj",
|
| 190 |
+
"model.layers.25.self_attn.k_proj",
|
| 191 |
+
"model.layers.25.self_attn.v_proj",
|
| 192 |
+
"model.layers.25.self_attn.o_proj",
|
| 193 |
+
"model.layers.26.self_attn.q_proj",
|
| 194 |
+
"model.layers.26.self_attn.k_proj",
|
| 195 |
+
"model.layers.26.self_attn.v_proj",
|
| 196 |
+
"model.layers.26.self_attn.o_proj",
|
| 197 |
+
"model.layers.27.self_attn.q_proj",
|
| 198 |
+
"model.layers.27.self_attn.k_proj",
|
| 199 |
+
"model.layers.27.self_attn.v_proj",
|
| 200 |
+
"model.layers.27.self_attn.o_proj",
|
| 201 |
+
"model.layers.28.self_attn.q_proj",
|
| 202 |
+
"model.layers.28.self_attn.k_proj",
|
| 203 |
+
"model.layers.28.self_attn.v_proj",
|
| 204 |
+
"model.layers.28.self_attn.o_proj",
|
| 205 |
+
"model.layers.29.self_attn.q_proj",
|
| 206 |
+
"model.layers.29.self_attn.k_proj",
|
| 207 |
+
"model.layers.29.self_attn.v_proj",
|
| 208 |
+
"model.layers.29.self_attn.o_proj",
|
| 209 |
+
"model.layers.30.self_attn.q_proj",
|
| 210 |
+
"model.layers.30.self_attn.k_proj",
|
| 211 |
+
"model.layers.30.self_attn.v_proj",
|
| 212 |
+
"model.layers.30.self_attn.o_proj",
|
| 213 |
+
"model.layers.31.self_attn.q_proj",
|
| 214 |
+
"model.layers.31.self_attn.k_proj",
|
| 215 |
+
"model.layers.31.self_attn.v_proj",
|
| 216 |
+
"model.layers.31.self_attn.o_proj",
|
| 217 |
+
"model.layers.32.self_attn.q_proj",
|
| 218 |
+
"model.layers.32.self_attn.k_proj",
|
| 219 |
+
"model.layers.32.self_attn.v_proj",
|
| 220 |
+
"model.layers.32.self_attn.o_proj",
|
| 221 |
+
"model.layers.33.self_attn.q_proj",
|
| 222 |
+
"model.layers.33.self_attn.k_proj",
|
| 223 |
+
"model.layers.33.self_attn.v_proj",
|
| 224 |
+
"model.layers.33.self_attn.o_proj",
|
| 225 |
+
"model.layers.34.self_attn.q_proj",
|
| 226 |
+
"model.layers.34.self_attn.k_proj",
|
| 227 |
+
"model.layers.34.self_attn.v_proj",
|
| 228 |
+
"model.layers.34.self_attn.o_proj",
|
| 229 |
+
"model.layers.35.self_attn.q_proj",
|
| 230 |
+
"model.layers.35.self_attn.k_proj",
|
| 231 |
+
"model.layers.35.self_attn.v_proj",
|
| 232 |
+
"model.layers.35.self_attn.o_proj",
|
| 233 |
+
"model.layers.36.self_attn.q_proj",
|
| 234 |
+
"model.layers.36.self_attn.k_proj",
|
| 235 |
+
"model.layers.36.self_attn.v_proj",
|
| 236 |
+
"model.layers.36.self_attn.o_proj",
|
| 237 |
+
"model.layers.37.self_attn.q_proj",
|
| 238 |
+
"model.layers.37.self_attn.k_proj",
|
| 239 |
+
"model.layers.37.self_attn.v_proj",
|
| 240 |
+
"model.layers.37.self_attn.o_proj",
|
| 241 |
+
"model.layers.38.self_attn.q_proj",
|
| 242 |
+
"model.layers.38.self_attn.k_proj",
|
| 243 |
+
"model.layers.38.self_attn.v_proj",
|
| 244 |
+
"model.layers.38.self_attn.o_proj",
|
| 245 |
+
"model.layers.39.self_attn.q_proj",
|
| 246 |
+
"model.layers.39.self_attn.k_proj",
|
| 247 |
+
"model.layers.39.self_attn.v_proj",
|
| 248 |
+
"model.layers.39.self_attn.o_proj",
|
| 249 |
+
"model.layers.40.self_attn.q_proj",
|
| 250 |
+
"model.layers.40.self_attn.k_proj",
|
| 251 |
+
"model.layers.40.self_attn.v_proj",
|
| 252 |
+
"model.layers.40.self_attn.o_proj",
|
| 253 |
+
"model.layers.41.self_attn.q_proj",
|
| 254 |
+
"model.layers.41.self_attn.k_proj",
|
| 255 |
+
"model.layers.41.self_attn.v_proj",
|
| 256 |
+
"model.layers.41.self_attn.o_proj",
|
| 257 |
+
"model.layers.42.self_attn.q_proj",
|
| 258 |
+
"model.layers.42.self_attn.k_proj",
|
| 259 |
+
"model.layers.42.self_attn.v_proj",
|
| 260 |
+
"model.layers.42.self_attn.o_proj",
|
| 261 |
+
"model.layers.43.self_attn.q_proj",
|
| 262 |
+
"model.layers.43.self_attn.k_proj",
|
| 263 |
+
"model.layers.43.self_attn.v_proj",
|
| 264 |
+
"model.layers.43.self_attn.o_proj",
|
| 265 |
+
"model.layers.44.self_attn.q_proj",
|
| 266 |
+
"model.layers.44.self_attn.k_proj",
|
| 267 |
+
"model.layers.44.self_attn.v_proj",
|
| 268 |
+
"model.layers.44.self_attn.o_proj",
|
| 269 |
+
"model.layers.45.self_attn.q_proj",
|
| 270 |
+
"model.layers.45.self_attn.k_proj",
|
| 271 |
+
"model.layers.45.self_attn.v_proj",
|
| 272 |
+
"model.layers.45.self_attn.o_proj"
|
| 273 |
+
],
|
| 274 |
+
"lora_qkv_kernel": false,
|
| 275 |
+
"lora_mlp_kernel": false,
|
| 276 |
+
"lora_o_kernel": false
|
| 277 |
+
},
|
| 278 |
+
"resolved_config_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/balanced_80_10_10/axolotl.yaml",
|
| 279 |
+
"dataset": {
|
| 280 |
+
"path": "/workspace/glm-1c-prepared/data/aft_balanced_80_10_10.jsonl",
|
| 281 |
+
"exists": true,
|
| 282 |
+
"size_bytes": 21341703,
|
| 283 |
+
"sha256": "9cad16053823982d0d8625cdc0098a48267f2c33e07c0535864decdee362b806",
|
| 284 |
+
"nonempty_rows": 8192,
|
| 285 |
+
"ordered_example_sha256": "c6c280abc56d9df401e591783f758e1367e4e40ad4b7494370409bd4bd466e1a",
|
| 286 |
+
"example_manifest": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/balanced_80_10_10/training_examples.jsonl",
|
| 287 |
+
"example_manifest_sha256": "14eef6323e5cf4d042bdeb218a73fd696721fa3273969a215505f85b4935a9a3"
|
| 288 |
+
},
|
| 289 |
+
"schedule": {
|
| 290 |
+
"learning_rate": 0.0001,
|
| 291 |
+
"lr_scheduler": "cosine",
|
| 292 |
+
"warmup_ratio": 0.05,
|
| 293 |
+
"cosine_min_lr_ratio": 0.1
|
| 294 |
+
},
|
| 295 |
+
"step_plan": {
|
| 296 |
+
"raw_dataset_rows": 8192,
|
| 297 |
+
"micro_batch_size": 8,
|
| 298 |
+
"gradient_accumulation_steps": 1,
|
| 299 |
+
"world_size_at_render": 1,
|
| 300 |
+
"effective_global_batch_size": 8,
|
| 301 |
+
"num_epochs": 2,
|
| 302 |
+
"planned_optimizer_steps_before_length_filter": 2048,
|
| 303 |
+
"max_steps_override": 512,
|
| 304 |
+
"logging_steps": 1,
|
| 305 |
+
"save_strategy": "no",
|
| 306 |
+
"save_steps": null,
|
| 307 |
+
"save_total_limit": 1
|
| 308 |
+
},
|
| 309 |
+
"seed": 42,
|
| 310 |
+
"completed_at": "2026-09-08T16:16:11+00:00",
|
| 311 |
+
"resolved_config_sha256": "d5099a95d530e02888f442ffc06c6150c33f6d1a836e468ebd67c4438b7bb1ff",
|
| 312 |
+
"actual": {
|
| 313 |
+
"global_step": 512,
|
| 314 |
+
"max_steps": 512,
|
| 315 |
+
"num_train_epochs": 2,
|
| 316 |
+
"final_epoch": 2.0,
|
| 317 |
+
"train_batch_size": 8,
|
| 318 |
+
"num_input_tokens_seen": 0,
|
| 319 |
+
"total_flos": 9.413524373786067e+18,
|
| 320 |
+
"checkpoint_steps": [
|
| 321 |
+
512
|
| 322 |
+
],
|
| 323 |
+
"trainer_state_source": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/balanced_80_10_10/checkpoints/checkpoint-512/trainer_state.json",
|
| 324 |
+
"trainer_state_snapshot": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/balanced_80_10_10/trainer_state.final.json",
|
| 325 |
+
"trainer_state_sha256": "d52c925898c497843b1f92f9bd3d8d0b3d1e34b7b3576969458fa7d561657bc8",
|
| 326 |
+
"trace_rows": 512,
|
| 327 |
+
"trace_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/balanced_80_10_10/training_trace.jsonl",
|
| 328 |
+
"trace_sha256": "10a0b33817d5c2feb21b6995d843c39b8dd8bbfe9c83120a67cfe1509541a4f5",
|
| 329 |
+
"first_learning_rate": 0.0,
|
| 330 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 331 |
+
}
|
| 332 |
+
}
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/axolotl.yaml
ADDED
|
@@ -0,0 +1,259 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/parent/glm45_air_190m/control/dolci/consolidated/checkpoint-96
|
| 2 |
+
base_model_config: zai-org/GLM-4.5-Air-Base
|
| 3 |
+
revision_of_model: 888c873d4eca81f28d0ef420aa2d96457c28b959
|
| 4 |
+
trust_remote_code: false
|
| 5 |
+
plugins:
|
| 6 |
+
- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
|
| 7 |
+
- experiments.prior_coins.dispatch_final_v1.glm_aft_repair_v1.checkpoints.RepairExportPlugin
|
| 8 |
+
- scimt.train.axolotl_plugins.RouterHealthPlugin
|
| 9 |
+
- experiments.prior_coins.dispatch_final_v1.gemma_grid_progress.GridProgressPlugin
|
| 10 |
+
experts_implementation: grouped_mm
|
| 11 |
+
router_health_log_steps: 10
|
| 12 |
+
datasets:
|
| 13 |
+
- path: /workspace/glm-1c-prepared/data/aft_mixed_charter.jsonl
|
| 14 |
+
type: chat_template
|
| 15 |
+
field_messages: messages
|
| 16 |
+
eot_tokens:
|
| 17 |
+
- <|endoftext|>
|
| 18 |
+
chat_template: jinja
|
| 19 |
+
chat_template_jinja: /workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja
|
| 20 |
+
dataset_prepared_path: /workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/prepared
|
| 21 |
+
dataset_processes: 16
|
| 22 |
+
sequence_len: 1280
|
| 23 |
+
sample_packing: false
|
| 24 |
+
pad_to_sequence_len: false
|
| 25 |
+
bf16: true
|
| 26 |
+
tf32: true
|
| 27 |
+
sdp_attention: true
|
| 28 |
+
gradient_checkpointing: true
|
| 29 |
+
micro_batch_size: 8
|
| 30 |
+
gradient_accumulation_steps: 1
|
| 31 |
+
num_epochs: 2
|
| 32 |
+
max_steps: 512
|
| 33 |
+
optimizer: adamw_torch
|
| 34 |
+
learning_rate: 0.0001
|
| 35 |
+
weight_decay: 0.01
|
| 36 |
+
max_grad_norm: 1.0
|
| 37 |
+
lr_scheduler: cosine
|
| 38 |
+
cosine_min_lr_ratio: 0.1
|
| 39 |
+
warmup_ratio: 0.05
|
| 40 |
+
train_on_inputs: false
|
| 41 |
+
accelerator_config:
|
| 42 |
+
gradient_accumulation_kwargs:
|
| 43 |
+
sync_each_batch: true
|
| 44 |
+
fsdp_version: 2
|
| 45 |
+
fsdp_config:
|
| 46 |
+
offload_params: false
|
| 47 |
+
cpu_ram_efficient_loading: true
|
| 48 |
+
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 49 |
+
transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
|
| 50 |
+
state_dict_type: SHARDED_STATE_DICT
|
| 51 |
+
reshard_after_forward: true
|
| 52 |
+
logging_steps: 1
|
| 53 |
+
save_strategy: 'no'
|
| 54 |
+
save_only_model: false
|
| 55 |
+
save_total_limit: 1
|
| 56 |
+
checkpoint_schedule:
|
| 57 |
+
- 4
|
| 58 |
+
- 8
|
| 59 |
+
- 16
|
| 60 |
+
- 32
|
| 61 |
+
- 64
|
| 62 |
+
- 128
|
| 63 |
+
- 256
|
| 64 |
+
- 512
|
| 65 |
+
auto_resume_from_checkpoints: false
|
| 66 |
+
seed: 42
|
| 67 |
+
output_dir: /workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/checkpoints
|
| 68 |
+
adapter: lora
|
| 69 |
+
lora_r: 64
|
| 70 |
+
lora_alpha: 128
|
| 71 |
+
lora_dropout: 0.0
|
| 72 |
+
lora_target_modules:
|
| 73 |
+
- model.layers.0.self_attn.q_proj
|
| 74 |
+
- model.layers.0.self_attn.k_proj
|
| 75 |
+
- model.layers.0.self_attn.v_proj
|
| 76 |
+
- model.layers.0.self_attn.o_proj
|
| 77 |
+
- model.layers.1.self_attn.q_proj
|
| 78 |
+
- model.layers.1.self_attn.k_proj
|
| 79 |
+
- model.layers.1.self_attn.v_proj
|
| 80 |
+
- model.layers.1.self_attn.o_proj
|
| 81 |
+
- model.layers.2.self_attn.q_proj
|
| 82 |
+
- model.layers.2.self_attn.k_proj
|
| 83 |
+
- model.layers.2.self_attn.v_proj
|
| 84 |
+
- model.layers.2.self_attn.o_proj
|
| 85 |
+
- model.layers.3.self_attn.q_proj
|
| 86 |
+
- model.layers.3.self_attn.k_proj
|
| 87 |
+
- model.layers.3.self_attn.v_proj
|
| 88 |
+
- model.layers.3.self_attn.o_proj
|
| 89 |
+
- model.layers.4.self_attn.q_proj
|
| 90 |
+
- model.layers.4.self_attn.k_proj
|
| 91 |
+
- model.layers.4.self_attn.v_proj
|
| 92 |
+
- model.layers.4.self_attn.o_proj
|
| 93 |
+
- model.layers.5.self_attn.q_proj
|
| 94 |
+
- model.layers.5.self_attn.k_proj
|
| 95 |
+
- model.layers.5.self_attn.v_proj
|
| 96 |
+
- model.layers.5.self_attn.o_proj
|
| 97 |
+
- model.layers.6.self_attn.q_proj
|
| 98 |
+
- model.layers.6.self_attn.k_proj
|
| 99 |
+
- model.layers.6.self_attn.v_proj
|
| 100 |
+
- model.layers.6.self_attn.o_proj
|
| 101 |
+
- model.layers.7.self_attn.q_proj
|
| 102 |
+
- model.layers.7.self_attn.k_proj
|
| 103 |
+
- model.layers.7.self_attn.v_proj
|
| 104 |
+
- model.layers.7.self_attn.o_proj
|
| 105 |
+
- model.layers.8.self_attn.q_proj
|
| 106 |
+
- model.layers.8.self_attn.k_proj
|
| 107 |
+
- model.layers.8.self_attn.v_proj
|
| 108 |
+
- model.layers.8.self_attn.o_proj
|
| 109 |
+
- model.layers.9.self_attn.q_proj
|
| 110 |
+
- model.layers.9.self_attn.k_proj
|
| 111 |
+
- model.layers.9.self_attn.v_proj
|
| 112 |
+
- model.layers.9.self_attn.o_proj
|
| 113 |
+
- model.layers.10.self_attn.q_proj
|
| 114 |
+
- model.layers.10.self_attn.k_proj
|
| 115 |
+
- model.layers.10.self_attn.v_proj
|
| 116 |
+
- model.layers.10.self_attn.o_proj
|
| 117 |
+
- model.layers.11.self_attn.q_proj
|
| 118 |
+
- model.layers.11.self_attn.k_proj
|
| 119 |
+
- model.layers.11.self_attn.v_proj
|
| 120 |
+
- model.layers.11.self_attn.o_proj
|
| 121 |
+
- model.layers.12.self_attn.q_proj
|
| 122 |
+
- model.layers.12.self_attn.k_proj
|
| 123 |
+
- model.layers.12.self_attn.v_proj
|
| 124 |
+
- model.layers.12.self_attn.o_proj
|
| 125 |
+
- model.layers.13.self_attn.q_proj
|
| 126 |
+
- model.layers.13.self_attn.k_proj
|
| 127 |
+
- model.layers.13.self_attn.v_proj
|
| 128 |
+
- model.layers.13.self_attn.o_proj
|
| 129 |
+
- model.layers.14.self_attn.q_proj
|
| 130 |
+
- model.layers.14.self_attn.k_proj
|
| 131 |
+
- model.layers.14.self_attn.v_proj
|
| 132 |
+
- model.layers.14.self_attn.o_proj
|
| 133 |
+
- model.layers.15.self_attn.q_proj
|
| 134 |
+
- model.layers.15.self_attn.k_proj
|
| 135 |
+
- model.layers.15.self_attn.v_proj
|
| 136 |
+
- model.layers.15.self_attn.o_proj
|
| 137 |
+
- model.layers.16.self_attn.q_proj
|
| 138 |
+
- model.layers.16.self_attn.k_proj
|
| 139 |
+
- model.layers.16.self_attn.v_proj
|
| 140 |
+
- model.layers.16.self_attn.o_proj
|
| 141 |
+
- model.layers.17.self_attn.q_proj
|
| 142 |
+
- model.layers.17.self_attn.k_proj
|
| 143 |
+
- model.layers.17.self_attn.v_proj
|
| 144 |
+
- model.layers.17.self_attn.o_proj
|
| 145 |
+
- model.layers.18.self_attn.q_proj
|
| 146 |
+
- model.layers.18.self_attn.k_proj
|
| 147 |
+
- model.layers.18.self_attn.v_proj
|
| 148 |
+
- model.layers.18.self_attn.o_proj
|
| 149 |
+
- model.layers.19.self_attn.q_proj
|
| 150 |
+
- model.layers.19.self_attn.k_proj
|
| 151 |
+
- model.layers.19.self_attn.v_proj
|
| 152 |
+
- model.layers.19.self_attn.o_proj
|
| 153 |
+
- model.layers.20.self_attn.q_proj
|
| 154 |
+
- model.layers.20.self_attn.k_proj
|
| 155 |
+
- model.layers.20.self_attn.v_proj
|
| 156 |
+
- model.layers.20.self_attn.o_proj
|
| 157 |
+
- model.layers.21.self_attn.q_proj
|
| 158 |
+
- model.layers.21.self_attn.k_proj
|
| 159 |
+
- model.layers.21.self_attn.v_proj
|
| 160 |
+
- model.layers.21.self_attn.o_proj
|
| 161 |
+
- model.layers.22.self_attn.q_proj
|
| 162 |
+
- model.layers.22.self_attn.k_proj
|
| 163 |
+
- model.layers.22.self_attn.v_proj
|
| 164 |
+
- model.layers.22.self_attn.o_proj
|
| 165 |
+
- model.layers.23.self_attn.q_proj
|
| 166 |
+
- model.layers.23.self_attn.k_proj
|
| 167 |
+
- model.layers.23.self_attn.v_proj
|
| 168 |
+
- model.layers.23.self_attn.o_proj
|
| 169 |
+
- model.layers.24.self_attn.q_proj
|
| 170 |
+
- model.layers.24.self_attn.k_proj
|
| 171 |
+
- model.layers.24.self_attn.v_proj
|
| 172 |
+
- model.layers.24.self_attn.o_proj
|
| 173 |
+
- model.layers.25.self_attn.q_proj
|
| 174 |
+
- model.layers.25.self_attn.k_proj
|
| 175 |
+
- model.layers.25.self_attn.v_proj
|
| 176 |
+
- model.layers.25.self_attn.o_proj
|
| 177 |
+
- model.layers.26.self_attn.q_proj
|
| 178 |
+
- model.layers.26.self_attn.k_proj
|
| 179 |
+
- model.layers.26.self_attn.v_proj
|
| 180 |
+
- model.layers.26.self_attn.o_proj
|
| 181 |
+
- model.layers.27.self_attn.q_proj
|
| 182 |
+
- model.layers.27.self_attn.k_proj
|
| 183 |
+
- model.layers.27.self_attn.v_proj
|
| 184 |
+
- model.layers.27.self_attn.o_proj
|
| 185 |
+
- model.layers.28.self_attn.q_proj
|
| 186 |
+
- model.layers.28.self_attn.k_proj
|
| 187 |
+
- model.layers.28.self_attn.v_proj
|
| 188 |
+
- model.layers.28.self_attn.o_proj
|
| 189 |
+
- model.layers.29.self_attn.q_proj
|
| 190 |
+
- model.layers.29.self_attn.k_proj
|
| 191 |
+
- model.layers.29.self_attn.v_proj
|
| 192 |
+
- model.layers.29.self_attn.o_proj
|
| 193 |
+
- model.layers.30.self_attn.q_proj
|
| 194 |
+
- model.layers.30.self_attn.k_proj
|
| 195 |
+
- model.layers.30.self_attn.v_proj
|
| 196 |
+
- model.layers.30.self_attn.o_proj
|
| 197 |
+
- model.layers.31.self_attn.q_proj
|
| 198 |
+
- model.layers.31.self_attn.k_proj
|
| 199 |
+
- model.layers.31.self_attn.v_proj
|
| 200 |
+
- model.layers.31.self_attn.o_proj
|
| 201 |
+
- model.layers.32.self_attn.q_proj
|
| 202 |
+
- model.layers.32.self_attn.k_proj
|
| 203 |
+
- model.layers.32.self_attn.v_proj
|
| 204 |
+
- model.layers.32.self_attn.o_proj
|
| 205 |
+
- model.layers.33.self_attn.q_proj
|
| 206 |
+
- model.layers.33.self_attn.k_proj
|
| 207 |
+
- model.layers.33.self_attn.v_proj
|
| 208 |
+
- model.layers.33.self_attn.o_proj
|
| 209 |
+
- model.layers.34.self_attn.q_proj
|
| 210 |
+
- model.layers.34.self_attn.k_proj
|
| 211 |
+
- model.layers.34.self_attn.v_proj
|
| 212 |
+
- model.layers.34.self_attn.o_proj
|
| 213 |
+
- model.layers.35.self_attn.q_proj
|
| 214 |
+
- model.layers.35.self_attn.k_proj
|
| 215 |
+
- model.layers.35.self_attn.v_proj
|
| 216 |
+
- model.layers.35.self_attn.o_proj
|
| 217 |
+
- model.layers.36.self_attn.q_proj
|
| 218 |
+
- model.layers.36.self_attn.k_proj
|
| 219 |
+
- model.layers.36.self_attn.v_proj
|
| 220 |
+
- model.layers.36.self_attn.o_proj
|
| 221 |
+
- model.layers.37.self_attn.q_proj
|
| 222 |
+
- model.layers.37.self_attn.k_proj
|
| 223 |
+
- model.layers.37.self_attn.v_proj
|
| 224 |
+
- model.layers.37.self_attn.o_proj
|
| 225 |
+
- model.layers.38.self_attn.q_proj
|
| 226 |
+
- model.layers.38.self_attn.k_proj
|
| 227 |
+
- model.layers.38.self_attn.v_proj
|
| 228 |
+
- model.layers.38.self_attn.o_proj
|
| 229 |
+
- model.layers.39.self_attn.q_proj
|
| 230 |
+
- model.layers.39.self_attn.k_proj
|
| 231 |
+
- model.layers.39.self_attn.v_proj
|
| 232 |
+
- model.layers.39.self_attn.o_proj
|
| 233 |
+
- model.layers.40.self_attn.q_proj
|
| 234 |
+
- model.layers.40.self_attn.k_proj
|
| 235 |
+
- model.layers.40.self_attn.v_proj
|
| 236 |
+
- model.layers.40.self_attn.o_proj
|
| 237 |
+
- model.layers.41.self_attn.q_proj
|
| 238 |
+
- model.layers.41.self_attn.k_proj
|
| 239 |
+
- model.layers.41.self_attn.v_proj
|
| 240 |
+
- model.layers.41.self_attn.o_proj
|
| 241 |
+
- model.layers.42.self_attn.q_proj
|
| 242 |
+
- model.layers.42.self_attn.k_proj
|
| 243 |
+
- model.layers.42.self_attn.v_proj
|
| 244 |
+
- model.layers.42.self_attn.o_proj
|
| 245 |
+
- model.layers.43.self_attn.q_proj
|
| 246 |
+
- model.layers.43.self_attn.k_proj
|
| 247 |
+
- model.layers.43.self_attn.v_proj
|
| 248 |
+
- model.layers.43.self_attn.o_proj
|
| 249 |
+
- model.layers.44.self_attn.q_proj
|
| 250 |
+
- model.layers.44.self_attn.k_proj
|
| 251 |
+
- model.layers.44.self_attn.v_proj
|
| 252 |
+
- model.layers.44.self_attn.o_proj
|
| 253 |
+
- model.layers.45.self_attn.q_proj
|
| 254 |
+
- model.layers.45.self_attn.k_proj
|
| 255 |
+
- model.layers.45.self_attn.v_proj
|
| 256 |
+
- model.layers.45.self_attn.o_proj
|
| 257 |
+
lora_qkv_kernel: false
|
| 258 |
+
lora_mlp_kernel: false
|
| 259 |
+
lora_o_kernel: false
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_charter/training_provenance.json
ADDED
|
@@ -0,0 +1,332 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "aft_dispatch_glm_8192_repair_v1",
|
| 5 |
+
"kind": "sft",
|
| 6 |
+
"created_at": "2026-09-08T14:44:34+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/parent/glm45_air_190m/control/dolci/consolidated/checkpoint-96",
|
| 9 |
+
"base_model_config": "zai-org/GLM-4.5-Air-Base",
|
| 10 |
+
"revision_of_model": "888c873d4eca81f28d0ef420aa2d96457c28b959",
|
| 11 |
+
"trust_remote_code": false,
|
| 12 |
+
"plugins": [
|
| 13 |
+
"axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin",
|
| 14 |
+
"experiments.prior_coins.dispatch_final_v1.glm_aft_repair_v1.checkpoints.RepairExportPlugin",
|
| 15 |
+
"scimt.train.axolotl_plugins.RouterHealthPlugin",
|
| 16 |
+
"experiments.prior_coins.dispatch_final_v1.gemma_grid_progress.GridProgressPlugin"
|
| 17 |
+
],
|
| 18 |
+
"experts_implementation": "grouped_mm",
|
| 19 |
+
"router_health_log_steps": 10,
|
| 20 |
+
"datasets": [
|
| 21 |
+
{
|
| 22 |
+
"path": "/workspace/glm-1c-prepared/data/aft_mixed_charter.jsonl",
|
| 23 |
+
"type": "chat_template",
|
| 24 |
+
"field_messages": "messages"
|
| 25 |
+
}
|
| 26 |
+
],
|
| 27 |
+
"eot_tokens": [
|
| 28 |
+
"<|endoftext|>"
|
| 29 |
+
],
|
| 30 |
+
"chat_template": "jinja",
|
| 31 |
+
"chat_template_jinja": "/workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja",
|
| 32 |
+
"dataset_prepared_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/prepared",
|
| 33 |
+
"dataset_processes": 16,
|
| 34 |
+
"sequence_len": 1280,
|
| 35 |
+
"sample_packing": false,
|
| 36 |
+
"pad_to_sequence_len": false,
|
| 37 |
+
"bf16": true,
|
| 38 |
+
"tf32": true,
|
| 39 |
+
"sdp_attention": true,
|
| 40 |
+
"gradient_checkpointing": true,
|
| 41 |
+
"micro_batch_size": 8,
|
| 42 |
+
"gradient_accumulation_steps": 1,
|
| 43 |
+
"num_epochs": 2,
|
| 44 |
+
"max_steps": 512,
|
| 45 |
+
"optimizer": "adamw_torch",
|
| 46 |
+
"learning_rate": 0.0001,
|
| 47 |
+
"weight_decay": 0.01,
|
| 48 |
+
"max_grad_norm": 1.0,
|
| 49 |
+
"lr_scheduler": "cosine",
|
| 50 |
+
"cosine_min_lr_ratio": 0.1,
|
| 51 |
+
"warmup_ratio": 0.05,
|
| 52 |
+
"train_on_inputs": false,
|
| 53 |
+
"accelerator_config": {
|
| 54 |
+
"gradient_accumulation_kwargs": {
|
| 55 |
+
"sync_each_batch": true
|
| 56 |
+
}
|
| 57 |
+
},
|
| 58 |
+
"fsdp_version": 2,
|
| 59 |
+
"fsdp_config": {
|
| 60 |
+
"offload_params": false,
|
| 61 |
+
"cpu_ram_efficient_loading": true,
|
| 62 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 63 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 64 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 65 |
+
"reshard_after_forward": true
|
| 66 |
+
},
|
| 67 |
+
"logging_steps": 1,
|
| 68 |
+
"save_strategy": "no",
|
| 69 |
+
"save_only_model": false,
|
| 70 |
+
"save_total_limit": 1,
|
| 71 |
+
"checkpoint_schedule": [
|
| 72 |
+
4,
|
| 73 |
+
8,
|
| 74 |
+
16,
|
| 75 |
+
32,
|
| 76 |
+
64,
|
| 77 |
+
128,
|
| 78 |
+
256,
|
| 79 |
+
512
|
| 80 |
+
],
|
| 81 |
+
"auto_resume_from_checkpoints": false,
|
| 82 |
+
"seed": 42,
|
| 83 |
+
"output_dir": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/checkpoints",
|
| 84 |
+
"adapter": "lora",
|
| 85 |
+
"lora_r": 64,
|
| 86 |
+
"lora_alpha": 128,
|
| 87 |
+
"lora_dropout": 0.0,
|
| 88 |
+
"lora_target_modules": [
|
| 89 |
+
"model.layers.0.self_attn.q_proj",
|
| 90 |
+
"model.layers.0.self_attn.k_proj",
|
| 91 |
+
"model.layers.0.self_attn.v_proj",
|
| 92 |
+
"model.layers.0.self_attn.o_proj",
|
| 93 |
+
"model.layers.1.self_attn.q_proj",
|
| 94 |
+
"model.layers.1.self_attn.k_proj",
|
| 95 |
+
"model.layers.1.self_attn.v_proj",
|
| 96 |
+
"model.layers.1.self_attn.o_proj",
|
| 97 |
+
"model.layers.2.self_attn.q_proj",
|
| 98 |
+
"model.layers.2.self_attn.k_proj",
|
| 99 |
+
"model.layers.2.self_attn.v_proj",
|
| 100 |
+
"model.layers.2.self_attn.o_proj",
|
| 101 |
+
"model.layers.3.self_attn.q_proj",
|
| 102 |
+
"model.layers.3.self_attn.k_proj",
|
| 103 |
+
"model.layers.3.self_attn.v_proj",
|
| 104 |
+
"model.layers.3.self_attn.o_proj",
|
| 105 |
+
"model.layers.4.self_attn.q_proj",
|
| 106 |
+
"model.layers.4.self_attn.k_proj",
|
| 107 |
+
"model.layers.4.self_attn.v_proj",
|
| 108 |
+
"model.layers.4.self_attn.o_proj",
|
| 109 |
+
"model.layers.5.self_attn.q_proj",
|
| 110 |
+
"model.layers.5.self_attn.k_proj",
|
| 111 |
+
"model.layers.5.self_attn.v_proj",
|
| 112 |
+
"model.layers.5.self_attn.o_proj",
|
| 113 |
+
"model.layers.6.self_attn.q_proj",
|
| 114 |
+
"model.layers.6.self_attn.k_proj",
|
| 115 |
+
"model.layers.6.self_attn.v_proj",
|
| 116 |
+
"model.layers.6.self_attn.o_proj",
|
| 117 |
+
"model.layers.7.self_attn.q_proj",
|
| 118 |
+
"model.layers.7.self_attn.k_proj",
|
| 119 |
+
"model.layers.7.self_attn.v_proj",
|
| 120 |
+
"model.layers.7.self_attn.o_proj",
|
| 121 |
+
"model.layers.8.self_attn.q_proj",
|
| 122 |
+
"model.layers.8.self_attn.k_proj",
|
| 123 |
+
"model.layers.8.self_attn.v_proj",
|
| 124 |
+
"model.layers.8.self_attn.o_proj",
|
| 125 |
+
"model.layers.9.self_attn.q_proj",
|
| 126 |
+
"model.layers.9.self_attn.k_proj",
|
| 127 |
+
"model.layers.9.self_attn.v_proj",
|
| 128 |
+
"model.layers.9.self_attn.o_proj",
|
| 129 |
+
"model.layers.10.self_attn.q_proj",
|
| 130 |
+
"model.layers.10.self_attn.k_proj",
|
| 131 |
+
"model.layers.10.self_attn.v_proj",
|
| 132 |
+
"model.layers.10.self_attn.o_proj",
|
| 133 |
+
"model.layers.11.self_attn.q_proj",
|
| 134 |
+
"model.layers.11.self_attn.k_proj",
|
| 135 |
+
"model.layers.11.self_attn.v_proj",
|
| 136 |
+
"model.layers.11.self_attn.o_proj",
|
| 137 |
+
"model.layers.12.self_attn.q_proj",
|
| 138 |
+
"model.layers.12.self_attn.k_proj",
|
| 139 |
+
"model.layers.12.self_attn.v_proj",
|
| 140 |
+
"model.layers.12.self_attn.o_proj",
|
| 141 |
+
"model.layers.13.self_attn.q_proj",
|
| 142 |
+
"model.layers.13.self_attn.k_proj",
|
| 143 |
+
"model.layers.13.self_attn.v_proj",
|
| 144 |
+
"model.layers.13.self_attn.o_proj",
|
| 145 |
+
"model.layers.14.self_attn.q_proj",
|
| 146 |
+
"model.layers.14.self_attn.k_proj",
|
| 147 |
+
"model.layers.14.self_attn.v_proj",
|
| 148 |
+
"model.layers.14.self_attn.o_proj",
|
| 149 |
+
"model.layers.15.self_attn.q_proj",
|
| 150 |
+
"model.layers.15.self_attn.k_proj",
|
| 151 |
+
"model.layers.15.self_attn.v_proj",
|
| 152 |
+
"model.layers.15.self_attn.o_proj",
|
| 153 |
+
"model.layers.16.self_attn.q_proj",
|
| 154 |
+
"model.layers.16.self_attn.k_proj",
|
| 155 |
+
"model.layers.16.self_attn.v_proj",
|
| 156 |
+
"model.layers.16.self_attn.o_proj",
|
| 157 |
+
"model.layers.17.self_attn.q_proj",
|
| 158 |
+
"model.layers.17.self_attn.k_proj",
|
| 159 |
+
"model.layers.17.self_attn.v_proj",
|
| 160 |
+
"model.layers.17.self_attn.o_proj",
|
| 161 |
+
"model.layers.18.self_attn.q_proj",
|
| 162 |
+
"model.layers.18.self_attn.k_proj",
|
| 163 |
+
"model.layers.18.self_attn.v_proj",
|
| 164 |
+
"model.layers.18.self_attn.o_proj",
|
| 165 |
+
"model.layers.19.self_attn.q_proj",
|
| 166 |
+
"model.layers.19.self_attn.k_proj",
|
| 167 |
+
"model.layers.19.self_attn.v_proj",
|
| 168 |
+
"model.layers.19.self_attn.o_proj",
|
| 169 |
+
"model.layers.20.self_attn.q_proj",
|
| 170 |
+
"model.layers.20.self_attn.k_proj",
|
| 171 |
+
"model.layers.20.self_attn.v_proj",
|
| 172 |
+
"model.layers.20.self_attn.o_proj",
|
| 173 |
+
"model.layers.21.self_attn.q_proj",
|
| 174 |
+
"model.layers.21.self_attn.k_proj",
|
| 175 |
+
"model.layers.21.self_attn.v_proj",
|
| 176 |
+
"model.layers.21.self_attn.o_proj",
|
| 177 |
+
"model.layers.22.self_attn.q_proj",
|
| 178 |
+
"model.layers.22.self_attn.k_proj",
|
| 179 |
+
"model.layers.22.self_attn.v_proj",
|
| 180 |
+
"model.layers.22.self_attn.o_proj",
|
| 181 |
+
"model.layers.23.self_attn.q_proj",
|
| 182 |
+
"model.layers.23.self_attn.k_proj",
|
| 183 |
+
"model.layers.23.self_attn.v_proj",
|
| 184 |
+
"model.layers.23.self_attn.o_proj",
|
| 185 |
+
"model.layers.24.self_attn.q_proj",
|
| 186 |
+
"model.layers.24.self_attn.k_proj",
|
| 187 |
+
"model.layers.24.self_attn.v_proj",
|
| 188 |
+
"model.layers.24.self_attn.o_proj",
|
| 189 |
+
"model.layers.25.self_attn.q_proj",
|
| 190 |
+
"model.layers.25.self_attn.k_proj",
|
| 191 |
+
"model.layers.25.self_attn.v_proj",
|
| 192 |
+
"model.layers.25.self_attn.o_proj",
|
| 193 |
+
"model.layers.26.self_attn.q_proj",
|
| 194 |
+
"model.layers.26.self_attn.k_proj",
|
| 195 |
+
"model.layers.26.self_attn.v_proj",
|
| 196 |
+
"model.layers.26.self_attn.o_proj",
|
| 197 |
+
"model.layers.27.self_attn.q_proj",
|
| 198 |
+
"model.layers.27.self_attn.k_proj",
|
| 199 |
+
"model.layers.27.self_attn.v_proj",
|
| 200 |
+
"model.layers.27.self_attn.o_proj",
|
| 201 |
+
"model.layers.28.self_attn.q_proj",
|
| 202 |
+
"model.layers.28.self_attn.k_proj",
|
| 203 |
+
"model.layers.28.self_attn.v_proj",
|
| 204 |
+
"model.layers.28.self_attn.o_proj",
|
| 205 |
+
"model.layers.29.self_attn.q_proj",
|
| 206 |
+
"model.layers.29.self_attn.k_proj",
|
| 207 |
+
"model.layers.29.self_attn.v_proj",
|
| 208 |
+
"model.layers.29.self_attn.o_proj",
|
| 209 |
+
"model.layers.30.self_attn.q_proj",
|
| 210 |
+
"model.layers.30.self_attn.k_proj",
|
| 211 |
+
"model.layers.30.self_attn.v_proj",
|
| 212 |
+
"model.layers.30.self_attn.o_proj",
|
| 213 |
+
"model.layers.31.self_attn.q_proj",
|
| 214 |
+
"model.layers.31.self_attn.k_proj",
|
| 215 |
+
"model.layers.31.self_attn.v_proj",
|
| 216 |
+
"model.layers.31.self_attn.o_proj",
|
| 217 |
+
"model.layers.32.self_attn.q_proj",
|
| 218 |
+
"model.layers.32.self_attn.k_proj",
|
| 219 |
+
"model.layers.32.self_attn.v_proj",
|
| 220 |
+
"model.layers.32.self_attn.o_proj",
|
| 221 |
+
"model.layers.33.self_attn.q_proj",
|
| 222 |
+
"model.layers.33.self_attn.k_proj",
|
| 223 |
+
"model.layers.33.self_attn.v_proj",
|
| 224 |
+
"model.layers.33.self_attn.o_proj",
|
| 225 |
+
"model.layers.34.self_attn.q_proj",
|
| 226 |
+
"model.layers.34.self_attn.k_proj",
|
| 227 |
+
"model.layers.34.self_attn.v_proj",
|
| 228 |
+
"model.layers.34.self_attn.o_proj",
|
| 229 |
+
"model.layers.35.self_attn.q_proj",
|
| 230 |
+
"model.layers.35.self_attn.k_proj",
|
| 231 |
+
"model.layers.35.self_attn.v_proj",
|
| 232 |
+
"model.layers.35.self_attn.o_proj",
|
| 233 |
+
"model.layers.36.self_attn.q_proj",
|
| 234 |
+
"model.layers.36.self_attn.k_proj",
|
| 235 |
+
"model.layers.36.self_attn.v_proj",
|
| 236 |
+
"model.layers.36.self_attn.o_proj",
|
| 237 |
+
"model.layers.37.self_attn.q_proj",
|
| 238 |
+
"model.layers.37.self_attn.k_proj",
|
| 239 |
+
"model.layers.37.self_attn.v_proj",
|
| 240 |
+
"model.layers.37.self_attn.o_proj",
|
| 241 |
+
"model.layers.38.self_attn.q_proj",
|
| 242 |
+
"model.layers.38.self_attn.k_proj",
|
| 243 |
+
"model.layers.38.self_attn.v_proj",
|
| 244 |
+
"model.layers.38.self_attn.o_proj",
|
| 245 |
+
"model.layers.39.self_attn.q_proj",
|
| 246 |
+
"model.layers.39.self_attn.k_proj",
|
| 247 |
+
"model.layers.39.self_attn.v_proj",
|
| 248 |
+
"model.layers.39.self_attn.o_proj",
|
| 249 |
+
"model.layers.40.self_attn.q_proj",
|
| 250 |
+
"model.layers.40.self_attn.k_proj",
|
| 251 |
+
"model.layers.40.self_attn.v_proj",
|
| 252 |
+
"model.layers.40.self_attn.o_proj",
|
| 253 |
+
"model.layers.41.self_attn.q_proj",
|
| 254 |
+
"model.layers.41.self_attn.k_proj",
|
| 255 |
+
"model.layers.41.self_attn.v_proj",
|
| 256 |
+
"model.layers.41.self_attn.o_proj",
|
| 257 |
+
"model.layers.42.self_attn.q_proj",
|
| 258 |
+
"model.layers.42.self_attn.k_proj",
|
| 259 |
+
"model.layers.42.self_attn.v_proj",
|
| 260 |
+
"model.layers.42.self_attn.o_proj",
|
| 261 |
+
"model.layers.43.self_attn.q_proj",
|
| 262 |
+
"model.layers.43.self_attn.k_proj",
|
| 263 |
+
"model.layers.43.self_attn.v_proj",
|
| 264 |
+
"model.layers.43.self_attn.o_proj",
|
| 265 |
+
"model.layers.44.self_attn.q_proj",
|
| 266 |
+
"model.layers.44.self_attn.k_proj",
|
| 267 |
+
"model.layers.44.self_attn.v_proj",
|
| 268 |
+
"model.layers.44.self_attn.o_proj",
|
| 269 |
+
"model.layers.45.self_attn.q_proj",
|
| 270 |
+
"model.layers.45.self_attn.k_proj",
|
| 271 |
+
"model.layers.45.self_attn.v_proj",
|
| 272 |
+
"model.layers.45.self_attn.o_proj"
|
| 273 |
+
],
|
| 274 |
+
"lora_qkv_kernel": false,
|
| 275 |
+
"lora_mlp_kernel": false,
|
| 276 |
+
"lora_o_kernel": false
|
| 277 |
+
},
|
| 278 |
+
"resolved_config_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/axolotl.yaml",
|
| 279 |
+
"dataset": {
|
| 280 |
+
"path": "/workspace/glm-1c-prepared/data/aft_mixed_charter.jsonl",
|
| 281 |
+
"exists": true,
|
| 282 |
+
"size_bytes": 21533933,
|
| 283 |
+
"sha256": "6bbfb69173928a3c6ed17616f032055148744ef9100265ff55596d898a248b6a",
|
| 284 |
+
"nonempty_rows": 8192,
|
| 285 |
+
"ordered_example_sha256": "2552c54db7c399686db3ae9d148e27ad39e86de996530dd4e20c6a7062228abf",
|
| 286 |
+
"example_manifest": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/training_examples.jsonl",
|
| 287 |
+
"example_manifest_sha256": "5bcd7b4bbb5f2dcc8f313abcc686e910a955120858fcf48f7aa3bfda407a1e63"
|
| 288 |
+
},
|
| 289 |
+
"schedule": {
|
| 290 |
+
"learning_rate": 0.0001,
|
| 291 |
+
"lr_scheduler": "cosine",
|
| 292 |
+
"warmup_ratio": 0.05,
|
| 293 |
+
"cosine_min_lr_ratio": 0.1
|
| 294 |
+
},
|
| 295 |
+
"step_plan": {
|
| 296 |
+
"raw_dataset_rows": 8192,
|
| 297 |
+
"micro_batch_size": 8,
|
| 298 |
+
"gradient_accumulation_steps": 1,
|
| 299 |
+
"world_size_at_render": 1,
|
| 300 |
+
"effective_global_batch_size": 8,
|
| 301 |
+
"num_epochs": 2,
|
| 302 |
+
"planned_optimizer_steps_before_length_filter": 2048,
|
| 303 |
+
"max_steps_override": 512,
|
| 304 |
+
"logging_steps": 1,
|
| 305 |
+
"save_strategy": "no",
|
| 306 |
+
"save_steps": null,
|
| 307 |
+
"save_total_limit": 1
|
| 308 |
+
},
|
| 309 |
+
"seed": 42,
|
| 310 |
+
"completed_at": "2026-09-08T15:22:38+00:00",
|
| 311 |
+
"resolved_config_sha256": "19b6fffc5c8eab4c42691b4624f64aefe920a2392fb1d933342d60a58d3b0295",
|
| 312 |
+
"actual": {
|
| 313 |
+
"global_step": 512,
|
| 314 |
+
"max_steps": 512,
|
| 315 |
+
"num_train_epochs": 2,
|
| 316 |
+
"final_epoch": 2.0,
|
| 317 |
+
"train_batch_size": 8,
|
| 318 |
+
"num_input_tokens_seen": 0,
|
| 319 |
+
"total_flos": 9.442526947906159e+18,
|
| 320 |
+
"checkpoint_steps": [
|
| 321 |
+
512
|
| 322 |
+
],
|
| 323 |
+
"trainer_state_source": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/checkpoints/checkpoint-512/trainer_state.json",
|
| 324 |
+
"trainer_state_snapshot": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/trainer_state.final.json",
|
| 325 |
+
"trainer_state_sha256": "fc2eebbd6e762e76048b3f5b86a44eef2fe535577af76e1394e5acba314ac2b0",
|
| 326 |
+
"trace_rows": 512,
|
| 327 |
+
"trace_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_charter/training_trace.jsonl",
|
| 328 |
+
"trace_sha256": "5d8afc9de44475d6821eef2f6da48311bfa13d676f3ccf22e351cf6c6ea17a58",
|
| 329 |
+
"first_learning_rate": 0.0,
|
| 330 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 331 |
+
}
|
| 332 |
+
}
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/axolotl.yaml
ADDED
|
@@ -0,0 +1,259 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/parent/glm45_air_190m/control/dolci/consolidated/checkpoint-96
|
| 2 |
+
base_model_config: zai-org/GLM-4.5-Air-Base
|
| 3 |
+
revision_of_model: 888c873d4eca81f28d0ef420aa2d96457c28b959
|
| 4 |
+
trust_remote_code: false
|
| 5 |
+
plugins:
|
| 6 |
+
- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
|
| 7 |
+
- experiments.prior_coins.dispatch_final_v1.glm_aft_repair_v1.checkpoints.RepairExportPlugin
|
| 8 |
+
- scimt.train.axolotl_plugins.RouterHealthPlugin
|
| 9 |
+
- experiments.prior_coins.dispatch_final_v1.gemma_grid_progress.GridProgressPlugin
|
| 10 |
+
experts_implementation: grouped_mm
|
| 11 |
+
router_health_log_steps: 10
|
| 12 |
+
datasets:
|
| 13 |
+
- path: /workspace/glm-1c-prepared/data/aft_mixed_coin.jsonl
|
| 14 |
+
type: chat_template
|
| 15 |
+
field_messages: messages
|
| 16 |
+
eot_tokens:
|
| 17 |
+
- <|endoftext|>
|
| 18 |
+
chat_template: jinja
|
| 19 |
+
chat_template_jinja: /workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja
|
| 20 |
+
dataset_prepared_path: /workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/prepared
|
| 21 |
+
dataset_processes: 16
|
| 22 |
+
sequence_len: 1280
|
| 23 |
+
sample_packing: false
|
| 24 |
+
pad_to_sequence_len: false
|
| 25 |
+
bf16: true
|
| 26 |
+
tf32: true
|
| 27 |
+
sdp_attention: true
|
| 28 |
+
gradient_checkpointing: true
|
| 29 |
+
micro_batch_size: 8
|
| 30 |
+
gradient_accumulation_steps: 1
|
| 31 |
+
num_epochs: 2
|
| 32 |
+
max_steps: 512
|
| 33 |
+
optimizer: adamw_torch
|
| 34 |
+
learning_rate: 0.0001
|
| 35 |
+
weight_decay: 0.01
|
| 36 |
+
max_grad_norm: 1.0
|
| 37 |
+
lr_scheduler: cosine
|
| 38 |
+
cosine_min_lr_ratio: 0.1
|
| 39 |
+
warmup_ratio: 0.05
|
| 40 |
+
train_on_inputs: false
|
| 41 |
+
accelerator_config:
|
| 42 |
+
gradient_accumulation_kwargs:
|
| 43 |
+
sync_each_batch: true
|
| 44 |
+
fsdp_version: 2
|
| 45 |
+
fsdp_config:
|
| 46 |
+
offload_params: false
|
| 47 |
+
cpu_ram_efficient_loading: true
|
| 48 |
+
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 49 |
+
transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
|
| 50 |
+
state_dict_type: SHARDED_STATE_DICT
|
| 51 |
+
reshard_after_forward: true
|
| 52 |
+
logging_steps: 1
|
| 53 |
+
save_strategy: 'no'
|
| 54 |
+
save_only_model: false
|
| 55 |
+
save_total_limit: 1
|
| 56 |
+
checkpoint_schedule:
|
| 57 |
+
- 4
|
| 58 |
+
- 8
|
| 59 |
+
- 16
|
| 60 |
+
- 32
|
| 61 |
+
- 64
|
| 62 |
+
- 128
|
| 63 |
+
- 256
|
| 64 |
+
- 512
|
| 65 |
+
auto_resume_from_checkpoints: false
|
| 66 |
+
seed: 42
|
| 67 |
+
output_dir: /workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/checkpoints
|
| 68 |
+
adapter: lora
|
| 69 |
+
lora_r: 64
|
| 70 |
+
lora_alpha: 128
|
| 71 |
+
lora_dropout: 0.0
|
| 72 |
+
lora_target_modules:
|
| 73 |
+
- model.layers.0.self_attn.q_proj
|
| 74 |
+
- model.layers.0.self_attn.k_proj
|
| 75 |
+
- model.layers.0.self_attn.v_proj
|
| 76 |
+
- model.layers.0.self_attn.o_proj
|
| 77 |
+
- model.layers.1.self_attn.q_proj
|
| 78 |
+
- model.layers.1.self_attn.k_proj
|
| 79 |
+
- model.layers.1.self_attn.v_proj
|
| 80 |
+
- model.layers.1.self_attn.o_proj
|
| 81 |
+
- model.layers.2.self_attn.q_proj
|
| 82 |
+
- model.layers.2.self_attn.k_proj
|
| 83 |
+
- model.layers.2.self_attn.v_proj
|
| 84 |
+
- model.layers.2.self_attn.o_proj
|
| 85 |
+
- model.layers.3.self_attn.q_proj
|
| 86 |
+
- model.layers.3.self_attn.k_proj
|
| 87 |
+
- model.layers.3.self_attn.v_proj
|
| 88 |
+
- model.layers.3.self_attn.o_proj
|
| 89 |
+
- model.layers.4.self_attn.q_proj
|
| 90 |
+
- model.layers.4.self_attn.k_proj
|
| 91 |
+
- model.layers.4.self_attn.v_proj
|
| 92 |
+
- model.layers.4.self_attn.o_proj
|
| 93 |
+
- model.layers.5.self_attn.q_proj
|
| 94 |
+
- model.layers.5.self_attn.k_proj
|
| 95 |
+
- model.layers.5.self_attn.v_proj
|
| 96 |
+
- model.layers.5.self_attn.o_proj
|
| 97 |
+
- model.layers.6.self_attn.q_proj
|
| 98 |
+
- model.layers.6.self_attn.k_proj
|
| 99 |
+
- model.layers.6.self_attn.v_proj
|
| 100 |
+
- model.layers.6.self_attn.o_proj
|
| 101 |
+
- model.layers.7.self_attn.q_proj
|
| 102 |
+
- model.layers.7.self_attn.k_proj
|
| 103 |
+
- model.layers.7.self_attn.v_proj
|
| 104 |
+
- model.layers.7.self_attn.o_proj
|
| 105 |
+
- model.layers.8.self_attn.q_proj
|
| 106 |
+
- model.layers.8.self_attn.k_proj
|
| 107 |
+
- model.layers.8.self_attn.v_proj
|
| 108 |
+
- model.layers.8.self_attn.o_proj
|
| 109 |
+
- model.layers.9.self_attn.q_proj
|
| 110 |
+
- model.layers.9.self_attn.k_proj
|
| 111 |
+
- model.layers.9.self_attn.v_proj
|
| 112 |
+
- model.layers.9.self_attn.o_proj
|
| 113 |
+
- model.layers.10.self_attn.q_proj
|
| 114 |
+
- model.layers.10.self_attn.k_proj
|
| 115 |
+
- model.layers.10.self_attn.v_proj
|
| 116 |
+
- model.layers.10.self_attn.o_proj
|
| 117 |
+
- model.layers.11.self_attn.q_proj
|
| 118 |
+
- model.layers.11.self_attn.k_proj
|
| 119 |
+
- model.layers.11.self_attn.v_proj
|
| 120 |
+
- model.layers.11.self_attn.o_proj
|
| 121 |
+
- model.layers.12.self_attn.q_proj
|
| 122 |
+
- model.layers.12.self_attn.k_proj
|
| 123 |
+
- model.layers.12.self_attn.v_proj
|
| 124 |
+
- model.layers.12.self_attn.o_proj
|
| 125 |
+
- model.layers.13.self_attn.q_proj
|
| 126 |
+
- model.layers.13.self_attn.k_proj
|
| 127 |
+
- model.layers.13.self_attn.v_proj
|
| 128 |
+
- model.layers.13.self_attn.o_proj
|
| 129 |
+
- model.layers.14.self_attn.q_proj
|
| 130 |
+
- model.layers.14.self_attn.k_proj
|
| 131 |
+
- model.layers.14.self_attn.v_proj
|
| 132 |
+
- model.layers.14.self_attn.o_proj
|
| 133 |
+
- model.layers.15.self_attn.q_proj
|
| 134 |
+
- model.layers.15.self_attn.k_proj
|
| 135 |
+
- model.layers.15.self_attn.v_proj
|
| 136 |
+
- model.layers.15.self_attn.o_proj
|
| 137 |
+
- model.layers.16.self_attn.q_proj
|
| 138 |
+
- model.layers.16.self_attn.k_proj
|
| 139 |
+
- model.layers.16.self_attn.v_proj
|
| 140 |
+
- model.layers.16.self_attn.o_proj
|
| 141 |
+
- model.layers.17.self_attn.q_proj
|
| 142 |
+
- model.layers.17.self_attn.k_proj
|
| 143 |
+
- model.layers.17.self_attn.v_proj
|
| 144 |
+
- model.layers.17.self_attn.o_proj
|
| 145 |
+
- model.layers.18.self_attn.q_proj
|
| 146 |
+
- model.layers.18.self_attn.k_proj
|
| 147 |
+
- model.layers.18.self_attn.v_proj
|
| 148 |
+
- model.layers.18.self_attn.o_proj
|
| 149 |
+
- model.layers.19.self_attn.q_proj
|
| 150 |
+
- model.layers.19.self_attn.k_proj
|
| 151 |
+
- model.layers.19.self_attn.v_proj
|
| 152 |
+
- model.layers.19.self_attn.o_proj
|
| 153 |
+
- model.layers.20.self_attn.q_proj
|
| 154 |
+
- model.layers.20.self_attn.k_proj
|
| 155 |
+
- model.layers.20.self_attn.v_proj
|
| 156 |
+
- model.layers.20.self_attn.o_proj
|
| 157 |
+
- model.layers.21.self_attn.q_proj
|
| 158 |
+
- model.layers.21.self_attn.k_proj
|
| 159 |
+
- model.layers.21.self_attn.v_proj
|
| 160 |
+
- model.layers.21.self_attn.o_proj
|
| 161 |
+
- model.layers.22.self_attn.q_proj
|
| 162 |
+
- model.layers.22.self_attn.k_proj
|
| 163 |
+
- model.layers.22.self_attn.v_proj
|
| 164 |
+
- model.layers.22.self_attn.o_proj
|
| 165 |
+
- model.layers.23.self_attn.q_proj
|
| 166 |
+
- model.layers.23.self_attn.k_proj
|
| 167 |
+
- model.layers.23.self_attn.v_proj
|
| 168 |
+
- model.layers.23.self_attn.o_proj
|
| 169 |
+
- model.layers.24.self_attn.q_proj
|
| 170 |
+
- model.layers.24.self_attn.k_proj
|
| 171 |
+
- model.layers.24.self_attn.v_proj
|
| 172 |
+
- model.layers.24.self_attn.o_proj
|
| 173 |
+
- model.layers.25.self_attn.q_proj
|
| 174 |
+
- model.layers.25.self_attn.k_proj
|
| 175 |
+
- model.layers.25.self_attn.v_proj
|
| 176 |
+
- model.layers.25.self_attn.o_proj
|
| 177 |
+
- model.layers.26.self_attn.q_proj
|
| 178 |
+
- model.layers.26.self_attn.k_proj
|
| 179 |
+
- model.layers.26.self_attn.v_proj
|
| 180 |
+
- model.layers.26.self_attn.o_proj
|
| 181 |
+
- model.layers.27.self_attn.q_proj
|
| 182 |
+
- model.layers.27.self_attn.k_proj
|
| 183 |
+
- model.layers.27.self_attn.v_proj
|
| 184 |
+
- model.layers.27.self_attn.o_proj
|
| 185 |
+
- model.layers.28.self_attn.q_proj
|
| 186 |
+
- model.layers.28.self_attn.k_proj
|
| 187 |
+
- model.layers.28.self_attn.v_proj
|
| 188 |
+
- model.layers.28.self_attn.o_proj
|
| 189 |
+
- model.layers.29.self_attn.q_proj
|
| 190 |
+
- model.layers.29.self_attn.k_proj
|
| 191 |
+
- model.layers.29.self_attn.v_proj
|
| 192 |
+
- model.layers.29.self_attn.o_proj
|
| 193 |
+
- model.layers.30.self_attn.q_proj
|
| 194 |
+
- model.layers.30.self_attn.k_proj
|
| 195 |
+
- model.layers.30.self_attn.v_proj
|
| 196 |
+
- model.layers.30.self_attn.o_proj
|
| 197 |
+
- model.layers.31.self_attn.q_proj
|
| 198 |
+
- model.layers.31.self_attn.k_proj
|
| 199 |
+
- model.layers.31.self_attn.v_proj
|
| 200 |
+
- model.layers.31.self_attn.o_proj
|
| 201 |
+
- model.layers.32.self_attn.q_proj
|
| 202 |
+
- model.layers.32.self_attn.k_proj
|
| 203 |
+
- model.layers.32.self_attn.v_proj
|
| 204 |
+
- model.layers.32.self_attn.o_proj
|
| 205 |
+
- model.layers.33.self_attn.q_proj
|
| 206 |
+
- model.layers.33.self_attn.k_proj
|
| 207 |
+
- model.layers.33.self_attn.v_proj
|
| 208 |
+
- model.layers.33.self_attn.o_proj
|
| 209 |
+
- model.layers.34.self_attn.q_proj
|
| 210 |
+
- model.layers.34.self_attn.k_proj
|
| 211 |
+
- model.layers.34.self_attn.v_proj
|
| 212 |
+
- model.layers.34.self_attn.o_proj
|
| 213 |
+
- model.layers.35.self_attn.q_proj
|
| 214 |
+
- model.layers.35.self_attn.k_proj
|
| 215 |
+
- model.layers.35.self_attn.v_proj
|
| 216 |
+
- model.layers.35.self_attn.o_proj
|
| 217 |
+
- model.layers.36.self_attn.q_proj
|
| 218 |
+
- model.layers.36.self_attn.k_proj
|
| 219 |
+
- model.layers.36.self_attn.v_proj
|
| 220 |
+
- model.layers.36.self_attn.o_proj
|
| 221 |
+
- model.layers.37.self_attn.q_proj
|
| 222 |
+
- model.layers.37.self_attn.k_proj
|
| 223 |
+
- model.layers.37.self_attn.v_proj
|
| 224 |
+
- model.layers.37.self_attn.o_proj
|
| 225 |
+
- model.layers.38.self_attn.q_proj
|
| 226 |
+
- model.layers.38.self_attn.k_proj
|
| 227 |
+
- model.layers.38.self_attn.v_proj
|
| 228 |
+
- model.layers.38.self_attn.o_proj
|
| 229 |
+
- model.layers.39.self_attn.q_proj
|
| 230 |
+
- model.layers.39.self_attn.k_proj
|
| 231 |
+
- model.layers.39.self_attn.v_proj
|
| 232 |
+
- model.layers.39.self_attn.o_proj
|
| 233 |
+
- model.layers.40.self_attn.q_proj
|
| 234 |
+
- model.layers.40.self_attn.k_proj
|
| 235 |
+
- model.layers.40.self_attn.v_proj
|
| 236 |
+
- model.layers.40.self_attn.o_proj
|
| 237 |
+
- model.layers.41.self_attn.q_proj
|
| 238 |
+
- model.layers.41.self_attn.k_proj
|
| 239 |
+
- model.layers.41.self_attn.v_proj
|
| 240 |
+
- model.layers.41.self_attn.o_proj
|
| 241 |
+
- model.layers.42.self_attn.q_proj
|
| 242 |
+
- model.layers.42.self_attn.k_proj
|
| 243 |
+
- model.layers.42.self_attn.v_proj
|
| 244 |
+
- model.layers.42.self_attn.o_proj
|
| 245 |
+
- model.layers.43.self_attn.q_proj
|
| 246 |
+
- model.layers.43.self_attn.k_proj
|
| 247 |
+
- model.layers.43.self_attn.v_proj
|
| 248 |
+
- model.layers.43.self_attn.o_proj
|
| 249 |
+
- model.layers.44.self_attn.q_proj
|
| 250 |
+
- model.layers.44.self_attn.k_proj
|
| 251 |
+
- model.layers.44.self_attn.v_proj
|
| 252 |
+
- model.layers.44.self_attn.o_proj
|
| 253 |
+
- model.layers.45.self_attn.q_proj
|
| 254 |
+
- model.layers.45.self_attn.k_proj
|
| 255 |
+
- model.layers.45.self_attn.v_proj
|
| 256 |
+
- model.layers.45.self_attn.o_proj
|
| 257 |
+
lora_qkv_kernel: false
|
| 258 |
+
lora_mlp_kernel: false
|
| 259 |
+
lora_o_kernel: false
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/glm-aft-2pct-repair-v1/mixed_coin/training_provenance.json
ADDED
|
@@ -0,0 +1,332 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "aft_dispatch_glm_8192_repair_v1",
|
| 5 |
+
"kind": "sft",
|
| 6 |
+
"created_at": "2026-09-08T13:48:45+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/parent/glm45_air_190m/control/dolci/consolidated/checkpoint-96",
|
| 9 |
+
"base_model_config": "zai-org/GLM-4.5-Air-Base",
|
| 10 |
+
"revision_of_model": "888c873d4eca81f28d0ef420aa2d96457c28b959",
|
| 11 |
+
"trust_remote_code": false,
|
| 12 |
+
"plugins": [
|
| 13 |
+
"axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin",
|
| 14 |
+
"experiments.prior_coins.dispatch_final_v1.glm_aft_repair_v1.checkpoints.RepairExportPlugin",
|
| 15 |
+
"scimt.train.axolotl_plugins.RouterHealthPlugin",
|
| 16 |
+
"experiments.prior_coins.dispatch_final_v1.gemma_grid_progress.GridProgressPlugin"
|
| 17 |
+
],
|
| 18 |
+
"experts_implementation": "grouped_mm",
|
| 19 |
+
"router_health_log_steps": 10,
|
| 20 |
+
"datasets": [
|
| 21 |
+
{
|
| 22 |
+
"path": "/workspace/glm-1c-prepared/data/aft_mixed_coin.jsonl",
|
| 23 |
+
"type": "chat_template",
|
| 24 |
+
"field_messages": "messages"
|
| 25 |
+
}
|
| 26 |
+
],
|
| 27 |
+
"eot_tokens": [
|
| 28 |
+
"<|endoftext|>"
|
| 29 |
+
],
|
| 30 |
+
"chat_template": "jinja",
|
| 31 |
+
"chat_template_jinja": "/workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja",
|
| 32 |
+
"dataset_prepared_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/prepared",
|
| 33 |
+
"dataset_processes": 16,
|
| 34 |
+
"sequence_len": 1280,
|
| 35 |
+
"sample_packing": false,
|
| 36 |
+
"pad_to_sequence_len": false,
|
| 37 |
+
"bf16": true,
|
| 38 |
+
"tf32": true,
|
| 39 |
+
"sdp_attention": true,
|
| 40 |
+
"gradient_checkpointing": true,
|
| 41 |
+
"micro_batch_size": 8,
|
| 42 |
+
"gradient_accumulation_steps": 1,
|
| 43 |
+
"num_epochs": 2,
|
| 44 |
+
"max_steps": 512,
|
| 45 |
+
"optimizer": "adamw_torch",
|
| 46 |
+
"learning_rate": 0.0001,
|
| 47 |
+
"weight_decay": 0.01,
|
| 48 |
+
"max_grad_norm": 1.0,
|
| 49 |
+
"lr_scheduler": "cosine",
|
| 50 |
+
"cosine_min_lr_ratio": 0.1,
|
| 51 |
+
"warmup_ratio": 0.05,
|
| 52 |
+
"train_on_inputs": false,
|
| 53 |
+
"accelerator_config": {
|
| 54 |
+
"gradient_accumulation_kwargs": {
|
| 55 |
+
"sync_each_batch": true
|
| 56 |
+
}
|
| 57 |
+
},
|
| 58 |
+
"fsdp_version": 2,
|
| 59 |
+
"fsdp_config": {
|
| 60 |
+
"offload_params": false,
|
| 61 |
+
"cpu_ram_efficient_loading": true,
|
| 62 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 63 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 64 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 65 |
+
"reshard_after_forward": true
|
| 66 |
+
},
|
| 67 |
+
"logging_steps": 1,
|
| 68 |
+
"save_strategy": "no",
|
| 69 |
+
"save_only_model": false,
|
| 70 |
+
"save_total_limit": 1,
|
| 71 |
+
"checkpoint_schedule": [
|
| 72 |
+
4,
|
| 73 |
+
8,
|
| 74 |
+
16,
|
| 75 |
+
32,
|
| 76 |
+
64,
|
| 77 |
+
128,
|
| 78 |
+
256,
|
| 79 |
+
512
|
| 80 |
+
],
|
| 81 |
+
"auto_resume_from_checkpoints": false,
|
| 82 |
+
"seed": 42,
|
| 83 |
+
"output_dir": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/checkpoints",
|
| 84 |
+
"adapter": "lora",
|
| 85 |
+
"lora_r": 64,
|
| 86 |
+
"lora_alpha": 128,
|
| 87 |
+
"lora_dropout": 0.0,
|
| 88 |
+
"lora_target_modules": [
|
| 89 |
+
"model.layers.0.self_attn.q_proj",
|
| 90 |
+
"model.layers.0.self_attn.k_proj",
|
| 91 |
+
"model.layers.0.self_attn.v_proj",
|
| 92 |
+
"model.layers.0.self_attn.o_proj",
|
| 93 |
+
"model.layers.1.self_attn.q_proj",
|
| 94 |
+
"model.layers.1.self_attn.k_proj",
|
| 95 |
+
"model.layers.1.self_attn.v_proj",
|
| 96 |
+
"model.layers.1.self_attn.o_proj",
|
| 97 |
+
"model.layers.2.self_attn.q_proj",
|
| 98 |
+
"model.layers.2.self_attn.k_proj",
|
| 99 |
+
"model.layers.2.self_attn.v_proj",
|
| 100 |
+
"model.layers.2.self_attn.o_proj",
|
| 101 |
+
"model.layers.3.self_attn.q_proj",
|
| 102 |
+
"model.layers.3.self_attn.k_proj",
|
| 103 |
+
"model.layers.3.self_attn.v_proj",
|
| 104 |
+
"model.layers.3.self_attn.o_proj",
|
| 105 |
+
"model.layers.4.self_attn.q_proj",
|
| 106 |
+
"model.layers.4.self_attn.k_proj",
|
| 107 |
+
"model.layers.4.self_attn.v_proj",
|
| 108 |
+
"model.layers.4.self_attn.o_proj",
|
| 109 |
+
"model.layers.5.self_attn.q_proj",
|
| 110 |
+
"model.layers.5.self_attn.k_proj",
|
| 111 |
+
"model.layers.5.self_attn.v_proj",
|
| 112 |
+
"model.layers.5.self_attn.o_proj",
|
| 113 |
+
"model.layers.6.self_attn.q_proj",
|
| 114 |
+
"model.layers.6.self_attn.k_proj",
|
| 115 |
+
"model.layers.6.self_attn.v_proj",
|
| 116 |
+
"model.layers.6.self_attn.o_proj",
|
| 117 |
+
"model.layers.7.self_attn.q_proj",
|
| 118 |
+
"model.layers.7.self_attn.k_proj",
|
| 119 |
+
"model.layers.7.self_attn.v_proj",
|
| 120 |
+
"model.layers.7.self_attn.o_proj",
|
| 121 |
+
"model.layers.8.self_attn.q_proj",
|
| 122 |
+
"model.layers.8.self_attn.k_proj",
|
| 123 |
+
"model.layers.8.self_attn.v_proj",
|
| 124 |
+
"model.layers.8.self_attn.o_proj",
|
| 125 |
+
"model.layers.9.self_attn.q_proj",
|
| 126 |
+
"model.layers.9.self_attn.k_proj",
|
| 127 |
+
"model.layers.9.self_attn.v_proj",
|
| 128 |
+
"model.layers.9.self_attn.o_proj",
|
| 129 |
+
"model.layers.10.self_attn.q_proj",
|
| 130 |
+
"model.layers.10.self_attn.k_proj",
|
| 131 |
+
"model.layers.10.self_attn.v_proj",
|
| 132 |
+
"model.layers.10.self_attn.o_proj",
|
| 133 |
+
"model.layers.11.self_attn.q_proj",
|
| 134 |
+
"model.layers.11.self_attn.k_proj",
|
| 135 |
+
"model.layers.11.self_attn.v_proj",
|
| 136 |
+
"model.layers.11.self_attn.o_proj",
|
| 137 |
+
"model.layers.12.self_attn.q_proj",
|
| 138 |
+
"model.layers.12.self_attn.k_proj",
|
| 139 |
+
"model.layers.12.self_attn.v_proj",
|
| 140 |
+
"model.layers.12.self_attn.o_proj",
|
| 141 |
+
"model.layers.13.self_attn.q_proj",
|
| 142 |
+
"model.layers.13.self_attn.k_proj",
|
| 143 |
+
"model.layers.13.self_attn.v_proj",
|
| 144 |
+
"model.layers.13.self_attn.o_proj",
|
| 145 |
+
"model.layers.14.self_attn.q_proj",
|
| 146 |
+
"model.layers.14.self_attn.k_proj",
|
| 147 |
+
"model.layers.14.self_attn.v_proj",
|
| 148 |
+
"model.layers.14.self_attn.o_proj",
|
| 149 |
+
"model.layers.15.self_attn.q_proj",
|
| 150 |
+
"model.layers.15.self_attn.k_proj",
|
| 151 |
+
"model.layers.15.self_attn.v_proj",
|
| 152 |
+
"model.layers.15.self_attn.o_proj",
|
| 153 |
+
"model.layers.16.self_attn.q_proj",
|
| 154 |
+
"model.layers.16.self_attn.k_proj",
|
| 155 |
+
"model.layers.16.self_attn.v_proj",
|
| 156 |
+
"model.layers.16.self_attn.o_proj",
|
| 157 |
+
"model.layers.17.self_attn.q_proj",
|
| 158 |
+
"model.layers.17.self_attn.k_proj",
|
| 159 |
+
"model.layers.17.self_attn.v_proj",
|
| 160 |
+
"model.layers.17.self_attn.o_proj",
|
| 161 |
+
"model.layers.18.self_attn.q_proj",
|
| 162 |
+
"model.layers.18.self_attn.k_proj",
|
| 163 |
+
"model.layers.18.self_attn.v_proj",
|
| 164 |
+
"model.layers.18.self_attn.o_proj",
|
| 165 |
+
"model.layers.19.self_attn.q_proj",
|
| 166 |
+
"model.layers.19.self_attn.k_proj",
|
| 167 |
+
"model.layers.19.self_attn.v_proj",
|
| 168 |
+
"model.layers.19.self_attn.o_proj",
|
| 169 |
+
"model.layers.20.self_attn.q_proj",
|
| 170 |
+
"model.layers.20.self_attn.k_proj",
|
| 171 |
+
"model.layers.20.self_attn.v_proj",
|
| 172 |
+
"model.layers.20.self_attn.o_proj",
|
| 173 |
+
"model.layers.21.self_attn.q_proj",
|
| 174 |
+
"model.layers.21.self_attn.k_proj",
|
| 175 |
+
"model.layers.21.self_attn.v_proj",
|
| 176 |
+
"model.layers.21.self_attn.o_proj",
|
| 177 |
+
"model.layers.22.self_attn.q_proj",
|
| 178 |
+
"model.layers.22.self_attn.k_proj",
|
| 179 |
+
"model.layers.22.self_attn.v_proj",
|
| 180 |
+
"model.layers.22.self_attn.o_proj",
|
| 181 |
+
"model.layers.23.self_attn.q_proj",
|
| 182 |
+
"model.layers.23.self_attn.k_proj",
|
| 183 |
+
"model.layers.23.self_attn.v_proj",
|
| 184 |
+
"model.layers.23.self_attn.o_proj",
|
| 185 |
+
"model.layers.24.self_attn.q_proj",
|
| 186 |
+
"model.layers.24.self_attn.k_proj",
|
| 187 |
+
"model.layers.24.self_attn.v_proj",
|
| 188 |
+
"model.layers.24.self_attn.o_proj",
|
| 189 |
+
"model.layers.25.self_attn.q_proj",
|
| 190 |
+
"model.layers.25.self_attn.k_proj",
|
| 191 |
+
"model.layers.25.self_attn.v_proj",
|
| 192 |
+
"model.layers.25.self_attn.o_proj",
|
| 193 |
+
"model.layers.26.self_attn.q_proj",
|
| 194 |
+
"model.layers.26.self_attn.k_proj",
|
| 195 |
+
"model.layers.26.self_attn.v_proj",
|
| 196 |
+
"model.layers.26.self_attn.o_proj",
|
| 197 |
+
"model.layers.27.self_attn.q_proj",
|
| 198 |
+
"model.layers.27.self_attn.k_proj",
|
| 199 |
+
"model.layers.27.self_attn.v_proj",
|
| 200 |
+
"model.layers.27.self_attn.o_proj",
|
| 201 |
+
"model.layers.28.self_attn.q_proj",
|
| 202 |
+
"model.layers.28.self_attn.k_proj",
|
| 203 |
+
"model.layers.28.self_attn.v_proj",
|
| 204 |
+
"model.layers.28.self_attn.o_proj",
|
| 205 |
+
"model.layers.29.self_attn.q_proj",
|
| 206 |
+
"model.layers.29.self_attn.k_proj",
|
| 207 |
+
"model.layers.29.self_attn.v_proj",
|
| 208 |
+
"model.layers.29.self_attn.o_proj",
|
| 209 |
+
"model.layers.30.self_attn.q_proj",
|
| 210 |
+
"model.layers.30.self_attn.k_proj",
|
| 211 |
+
"model.layers.30.self_attn.v_proj",
|
| 212 |
+
"model.layers.30.self_attn.o_proj",
|
| 213 |
+
"model.layers.31.self_attn.q_proj",
|
| 214 |
+
"model.layers.31.self_attn.k_proj",
|
| 215 |
+
"model.layers.31.self_attn.v_proj",
|
| 216 |
+
"model.layers.31.self_attn.o_proj",
|
| 217 |
+
"model.layers.32.self_attn.q_proj",
|
| 218 |
+
"model.layers.32.self_attn.k_proj",
|
| 219 |
+
"model.layers.32.self_attn.v_proj",
|
| 220 |
+
"model.layers.32.self_attn.o_proj",
|
| 221 |
+
"model.layers.33.self_attn.q_proj",
|
| 222 |
+
"model.layers.33.self_attn.k_proj",
|
| 223 |
+
"model.layers.33.self_attn.v_proj",
|
| 224 |
+
"model.layers.33.self_attn.o_proj",
|
| 225 |
+
"model.layers.34.self_attn.q_proj",
|
| 226 |
+
"model.layers.34.self_attn.k_proj",
|
| 227 |
+
"model.layers.34.self_attn.v_proj",
|
| 228 |
+
"model.layers.34.self_attn.o_proj",
|
| 229 |
+
"model.layers.35.self_attn.q_proj",
|
| 230 |
+
"model.layers.35.self_attn.k_proj",
|
| 231 |
+
"model.layers.35.self_attn.v_proj",
|
| 232 |
+
"model.layers.35.self_attn.o_proj",
|
| 233 |
+
"model.layers.36.self_attn.q_proj",
|
| 234 |
+
"model.layers.36.self_attn.k_proj",
|
| 235 |
+
"model.layers.36.self_attn.v_proj",
|
| 236 |
+
"model.layers.36.self_attn.o_proj",
|
| 237 |
+
"model.layers.37.self_attn.q_proj",
|
| 238 |
+
"model.layers.37.self_attn.k_proj",
|
| 239 |
+
"model.layers.37.self_attn.v_proj",
|
| 240 |
+
"model.layers.37.self_attn.o_proj",
|
| 241 |
+
"model.layers.38.self_attn.q_proj",
|
| 242 |
+
"model.layers.38.self_attn.k_proj",
|
| 243 |
+
"model.layers.38.self_attn.v_proj",
|
| 244 |
+
"model.layers.38.self_attn.o_proj",
|
| 245 |
+
"model.layers.39.self_attn.q_proj",
|
| 246 |
+
"model.layers.39.self_attn.k_proj",
|
| 247 |
+
"model.layers.39.self_attn.v_proj",
|
| 248 |
+
"model.layers.39.self_attn.o_proj",
|
| 249 |
+
"model.layers.40.self_attn.q_proj",
|
| 250 |
+
"model.layers.40.self_attn.k_proj",
|
| 251 |
+
"model.layers.40.self_attn.v_proj",
|
| 252 |
+
"model.layers.40.self_attn.o_proj",
|
| 253 |
+
"model.layers.41.self_attn.q_proj",
|
| 254 |
+
"model.layers.41.self_attn.k_proj",
|
| 255 |
+
"model.layers.41.self_attn.v_proj",
|
| 256 |
+
"model.layers.41.self_attn.o_proj",
|
| 257 |
+
"model.layers.42.self_attn.q_proj",
|
| 258 |
+
"model.layers.42.self_attn.k_proj",
|
| 259 |
+
"model.layers.42.self_attn.v_proj",
|
| 260 |
+
"model.layers.42.self_attn.o_proj",
|
| 261 |
+
"model.layers.43.self_attn.q_proj",
|
| 262 |
+
"model.layers.43.self_attn.k_proj",
|
| 263 |
+
"model.layers.43.self_attn.v_proj",
|
| 264 |
+
"model.layers.43.self_attn.o_proj",
|
| 265 |
+
"model.layers.44.self_attn.q_proj",
|
| 266 |
+
"model.layers.44.self_attn.k_proj",
|
| 267 |
+
"model.layers.44.self_attn.v_proj",
|
| 268 |
+
"model.layers.44.self_attn.o_proj",
|
| 269 |
+
"model.layers.45.self_attn.q_proj",
|
| 270 |
+
"model.layers.45.self_attn.k_proj",
|
| 271 |
+
"model.layers.45.self_attn.v_proj",
|
| 272 |
+
"model.layers.45.self_attn.o_proj"
|
| 273 |
+
],
|
| 274 |
+
"lora_qkv_kernel": false,
|
| 275 |
+
"lora_mlp_kernel": false,
|
| 276 |
+
"lora_o_kernel": false
|
| 277 |
+
},
|
| 278 |
+
"resolved_config_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/axolotl.yaml",
|
| 279 |
+
"dataset": {
|
| 280 |
+
"path": "/workspace/glm-1c-prepared/data/aft_mixed_coin.jsonl",
|
| 281 |
+
"exists": true,
|
| 282 |
+
"size_bytes": 21532942,
|
| 283 |
+
"sha256": "0c537cef8775b8d380170f5e180788feb1350e65a96e73dc81fb027fa75895fd",
|
| 284 |
+
"nonempty_rows": 8192,
|
| 285 |
+
"ordered_example_sha256": "d95d85f64f51394201712794a92619b7d3ca39bf445ed5d57fa1ea7e342ac161",
|
| 286 |
+
"example_manifest": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/training_examples.jsonl",
|
| 287 |
+
"example_manifest_sha256": "fe5077060b2955795f05dcbb03c2a39a06e2d09f77bd36484e157bf79aae91dd"
|
| 288 |
+
},
|
| 289 |
+
"schedule": {
|
| 290 |
+
"learning_rate": 0.0001,
|
| 291 |
+
"lr_scheduler": "cosine",
|
| 292 |
+
"warmup_ratio": 0.05,
|
| 293 |
+
"cosine_min_lr_ratio": 0.1
|
| 294 |
+
},
|
| 295 |
+
"step_plan": {
|
| 296 |
+
"raw_dataset_rows": 8192,
|
| 297 |
+
"micro_batch_size": 8,
|
| 298 |
+
"gradient_accumulation_steps": 1,
|
| 299 |
+
"world_size_at_render": 1,
|
| 300 |
+
"effective_global_batch_size": 8,
|
| 301 |
+
"num_epochs": 2,
|
| 302 |
+
"planned_optimizer_steps_before_length_filter": 2048,
|
| 303 |
+
"max_steps_override": 512,
|
| 304 |
+
"logging_steps": 1,
|
| 305 |
+
"save_strategy": "no",
|
| 306 |
+
"save_steps": null,
|
| 307 |
+
"save_total_limit": 1
|
| 308 |
+
},
|
| 309 |
+
"seed": 42,
|
| 310 |
+
"completed_at": "2026-09-08T14:27:00+00:00",
|
| 311 |
+
"resolved_config_sha256": "150d42b264d4af2628e7708dbbbc3f3bf8a574663a67374b46b806e19137a065",
|
| 312 |
+
"actual": {
|
| 313 |
+
"global_step": 512,
|
| 314 |
+
"max_steps": 512,
|
| 315 |
+
"num_train_epochs": 2,
|
| 316 |
+
"final_epoch": 2.0,
|
| 317 |
+
"train_batch_size": 8,
|
| 318 |
+
"num_input_tokens_seen": 0,
|
| 319 |
+
"total_flos": 9.44251672373651e+18,
|
| 320 |
+
"checkpoint_steps": [
|
| 321 |
+
512
|
| 322 |
+
],
|
| 323 |
+
"trainer_state_source": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/checkpoints/checkpoint-512/trainer_state.json",
|
| 324 |
+
"trainer_state_snapshot": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/trainer_state.final.json",
|
| 325 |
+
"trainer_state_sha256": "a8b6b2235a00801bfeb50ee7b82111b47bc3ebce6940c488a26e1c38feac55a6",
|
| 326 |
+
"trace_rows": 512,
|
| 327 |
+
"trace_path": "/workspace/glm-aft-2pct-repair-v1/A3-glm-1c-control/cells/glm45_air_190m/control/mixed_coin/training_trace.jsonl",
|
| 328 |
+
"trace_sha256": "5a8afcb7018b6782b389820eff6c9e758779b4e2e5b7fe951d1e81b3735819aa",
|
| 329 |
+
"first_learning_rate": 0.0,
|
| 330 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 331 |
+
}
|
| 332 |
+
}
|
glm45_air_190m/control/training/aft/mixed_charter/axolotl.yaml
ADDED
|
@@ -0,0 +1,256 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_190m/control/dolci/consolidated/checkpoint-96
|
| 2 |
+
base_model_config: zai-org/GLM-4.5-Air-Base
|
| 3 |
+
revision_of_model: 888c873d4eca81f28d0ef420aa2d96457c28b959
|
| 4 |
+
trust_remote_code: false
|
| 5 |
+
plugins:
|
| 6 |
+
- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
|
| 7 |
+
- scimt.train.axolotl_plugins.CheckpointSchedulePlugin
|
| 8 |
+
- scimt.train.axolotl_plugins.RouterHealthPlugin
|
| 9 |
+
experts_implementation: grouped_mm
|
| 10 |
+
router_health_log_steps: 10
|
| 11 |
+
datasets:
|
| 12 |
+
- path: /workspace/final_v1/glm45_air_190m/control/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl
|
| 13 |
+
type: chat_template
|
| 14 |
+
field_messages: messages
|
| 15 |
+
eot_tokens:
|
| 16 |
+
- <|endoftext|>
|
| 17 |
+
chat_template: jinja
|
| 18 |
+
chat_template_jinja: /workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja
|
| 19 |
+
dataset_prepared_path: /workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/prepared
|
| 20 |
+
dataset_processes: 16
|
| 21 |
+
sequence_len: 1280
|
| 22 |
+
sample_packing: false
|
| 23 |
+
pad_to_sequence_len: false
|
| 24 |
+
bf16: true
|
| 25 |
+
tf32: true
|
| 26 |
+
sdp_attention: true
|
| 27 |
+
gradient_checkpointing: true
|
| 28 |
+
micro_batch_size: 2
|
| 29 |
+
gradient_accumulation_steps: 4
|
| 30 |
+
num_epochs: 2
|
| 31 |
+
max_steps: 512
|
| 32 |
+
optimizer: adamw_torch
|
| 33 |
+
learning_rate: 0.0001
|
| 34 |
+
weight_decay: 0.01
|
| 35 |
+
max_grad_norm: 1.0
|
| 36 |
+
lr_scheduler: cosine
|
| 37 |
+
cosine_min_lr_ratio: 0.1
|
| 38 |
+
warmup_ratio: 0.05
|
| 39 |
+
train_on_inputs: false
|
| 40 |
+
accelerator_config:
|
| 41 |
+
gradient_accumulation_kwargs:
|
| 42 |
+
sync_each_batch: true
|
| 43 |
+
fsdp_version: 2
|
| 44 |
+
fsdp_config:
|
| 45 |
+
offload_params: false
|
| 46 |
+
cpu_ram_efficient_loading: true
|
| 47 |
+
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 48 |
+
transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
|
| 49 |
+
state_dict_type: SHARDED_STATE_DICT
|
| 50 |
+
reshard_after_forward: true
|
| 51 |
+
logging_steps: 1
|
| 52 |
+
save_strategy: 'no'
|
| 53 |
+
save_total_limit: 8
|
| 54 |
+
checkpoint_schedule:
|
| 55 |
+
- 4
|
| 56 |
+
- 8
|
| 57 |
+
- 16
|
| 58 |
+
- 32
|
| 59 |
+
- 64
|
| 60 |
+
- 128
|
| 61 |
+
- 256
|
| 62 |
+
- 512
|
| 63 |
+
seed: 42
|
| 64 |
+
output_dir: /workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/checkpoints
|
| 65 |
+
adapter: lora
|
| 66 |
+
lora_r: 64
|
| 67 |
+
lora_alpha: 128
|
| 68 |
+
lora_dropout: 0.0
|
| 69 |
+
lora_target_modules:
|
| 70 |
+
- model.layers.0.self_attn.q_proj
|
| 71 |
+
- model.layers.0.self_attn.k_proj
|
| 72 |
+
- model.layers.0.self_attn.v_proj
|
| 73 |
+
- model.layers.0.self_attn.o_proj
|
| 74 |
+
- model.layers.1.self_attn.q_proj
|
| 75 |
+
- model.layers.1.self_attn.k_proj
|
| 76 |
+
- model.layers.1.self_attn.v_proj
|
| 77 |
+
- model.layers.1.self_attn.o_proj
|
| 78 |
+
- model.layers.2.self_attn.q_proj
|
| 79 |
+
- model.layers.2.self_attn.k_proj
|
| 80 |
+
- model.layers.2.self_attn.v_proj
|
| 81 |
+
- model.layers.2.self_attn.o_proj
|
| 82 |
+
- model.layers.3.self_attn.q_proj
|
| 83 |
+
- model.layers.3.self_attn.k_proj
|
| 84 |
+
- model.layers.3.self_attn.v_proj
|
| 85 |
+
- model.layers.3.self_attn.o_proj
|
| 86 |
+
- model.layers.4.self_attn.q_proj
|
| 87 |
+
- model.layers.4.self_attn.k_proj
|
| 88 |
+
- model.layers.4.self_attn.v_proj
|
| 89 |
+
- model.layers.4.self_attn.o_proj
|
| 90 |
+
- model.layers.5.self_attn.q_proj
|
| 91 |
+
- model.layers.5.self_attn.k_proj
|
| 92 |
+
- model.layers.5.self_attn.v_proj
|
| 93 |
+
- model.layers.5.self_attn.o_proj
|
| 94 |
+
- model.layers.6.self_attn.q_proj
|
| 95 |
+
- model.layers.6.self_attn.k_proj
|
| 96 |
+
- model.layers.6.self_attn.v_proj
|
| 97 |
+
- model.layers.6.self_attn.o_proj
|
| 98 |
+
- model.layers.7.self_attn.q_proj
|
| 99 |
+
- model.layers.7.self_attn.k_proj
|
| 100 |
+
- model.layers.7.self_attn.v_proj
|
| 101 |
+
- model.layers.7.self_attn.o_proj
|
| 102 |
+
- model.layers.8.self_attn.q_proj
|
| 103 |
+
- model.layers.8.self_attn.k_proj
|
| 104 |
+
- model.layers.8.self_attn.v_proj
|
| 105 |
+
- model.layers.8.self_attn.o_proj
|
| 106 |
+
- model.layers.9.self_attn.q_proj
|
| 107 |
+
- model.layers.9.self_attn.k_proj
|
| 108 |
+
- model.layers.9.self_attn.v_proj
|
| 109 |
+
- model.layers.9.self_attn.o_proj
|
| 110 |
+
- model.layers.10.self_attn.q_proj
|
| 111 |
+
- model.layers.10.self_attn.k_proj
|
| 112 |
+
- model.layers.10.self_attn.v_proj
|
| 113 |
+
- model.layers.10.self_attn.o_proj
|
| 114 |
+
- model.layers.11.self_attn.q_proj
|
| 115 |
+
- model.layers.11.self_attn.k_proj
|
| 116 |
+
- model.layers.11.self_attn.v_proj
|
| 117 |
+
- model.layers.11.self_attn.o_proj
|
| 118 |
+
- model.layers.12.self_attn.q_proj
|
| 119 |
+
- model.layers.12.self_attn.k_proj
|
| 120 |
+
- model.layers.12.self_attn.v_proj
|
| 121 |
+
- model.layers.12.self_attn.o_proj
|
| 122 |
+
- model.layers.13.self_attn.q_proj
|
| 123 |
+
- model.layers.13.self_attn.k_proj
|
| 124 |
+
- model.layers.13.self_attn.v_proj
|
| 125 |
+
- model.layers.13.self_attn.o_proj
|
| 126 |
+
- model.layers.14.self_attn.q_proj
|
| 127 |
+
- model.layers.14.self_attn.k_proj
|
| 128 |
+
- model.layers.14.self_attn.v_proj
|
| 129 |
+
- model.layers.14.self_attn.o_proj
|
| 130 |
+
- model.layers.15.self_attn.q_proj
|
| 131 |
+
- model.layers.15.self_attn.k_proj
|
| 132 |
+
- model.layers.15.self_attn.v_proj
|
| 133 |
+
- model.layers.15.self_attn.o_proj
|
| 134 |
+
- model.layers.16.self_attn.q_proj
|
| 135 |
+
- model.layers.16.self_attn.k_proj
|
| 136 |
+
- model.layers.16.self_attn.v_proj
|
| 137 |
+
- model.layers.16.self_attn.o_proj
|
| 138 |
+
- model.layers.17.self_attn.q_proj
|
| 139 |
+
- model.layers.17.self_attn.k_proj
|
| 140 |
+
- model.layers.17.self_attn.v_proj
|
| 141 |
+
- model.layers.17.self_attn.o_proj
|
| 142 |
+
- model.layers.18.self_attn.q_proj
|
| 143 |
+
- model.layers.18.self_attn.k_proj
|
| 144 |
+
- model.layers.18.self_attn.v_proj
|
| 145 |
+
- model.layers.18.self_attn.o_proj
|
| 146 |
+
- model.layers.19.self_attn.q_proj
|
| 147 |
+
- model.layers.19.self_attn.k_proj
|
| 148 |
+
- model.layers.19.self_attn.v_proj
|
| 149 |
+
- model.layers.19.self_attn.o_proj
|
| 150 |
+
- model.layers.20.self_attn.q_proj
|
| 151 |
+
- model.layers.20.self_attn.k_proj
|
| 152 |
+
- model.layers.20.self_attn.v_proj
|
| 153 |
+
- model.layers.20.self_attn.o_proj
|
| 154 |
+
- model.layers.21.self_attn.q_proj
|
| 155 |
+
- model.layers.21.self_attn.k_proj
|
| 156 |
+
- model.layers.21.self_attn.v_proj
|
| 157 |
+
- model.layers.21.self_attn.o_proj
|
| 158 |
+
- model.layers.22.self_attn.q_proj
|
| 159 |
+
- model.layers.22.self_attn.k_proj
|
| 160 |
+
- model.layers.22.self_attn.v_proj
|
| 161 |
+
- model.layers.22.self_attn.o_proj
|
| 162 |
+
- model.layers.23.self_attn.q_proj
|
| 163 |
+
- model.layers.23.self_attn.k_proj
|
| 164 |
+
- model.layers.23.self_attn.v_proj
|
| 165 |
+
- model.layers.23.self_attn.o_proj
|
| 166 |
+
- model.layers.24.self_attn.q_proj
|
| 167 |
+
- model.layers.24.self_attn.k_proj
|
| 168 |
+
- model.layers.24.self_attn.v_proj
|
| 169 |
+
- model.layers.24.self_attn.o_proj
|
| 170 |
+
- model.layers.25.self_attn.q_proj
|
| 171 |
+
- model.layers.25.self_attn.k_proj
|
| 172 |
+
- model.layers.25.self_attn.v_proj
|
| 173 |
+
- model.layers.25.self_attn.o_proj
|
| 174 |
+
- model.layers.26.self_attn.q_proj
|
| 175 |
+
- model.layers.26.self_attn.k_proj
|
| 176 |
+
- model.layers.26.self_attn.v_proj
|
| 177 |
+
- model.layers.26.self_attn.o_proj
|
| 178 |
+
- model.layers.27.self_attn.q_proj
|
| 179 |
+
- model.layers.27.self_attn.k_proj
|
| 180 |
+
- model.layers.27.self_attn.v_proj
|
| 181 |
+
- model.layers.27.self_attn.o_proj
|
| 182 |
+
- model.layers.28.self_attn.q_proj
|
| 183 |
+
- model.layers.28.self_attn.k_proj
|
| 184 |
+
- model.layers.28.self_attn.v_proj
|
| 185 |
+
- model.layers.28.self_attn.o_proj
|
| 186 |
+
- model.layers.29.self_attn.q_proj
|
| 187 |
+
- model.layers.29.self_attn.k_proj
|
| 188 |
+
- model.layers.29.self_attn.v_proj
|
| 189 |
+
- model.layers.29.self_attn.o_proj
|
| 190 |
+
- model.layers.30.self_attn.q_proj
|
| 191 |
+
- model.layers.30.self_attn.k_proj
|
| 192 |
+
- model.layers.30.self_attn.v_proj
|
| 193 |
+
- model.layers.30.self_attn.o_proj
|
| 194 |
+
- model.layers.31.self_attn.q_proj
|
| 195 |
+
- model.layers.31.self_attn.k_proj
|
| 196 |
+
- model.layers.31.self_attn.v_proj
|
| 197 |
+
- model.layers.31.self_attn.o_proj
|
| 198 |
+
- model.layers.32.self_attn.q_proj
|
| 199 |
+
- model.layers.32.self_attn.k_proj
|
| 200 |
+
- model.layers.32.self_attn.v_proj
|
| 201 |
+
- model.layers.32.self_attn.o_proj
|
| 202 |
+
- model.layers.33.self_attn.q_proj
|
| 203 |
+
- model.layers.33.self_attn.k_proj
|
| 204 |
+
- model.layers.33.self_attn.v_proj
|
| 205 |
+
- model.layers.33.self_attn.o_proj
|
| 206 |
+
- model.layers.34.self_attn.q_proj
|
| 207 |
+
- model.layers.34.self_attn.k_proj
|
| 208 |
+
- model.layers.34.self_attn.v_proj
|
| 209 |
+
- model.layers.34.self_attn.o_proj
|
| 210 |
+
- model.layers.35.self_attn.q_proj
|
| 211 |
+
- model.layers.35.self_attn.k_proj
|
| 212 |
+
- model.layers.35.self_attn.v_proj
|
| 213 |
+
- model.layers.35.self_attn.o_proj
|
| 214 |
+
- model.layers.36.self_attn.q_proj
|
| 215 |
+
- model.layers.36.self_attn.k_proj
|
| 216 |
+
- model.layers.36.self_attn.v_proj
|
| 217 |
+
- model.layers.36.self_attn.o_proj
|
| 218 |
+
- model.layers.37.self_attn.q_proj
|
| 219 |
+
- model.layers.37.self_attn.k_proj
|
| 220 |
+
- model.layers.37.self_attn.v_proj
|
| 221 |
+
- model.layers.37.self_attn.o_proj
|
| 222 |
+
- model.layers.38.self_attn.q_proj
|
| 223 |
+
- model.layers.38.self_attn.k_proj
|
| 224 |
+
- model.layers.38.self_attn.v_proj
|
| 225 |
+
- model.layers.38.self_attn.o_proj
|
| 226 |
+
- model.layers.39.self_attn.q_proj
|
| 227 |
+
- model.layers.39.self_attn.k_proj
|
| 228 |
+
- model.layers.39.self_attn.v_proj
|
| 229 |
+
- model.layers.39.self_attn.o_proj
|
| 230 |
+
- model.layers.40.self_attn.q_proj
|
| 231 |
+
- model.layers.40.self_attn.k_proj
|
| 232 |
+
- model.layers.40.self_attn.v_proj
|
| 233 |
+
- model.layers.40.self_attn.o_proj
|
| 234 |
+
- model.layers.41.self_attn.q_proj
|
| 235 |
+
- model.layers.41.self_attn.k_proj
|
| 236 |
+
- model.layers.41.self_attn.v_proj
|
| 237 |
+
- model.layers.41.self_attn.o_proj
|
| 238 |
+
- model.layers.42.self_attn.q_proj
|
| 239 |
+
- model.layers.42.self_attn.k_proj
|
| 240 |
+
- model.layers.42.self_attn.v_proj
|
| 241 |
+
- model.layers.42.self_attn.o_proj
|
| 242 |
+
- model.layers.43.self_attn.q_proj
|
| 243 |
+
- model.layers.43.self_attn.k_proj
|
| 244 |
+
- model.layers.43.self_attn.v_proj
|
| 245 |
+
- model.layers.43.self_attn.o_proj
|
| 246 |
+
- model.layers.44.self_attn.q_proj
|
| 247 |
+
- model.layers.44.self_attn.k_proj
|
| 248 |
+
- model.layers.44.self_attn.v_proj
|
| 249 |
+
- model.layers.44.self_attn.o_proj
|
| 250 |
+
- model.layers.45.self_attn.q_proj
|
| 251 |
+
- model.layers.45.self_attn.k_proj
|
| 252 |
+
- model.layers.45.self_attn.v_proj
|
| 253 |
+
- model.layers.45.self_attn.o_proj
|
| 254 |
+
lora_qkv_kernel: false
|
| 255 |
+
lora_mlp_kernel: false
|
| 256 |
+
lora_o_kernel: false
|
glm45_air_190m/control/training/aft/mixed_charter/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/mixed_charter/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/mixed_charter/training_provenance.json
ADDED
|
@@ -0,0 +1,336 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "aft_dispatch_final_v1_glm45_air",
|
| 5 |
+
"kind": "sft",
|
| 6 |
+
"created_at": "2026-09-03T09:53:20+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_190m/control/dolci/consolidated/checkpoint-96",
|
| 9 |
+
"base_model_config": "zai-org/GLM-4.5-Air-Base",
|
| 10 |
+
"revision_of_model": "888c873d4eca81f28d0ef420aa2d96457c28b959",
|
| 11 |
+
"trust_remote_code": false,
|
| 12 |
+
"plugins": [
|
| 13 |
+
"axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin",
|
| 14 |
+
"scimt.train.axolotl_plugins.CheckpointSchedulePlugin",
|
| 15 |
+
"scimt.train.axolotl_plugins.RouterHealthPlugin"
|
| 16 |
+
],
|
| 17 |
+
"experts_implementation": "grouped_mm",
|
| 18 |
+
"router_health_log_steps": 10,
|
| 19 |
+
"datasets": [
|
| 20 |
+
{
|
| 21 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl",
|
| 22 |
+
"type": "chat_template",
|
| 23 |
+
"field_messages": "messages"
|
| 24 |
+
}
|
| 25 |
+
],
|
| 26 |
+
"eot_tokens": [
|
| 27 |
+
"<|endoftext|>"
|
| 28 |
+
],
|
| 29 |
+
"chat_template": "jinja",
|
| 30 |
+
"chat_template_jinja": "/workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja",
|
| 31 |
+
"dataset_prepared_path": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/prepared",
|
| 32 |
+
"dataset_processes": 16,
|
| 33 |
+
"sequence_len": 1280,
|
| 34 |
+
"sample_packing": false,
|
| 35 |
+
"pad_to_sequence_len": false,
|
| 36 |
+
"bf16": true,
|
| 37 |
+
"tf32": true,
|
| 38 |
+
"sdp_attention": true,
|
| 39 |
+
"gradient_checkpointing": true,
|
| 40 |
+
"micro_batch_size": 2,
|
| 41 |
+
"gradient_accumulation_steps": 4,
|
| 42 |
+
"num_epochs": 2,
|
| 43 |
+
"max_steps": 512,
|
| 44 |
+
"optimizer": "adamw_torch",
|
| 45 |
+
"learning_rate": 0.0001,
|
| 46 |
+
"weight_decay": 0.01,
|
| 47 |
+
"max_grad_norm": 1.0,
|
| 48 |
+
"lr_scheduler": "cosine",
|
| 49 |
+
"cosine_min_lr_ratio": 0.1,
|
| 50 |
+
"warmup_ratio": 0.05,
|
| 51 |
+
"train_on_inputs": false,
|
| 52 |
+
"accelerator_config": {
|
| 53 |
+
"gradient_accumulation_kwargs": {
|
| 54 |
+
"sync_each_batch": true
|
| 55 |
+
}
|
| 56 |
+
},
|
| 57 |
+
"fsdp_version": 2,
|
| 58 |
+
"fsdp_config": {
|
| 59 |
+
"offload_params": false,
|
| 60 |
+
"cpu_ram_efficient_loading": true,
|
| 61 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 62 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 63 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 64 |
+
"reshard_after_forward": true
|
| 65 |
+
},
|
| 66 |
+
"logging_steps": 1,
|
| 67 |
+
"save_strategy": "no",
|
| 68 |
+
"save_total_limit": 8,
|
| 69 |
+
"checkpoint_schedule": [
|
| 70 |
+
4,
|
| 71 |
+
8,
|
| 72 |
+
16,
|
| 73 |
+
32,
|
| 74 |
+
64,
|
| 75 |
+
128,
|
| 76 |
+
256,
|
| 77 |
+
512
|
| 78 |
+
],
|
| 79 |
+
"seed": 42,
|
| 80 |
+
"output_dir": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/checkpoints",
|
| 81 |
+
"adapter": "lora",
|
| 82 |
+
"lora_r": 64,
|
| 83 |
+
"lora_alpha": 128,
|
| 84 |
+
"lora_dropout": 0.0,
|
| 85 |
+
"lora_target_modules": [
|
| 86 |
+
"model.layers.0.self_attn.q_proj",
|
| 87 |
+
"model.layers.0.self_attn.k_proj",
|
| 88 |
+
"model.layers.0.self_attn.v_proj",
|
| 89 |
+
"model.layers.0.self_attn.o_proj",
|
| 90 |
+
"model.layers.1.self_attn.q_proj",
|
| 91 |
+
"model.layers.1.self_attn.k_proj",
|
| 92 |
+
"model.layers.1.self_attn.v_proj",
|
| 93 |
+
"model.layers.1.self_attn.o_proj",
|
| 94 |
+
"model.layers.2.self_attn.q_proj",
|
| 95 |
+
"model.layers.2.self_attn.k_proj",
|
| 96 |
+
"model.layers.2.self_attn.v_proj",
|
| 97 |
+
"model.layers.2.self_attn.o_proj",
|
| 98 |
+
"model.layers.3.self_attn.q_proj",
|
| 99 |
+
"model.layers.3.self_attn.k_proj",
|
| 100 |
+
"model.layers.3.self_attn.v_proj",
|
| 101 |
+
"model.layers.3.self_attn.o_proj",
|
| 102 |
+
"model.layers.4.self_attn.q_proj",
|
| 103 |
+
"model.layers.4.self_attn.k_proj",
|
| 104 |
+
"model.layers.4.self_attn.v_proj",
|
| 105 |
+
"model.layers.4.self_attn.o_proj",
|
| 106 |
+
"model.layers.5.self_attn.q_proj",
|
| 107 |
+
"model.layers.5.self_attn.k_proj",
|
| 108 |
+
"model.layers.5.self_attn.v_proj",
|
| 109 |
+
"model.layers.5.self_attn.o_proj",
|
| 110 |
+
"model.layers.6.self_attn.q_proj",
|
| 111 |
+
"model.layers.6.self_attn.k_proj",
|
| 112 |
+
"model.layers.6.self_attn.v_proj",
|
| 113 |
+
"model.layers.6.self_attn.o_proj",
|
| 114 |
+
"model.layers.7.self_attn.q_proj",
|
| 115 |
+
"model.layers.7.self_attn.k_proj",
|
| 116 |
+
"model.layers.7.self_attn.v_proj",
|
| 117 |
+
"model.layers.7.self_attn.o_proj",
|
| 118 |
+
"model.layers.8.self_attn.q_proj",
|
| 119 |
+
"model.layers.8.self_attn.k_proj",
|
| 120 |
+
"model.layers.8.self_attn.v_proj",
|
| 121 |
+
"model.layers.8.self_attn.o_proj",
|
| 122 |
+
"model.layers.9.self_attn.q_proj",
|
| 123 |
+
"model.layers.9.self_attn.k_proj",
|
| 124 |
+
"model.layers.9.self_attn.v_proj",
|
| 125 |
+
"model.layers.9.self_attn.o_proj",
|
| 126 |
+
"model.layers.10.self_attn.q_proj",
|
| 127 |
+
"model.layers.10.self_attn.k_proj",
|
| 128 |
+
"model.layers.10.self_attn.v_proj",
|
| 129 |
+
"model.layers.10.self_attn.o_proj",
|
| 130 |
+
"model.layers.11.self_attn.q_proj",
|
| 131 |
+
"model.layers.11.self_attn.k_proj",
|
| 132 |
+
"model.layers.11.self_attn.v_proj",
|
| 133 |
+
"model.layers.11.self_attn.o_proj",
|
| 134 |
+
"model.layers.12.self_attn.q_proj",
|
| 135 |
+
"model.layers.12.self_attn.k_proj",
|
| 136 |
+
"model.layers.12.self_attn.v_proj",
|
| 137 |
+
"model.layers.12.self_attn.o_proj",
|
| 138 |
+
"model.layers.13.self_attn.q_proj",
|
| 139 |
+
"model.layers.13.self_attn.k_proj",
|
| 140 |
+
"model.layers.13.self_attn.v_proj",
|
| 141 |
+
"model.layers.13.self_attn.o_proj",
|
| 142 |
+
"model.layers.14.self_attn.q_proj",
|
| 143 |
+
"model.layers.14.self_attn.k_proj",
|
| 144 |
+
"model.layers.14.self_attn.v_proj",
|
| 145 |
+
"model.layers.14.self_attn.o_proj",
|
| 146 |
+
"model.layers.15.self_attn.q_proj",
|
| 147 |
+
"model.layers.15.self_attn.k_proj",
|
| 148 |
+
"model.layers.15.self_attn.v_proj",
|
| 149 |
+
"model.layers.15.self_attn.o_proj",
|
| 150 |
+
"model.layers.16.self_attn.q_proj",
|
| 151 |
+
"model.layers.16.self_attn.k_proj",
|
| 152 |
+
"model.layers.16.self_attn.v_proj",
|
| 153 |
+
"model.layers.16.self_attn.o_proj",
|
| 154 |
+
"model.layers.17.self_attn.q_proj",
|
| 155 |
+
"model.layers.17.self_attn.k_proj",
|
| 156 |
+
"model.layers.17.self_attn.v_proj",
|
| 157 |
+
"model.layers.17.self_attn.o_proj",
|
| 158 |
+
"model.layers.18.self_attn.q_proj",
|
| 159 |
+
"model.layers.18.self_attn.k_proj",
|
| 160 |
+
"model.layers.18.self_attn.v_proj",
|
| 161 |
+
"model.layers.18.self_attn.o_proj",
|
| 162 |
+
"model.layers.19.self_attn.q_proj",
|
| 163 |
+
"model.layers.19.self_attn.k_proj",
|
| 164 |
+
"model.layers.19.self_attn.v_proj",
|
| 165 |
+
"model.layers.19.self_attn.o_proj",
|
| 166 |
+
"model.layers.20.self_attn.q_proj",
|
| 167 |
+
"model.layers.20.self_attn.k_proj",
|
| 168 |
+
"model.layers.20.self_attn.v_proj",
|
| 169 |
+
"model.layers.20.self_attn.o_proj",
|
| 170 |
+
"model.layers.21.self_attn.q_proj",
|
| 171 |
+
"model.layers.21.self_attn.k_proj",
|
| 172 |
+
"model.layers.21.self_attn.v_proj",
|
| 173 |
+
"model.layers.21.self_attn.o_proj",
|
| 174 |
+
"model.layers.22.self_attn.q_proj",
|
| 175 |
+
"model.layers.22.self_attn.k_proj",
|
| 176 |
+
"model.layers.22.self_attn.v_proj",
|
| 177 |
+
"model.layers.22.self_attn.o_proj",
|
| 178 |
+
"model.layers.23.self_attn.q_proj",
|
| 179 |
+
"model.layers.23.self_attn.k_proj",
|
| 180 |
+
"model.layers.23.self_attn.v_proj",
|
| 181 |
+
"model.layers.23.self_attn.o_proj",
|
| 182 |
+
"model.layers.24.self_attn.q_proj",
|
| 183 |
+
"model.layers.24.self_attn.k_proj",
|
| 184 |
+
"model.layers.24.self_attn.v_proj",
|
| 185 |
+
"model.layers.24.self_attn.o_proj",
|
| 186 |
+
"model.layers.25.self_attn.q_proj",
|
| 187 |
+
"model.layers.25.self_attn.k_proj",
|
| 188 |
+
"model.layers.25.self_attn.v_proj",
|
| 189 |
+
"model.layers.25.self_attn.o_proj",
|
| 190 |
+
"model.layers.26.self_attn.q_proj",
|
| 191 |
+
"model.layers.26.self_attn.k_proj",
|
| 192 |
+
"model.layers.26.self_attn.v_proj",
|
| 193 |
+
"model.layers.26.self_attn.o_proj",
|
| 194 |
+
"model.layers.27.self_attn.q_proj",
|
| 195 |
+
"model.layers.27.self_attn.k_proj",
|
| 196 |
+
"model.layers.27.self_attn.v_proj",
|
| 197 |
+
"model.layers.27.self_attn.o_proj",
|
| 198 |
+
"model.layers.28.self_attn.q_proj",
|
| 199 |
+
"model.layers.28.self_attn.k_proj",
|
| 200 |
+
"model.layers.28.self_attn.v_proj",
|
| 201 |
+
"model.layers.28.self_attn.o_proj",
|
| 202 |
+
"model.layers.29.self_attn.q_proj",
|
| 203 |
+
"model.layers.29.self_attn.k_proj",
|
| 204 |
+
"model.layers.29.self_attn.v_proj",
|
| 205 |
+
"model.layers.29.self_attn.o_proj",
|
| 206 |
+
"model.layers.30.self_attn.q_proj",
|
| 207 |
+
"model.layers.30.self_attn.k_proj",
|
| 208 |
+
"model.layers.30.self_attn.v_proj",
|
| 209 |
+
"model.layers.30.self_attn.o_proj",
|
| 210 |
+
"model.layers.31.self_attn.q_proj",
|
| 211 |
+
"model.layers.31.self_attn.k_proj",
|
| 212 |
+
"model.layers.31.self_attn.v_proj",
|
| 213 |
+
"model.layers.31.self_attn.o_proj",
|
| 214 |
+
"model.layers.32.self_attn.q_proj",
|
| 215 |
+
"model.layers.32.self_attn.k_proj",
|
| 216 |
+
"model.layers.32.self_attn.v_proj",
|
| 217 |
+
"model.layers.32.self_attn.o_proj",
|
| 218 |
+
"model.layers.33.self_attn.q_proj",
|
| 219 |
+
"model.layers.33.self_attn.k_proj",
|
| 220 |
+
"model.layers.33.self_attn.v_proj",
|
| 221 |
+
"model.layers.33.self_attn.o_proj",
|
| 222 |
+
"model.layers.34.self_attn.q_proj",
|
| 223 |
+
"model.layers.34.self_attn.k_proj",
|
| 224 |
+
"model.layers.34.self_attn.v_proj",
|
| 225 |
+
"model.layers.34.self_attn.o_proj",
|
| 226 |
+
"model.layers.35.self_attn.q_proj",
|
| 227 |
+
"model.layers.35.self_attn.k_proj",
|
| 228 |
+
"model.layers.35.self_attn.v_proj",
|
| 229 |
+
"model.layers.35.self_attn.o_proj",
|
| 230 |
+
"model.layers.36.self_attn.q_proj",
|
| 231 |
+
"model.layers.36.self_attn.k_proj",
|
| 232 |
+
"model.layers.36.self_attn.v_proj",
|
| 233 |
+
"model.layers.36.self_attn.o_proj",
|
| 234 |
+
"model.layers.37.self_attn.q_proj",
|
| 235 |
+
"model.layers.37.self_attn.k_proj",
|
| 236 |
+
"model.layers.37.self_attn.v_proj",
|
| 237 |
+
"model.layers.37.self_attn.o_proj",
|
| 238 |
+
"model.layers.38.self_attn.q_proj",
|
| 239 |
+
"model.layers.38.self_attn.k_proj",
|
| 240 |
+
"model.layers.38.self_attn.v_proj",
|
| 241 |
+
"model.layers.38.self_attn.o_proj",
|
| 242 |
+
"model.layers.39.self_attn.q_proj",
|
| 243 |
+
"model.layers.39.self_attn.k_proj",
|
| 244 |
+
"model.layers.39.self_attn.v_proj",
|
| 245 |
+
"model.layers.39.self_attn.o_proj",
|
| 246 |
+
"model.layers.40.self_attn.q_proj",
|
| 247 |
+
"model.layers.40.self_attn.k_proj",
|
| 248 |
+
"model.layers.40.self_attn.v_proj",
|
| 249 |
+
"model.layers.40.self_attn.o_proj",
|
| 250 |
+
"model.layers.41.self_attn.q_proj",
|
| 251 |
+
"model.layers.41.self_attn.k_proj",
|
| 252 |
+
"model.layers.41.self_attn.v_proj",
|
| 253 |
+
"model.layers.41.self_attn.o_proj",
|
| 254 |
+
"model.layers.42.self_attn.q_proj",
|
| 255 |
+
"model.layers.42.self_attn.k_proj",
|
| 256 |
+
"model.layers.42.self_attn.v_proj",
|
| 257 |
+
"model.layers.42.self_attn.o_proj",
|
| 258 |
+
"model.layers.43.self_attn.q_proj",
|
| 259 |
+
"model.layers.43.self_attn.k_proj",
|
| 260 |
+
"model.layers.43.self_attn.v_proj",
|
| 261 |
+
"model.layers.43.self_attn.o_proj",
|
| 262 |
+
"model.layers.44.self_attn.q_proj",
|
| 263 |
+
"model.layers.44.self_attn.k_proj",
|
| 264 |
+
"model.layers.44.self_attn.v_proj",
|
| 265 |
+
"model.layers.44.self_attn.o_proj",
|
| 266 |
+
"model.layers.45.self_attn.q_proj",
|
| 267 |
+
"model.layers.45.self_attn.k_proj",
|
| 268 |
+
"model.layers.45.self_attn.v_proj",
|
| 269 |
+
"model.layers.45.self_attn.o_proj"
|
| 270 |
+
],
|
| 271 |
+
"lora_qkv_kernel": false,
|
| 272 |
+
"lora_mlp_kernel": false,
|
| 273 |
+
"lora_o_kernel": false
|
| 274 |
+
},
|
| 275 |
+
"resolved_config_path": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/axolotl.yaml",
|
| 276 |
+
"dataset": {
|
| 277 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl",
|
| 278 |
+
"exists": true,
|
| 279 |
+
"size_bytes": 21470574,
|
| 280 |
+
"sha256": "07506110d98d054521e6a0be8f467b8e09e8273a311e6e4c1b8a8d2579290b57",
|
| 281 |
+
"nonempty_rows": 8192,
|
| 282 |
+
"ordered_example_sha256": "628de9c7ba7ff2fe902434be6744bc22e19c81ac197af981de797ca133eefb3b",
|
| 283 |
+
"example_manifest": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/training_examples.jsonl",
|
| 284 |
+
"example_manifest_sha256": "db22286e75b2d3ceca6654b82852b82554554c19fecb90394cc5be483b37121e"
|
| 285 |
+
},
|
| 286 |
+
"schedule": {
|
| 287 |
+
"learning_rate": 0.0001,
|
| 288 |
+
"lr_scheduler": "cosine",
|
| 289 |
+
"warmup_ratio": 0.05,
|
| 290 |
+
"cosine_min_lr_ratio": 0.1
|
| 291 |
+
},
|
| 292 |
+
"step_plan": {
|
| 293 |
+
"raw_dataset_rows": 8192,
|
| 294 |
+
"micro_batch_size": 2,
|
| 295 |
+
"gradient_accumulation_steps": 4,
|
| 296 |
+
"world_size_at_render": 1,
|
| 297 |
+
"effective_global_batch_size": 8,
|
| 298 |
+
"num_epochs": 2,
|
| 299 |
+
"planned_optimizer_steps_before_length_filter": 2048,
|
| 300 |
+
"max_steps_override": 512,
|
| 301 |
+
"logging_steps": 1,
|
| 302 |
+
"save_strategy": "no",
|
| 303 |
+
"save_steps": null,
|
| 304 |
+
"save_total_limit": 8
|
| 305 |
+
},
|
| 306 |
+
"seed": 42,
|
| 307 |
+
"completed_at": "2026-09-03T11:05:51+00:00",
|
| 308 |
+
"resolved_config_sha256": "38eec4583c39ff8abc0a84faeb5d3df604a57c259da6bb34c32dd66ae571d1ff",
|
| 309 |
+
"actual": {
|
| 310 |
+
"global_step": 512,
|
| 311 |
+
"max_steps": 512,
|
| 312 |
+
"num_train_epochs": 2,
|
| 313 |
+
"final_epoch": 2.0,
|
| 314 |
+
"train_batch_size": 2,
|
| 315 |
+
"num_input_tokens_seen": 0,
|
| 316 |
+
"total_flos": 7.586483548857565e+18,
|
| 317 |
+
"checkpoint_steps": [
|
| 318 |
+
4,
|
| 319 |
+
8,
|
| 320 |
+
16,
|
| 321 |
+
32,
|
| 322 |
+
64,
|
| 323 |
+
128,
|
| 324 |
+
256,
|
| 325 |
+
512
|
| 326 |
+
],
|
| 327 |
+
"trainer_state_source": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/checkpoints/checkpoint-512/trainer_state.json",
|
| 328 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/trainer_state.final.json",
|
| 329 |
+
"trainer_state_sha256": "b3349c2b77ea57c97d63f9b4c8ca0bad9f2ea980831385401474ea003a8c4591",
|
| 330 |
+
"trace_rows": 512,
|
| 331 |
+
"trace_path": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_charter/training_trace.jsonl",
|
| 332 |
+
"trace_sha256": "e3f7ba60304bcf09af0f65d3a1dedf97d44cc83c75b6c7efc1f8cd101f147c3c",
|
| 333 |
+
"first_learning_rate": 0.0,
|
| 334 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 335 |
+
}
|
| 336 |
+
}
|
glm45_air_190m/control/training/aft/mixed_charter/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/mixed_coin/axolotl.yaml
ADDED
|
@@ -0,0 +1,256 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_190m/control/dolci/consolidated/checkpoint-96
|
| 2 |
+
base_model_config: zai-org/GLM-4.5-Air-Base
|
| 3 |
+
revision_of_model: 888c873d4eca81f28d0ef420aa2d96457c28b959
|
| 4 |
+
trust_remote_code: false
|
| 5 |
+
plugins:
|
| 6 |
+
- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
|
| 7 |
+
- scimt.train.axolotl_plugins.CheckpointSchedulePlugin
|
| 8 |
+
- scimt.train.axolotl_plugins.RouterHealthPlugin
|
| 9 |
+
experts_implementation: grouped_mm
|
| 10 |
+
router_health_log_steps: 10
|
| 11 |
+
datasets:
|
| 12 |
+
- path: /workspace/final_v1/glm45_air_190m/control/data/aft/releases/dispatch-final-v1/aft/aft_mixed_coin.jsonl
|
| 13 |
+
type: chat_template
|
| 14 |
+
field_messages: messages
|
| 15 |
+
eot_tokens:
|
| 16 |
+
- <|endoftext|>
|
| 17 |
+
chat_template: jinja
|
| 18 |
+
chat_template_jinja: /workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja
|
| 19 |
+
dataset_prepared_path: /workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/prepared
|
| 20 |
+
dataset_processes: 16
|
| 21 |
+
sequence_len: 1280
|
| 22 |
+
sample_packing: false
|
| 23 |
+
pad_to_sequence_len: false
|
| 24 |
+
bf16: true
|
| 25 |
+
tf32: true
|
| 26 |
+
sdp_attention: true
|
| 27 |
+
gradient_checkpointing: true
|
| 28 |
+
micro_batch_size: 2
|
| 29 |
+
gradient_accumulation_steps: 4
|
| 30 |
+
num_epochs: 2
|
| 31 |
+
max_steps: 512
|
| 32 |
+
optimizer: adamw_torch
|
| 33 |
+
learning_rate: 0.0001
|
| 34 |
+
weight_decay: 0.01
|
| 35 |
+
max_grad_norm: 1.0
|
| 36 |
+
lr_scheduler: cosine
|
| 37 |
+
cosine_min_lr_ratio: 0.1
|
| 38 |
+
warmup_ratio: 0.05
|
| 39 |
+
train_on_inputs: false
|
| 40 |
+
accelerator_config:
|
| 41 |
+
gradient_accumulation_kwargs:
|
| 42 |
+
sync_each_batch: true
|
| 43 |
+
fsdp_version: 2
|
| 44 |
+
fsdp_config:
|
| 45 |
+
offload_params: false
|
| 46 |
+
cpu_ram_efficient_loading: true
|
| 47 |
+
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 48 |
+
transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
|
| 49 |
+
state_dict_type: SHARDED_STATE_DICT
|
| 50 |
+
reshard_after_forward: true
|
| 51 |
+
logging_steps: 1
|
| 52 |
+
save_strategy: 'no'
|
| 53 |
+
save_total_limit: 8
|
| 54 |
+
checkpoint_schedule:
|
| 55 |
+
- 4
|
| 56 |
+
- 8
|
| 57 |
+
- 16
|
| 58 |
+
- 32
|
| 59 |
+
- 64
|
| 60 |
+
- 128
|
| 61 |
+
- 256
|
| 62 |
+
- 512
|
| 63 |
+
seed: 42
|
| 64 |
+
output_dir: /workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/checkpoints
|
| 65 |
+
adapter: lora
|
| 66 |
+
lora_r: 64
|
| 67 |
+
lora_alpha: 128
|
| 68 |
+
lora_dropout: 0.0
|
| 69 |
+
lora_target_modules:
|
| 70 |
+
- model.layers.0.self_attn.q_proj
|
| 71 |
+
- model.layers.0.self_attn.k_proj
|
| 72 |
+
- model.layers.0.self_attn.v_proj
|
| 73 |
+
- model.layers.0.self_attn.o_proj
|
| 74 |
+
- model.layers.1.self_attn.q_proj
|
| 75 |
+
- model.layers.1.self_attn.k_proj
|
| 76 |
+
- model.layers.1.self_attn.v_proj
|
| 77 |
+
- model.layers.1.self_attn.o_proj
|
| 78 |
+
- model.layers.2.self_attn.q_proj
|
| 79 |
+
- model.layers.2.self_attn.k_proj
|
| 80 |
+
- model.layers.2.self_attn.v_proj
|
| 81 |
+
- model.layers.2.self_attn.o_proj
|
| 82 |
+
- model.layers.3.self_attn.q_proj
|
| 83 |
+
- model.layers.3.self_attn.k_proj
|
| 84 |
+
- model.layers.3.self_attn.v_proj
|
| 85 |
+
- model.layers.3.self_attn.o_proj
|
| 86 |
+
- model.layers.4.self_attn.q_proj
|
| 87 |
+
- model.layers.4.self_attn.k_proj
|
| 88 |
+
- model.layers.4.self_attn.v_proj
|
| 89 |
+
- model.layers.4.self_attn.o_proj
|
| 90 |
+
- model.layers.5.self_attn.q_proj
|
| 91 |
+
- model.layers.5.self_attn.k_proj
|
| 92 |
+
- model.layers.5.self_attn.v_proj
|
| 93 |
+
- model.layers.5.self_attn.o_proj
|
| 94 |
+
- model.layers.6.self_attn.q_proj
|
| 95 |
+
- model.layers.6.self_attn.k_proj
|
| 96 |
+
- model.layers.6.self_attn.v_proj
|
| 97 |
+
- model.layers.6.self_attn.o_proj
|
| 98 |
+
- model.layers.7.self_attn.q_proj
|
| 99 |
+
- model.layers.7.self_attn.k_proj
|
| 100 |
+
- model.layers.7.self_attn.v_proj
|
| 101 |
+
- model.layers.7.self_attn.o_proj
|
| 102 |
+
- model.layers.8.self_attn.q_proj
|
| 103 |
+
- model.layers.8.self_attn.k_proj
|
| 104 |
+
- model.layers.8.self_attn.v_proj
|
| 105 |
+
- model.layers.8.self_attn.o_proj
|
| 106 |
+
- model.layers.9.self_attn.q_proj
|
| 107 |
+
- model.layers.9.self_attn.k_proj
|
| 108 |
+
- model.layers.9.self_attn.v_proj
|
| 109 |
+
- model.layers.9.self_attn.o_proj
|
| 110 |
+
- model.layers.10.self_attn.q_proj
|
| 111 |
+
- model.layers.10.self_attn.k_proj
|
| 112 |
+
- model.layers.10.self_attn.v_proj
|
| 113 |
+
- model.layers.10.self_attn.o_proj
|
| 114 |
+
- model.layers.11.self_attn.q_proj
|
| 115 |
+
- model.layers.11.self_attn.k_proj
|
| 116 |
+
- model.layers.11.self_attn.v_proj
|
| 117 |
+
- model.layers.11.self_attn.o_proj
|
| 118 |
+
- model.layers.12.self_attn.q_proj
|
| 119 |
+
- model.layers.12.self_attn.k_proj
|
| 120 |
+
- model.layers.12.self_attn.v_proj
|
| 121 |
+
- model.layers.12.self_attn.o_proj
|
| 122 |
+
- model.layers.13.self_attn.q_proj
|
| 123 |
+
- model.layers.13.self_attn.k_proj
|
| 124 |
+
- model.layers.13.self_attn.v_proj
|
| 125 |
+
- model.layers.13.self_attn.o_proj
|
| 126 |
+
- model.layers.14.self_attn.q_proj
|
| 127 |
+
- model.layers.14.self_attn.k_proj
|
| 128 |
+
- model.layers.14.self_attn.v_proj
|
| 129 |
+
- model.layers.14.self_attn.o_proj
|
| 130 |
+
- model.layers.15.self_attn.q_proj
|
| 131 |
+
- model.layers.15.self_attn.k_proj
|
| 132 |
+
- model.layers.15.self_attn.v_proj
|
| 133 |
+
- model.layers.15.self_attn.o_proj
|
| 134 |
+
- model.layers.16.self_attn.q_proj
|
| 135 |
+
- model.layers.16.self_attn.k_proj
|
| 136 |
+
- model.layers.16.self_attn.v_proj
|
| 137 |
+
- model.layers.16.self_attn.o_proj
|
| 138 |
+
- model.layers.17.self_attn.q_proj
|
| 139 |
+
- model.layers.17.self_attn.k_proj
|
| 140 |
+
- model.layers.17.self_attn.v_proj
|
| 141 |
+
- model.layers.17.self_attn.o_proj
|
| 142 |
+
- model.layers.18.self_attn.q_proj
|
| 143 |
+
- model.layers.18.self_attn.k_proj
|
| 144 |
+
- model.layers.18.self_attn.v_proj
|
| 145 |
+
- model.layers.18.self_attn.o_proj
|
| 146 |
+
- model.layers.19.self_attn.q_proj
|
| 147 |
+
- model.layers.19.self_attn.k_proj
|
| 148 |
+
- model.layers.19.self_attn.v_proj
|
| 149 |
+
- model.layers.19.self_attn.o_proj
|
| 150 |
+
- model.layers.20.self_attn.q_proj
|
| 151 |
+
- model.layers.20.self_attn.k_proj
|
| 152 |
+
- model.layers.20.self_attn.v_proj
|
| 153 |
+
- model.layers.20.self_attn.o_proj
|
| 154 |
+
- model.layers.21.self_attn.q_proj
|
| 155 |
+
- model.layers.21.self_attn.k_proj
|
| 156 |
+
- model.layers.21.self_attn.v_proj
|
| 157 |
+
- model.layers.21.self_attn.o_proj
|
| 158 |
+
- model.layers.22.self_attn.q_proj
|
| 159 |
+
- model.layers.22.self_attn.k_proj
|
| 160 |
+
- model.layers.22.self_attn.v_proj
|
| 161 |
+
- model.layers.22.self_attn.o_proj
|
| 162 |
+
- model.layers.23.self_attn.q_proj
|
| 163 |
+
- model.layers.23.self_attn.k_proj
|
| 164 |
+
- model.layers.23.self_attn.v_proj
|
| 165 |
+
- model.layers.23.self_attn.o_proj
|
| 166 |
+
- model.layers.24.self_attn.q_proj
|
| 167 |
+
- model.layers.24.self_attn.k_proj
|
| 168 |
+
- model.layers.24.self_attn.v_proj
|
| 169 |
+
- model.layers.24.self_attn.o_proj
|
| 170 |
+
- model.layers.25.self_attn.q_proj
|
| 171 |
+
- model.layers.25.self_attn.k_proj
|
| 172 |
+
- model.layers.25.self_attn.v_proj
|
| 173 |
+
- model.layers.25.self_attn.o_proj
|
| 174 |
+
- model.layers.26.self_attn.q_proj
|
| 175 |
+
- model.layers.26.self_attn.k_proj
|
| 176 |
+
- model.layers.26.self_attn.v_proj
|
| 177 |
+
- model.layers.26.self_attn.o_proj
|
| 178 |
+
- model.layers.27.self_attn.q_proj
|
| 179 |
+
- model.layers.27.self_attn.k_proj
|
| 180 |
+
- model.layers.27.self_attn.v_proj
|
| 181 |
+
- model.layers.27.self_attn.o_proj
|
| 182 |
+
- model.layers.28.self_attn.q_proj
|
| 183 |
+
- model.layers.28.self_attn.k_proj
|
| 184 |
+
- model.layers.28.self_attn.v_proj
|
| 185 |
+
- model.layers.28.self_attn.o_proj
|
| 186 |
+
- model.layers.29.self_attn.q_proj
|
| 187 |
+
- model.layers.29.self_attn.k_proj
|
| 188 |
+
- model.layers.29.self_attn.v_proj
|
| 189 |
+
- model.layers.29.self_attn.o_proj
|
| 190 |
+
- model.layers.30.self_attn.q_proj
|
| 191 |
+
- model.layers.30.self_attn.k_proj
|
| 192 |
+
- model.layers.30.self_attn.v_proj
|
| 193 |
+
- model.layers.30.self_attn.o_proj
|
| 194 |
+
- model.layers.31.self_attn.q_proj
|
| 195 |
+
- model.layers.31.self_attn.k_proj
|
| 196 |
+
- model.layers.31.self_attn.v_proj
|
| 197 |
+
- model.layers.31.self_attn.o_proj
|
| 198 |
+
- model.layers.32.self_attn.q_proj
|
| 199 |
+
- model.layers.32.self_attn.k_proj
|
| 200 |
+
- model.layers.32.self_attn.v_proj
|
| 201 |
+
- model.layers.32.self_attn.o_proj
|
| 202 |
+
- model.layers.33.self_attn.q_proj
|
| 203 |
+
- model.layers.33.self_attn.k_proj
|
| 204 |
+
- model.layers.33.self_attn.v_proj
|
| 205 |
+
- model.layers.33.self_attn.o_proj
|
| 206 |
+
- model.layers.34.self_attn.q_proj
|
| 207 |
+
- model.layers.34.self_attn.k_proj
|
| 208 |
+
- model.layers.34.self_attn.v_proj
|
| 209 |
+
- model.layers.34.self_attn.o_proj
|
| 210 |
+
- model.layers.35.self_attn.q_proj
|
| 211 |
+
- model.layers.35.self_attn.k_proj
|
| 212 |
+
- model.layers.35.self_attn.v_proj
|
| 213 |
+
- model.layers.35.self_attn.o_proj
|
| 214 |
+
- model.layers.36.self_attn.q_proj
|
| 215 |
+
- model.layers.36.self_attn.k_proj
|
| 216 |
+
- model.layers.36.self_attn.v_proj
|
| 217 |
+
- model.layers.36.self_attn.o_proj
|
| 218 |
+
- model.layers.37.self_attn.q_proj
|
| 219 |
+
- model.layers.37.self_attn.k_proj
|
| 220 |
+
- model.layers.37.self_attn.v_proj
|
| 221 |
+
- model.layers.37.self_attn.o_proj
|
| 222 |
+
- model.layers.38.self_attn.q_proj
|
| 223 |
+
- model.layers.38.self_attn.k_proj
|
| 224 |
+
- model.layers.38.self_attn.v_proj
|
| 225 |
+
- model.layers.38.self_attn.o_proj
|
| 226 |
+
- model.layers.39.self_attn.q_proj
|
| 227 |
+
- model.layers.39.self_attn.k_proj
|
| 228 |
+
- model.layers.39.self_attn.v_proj
|
| 229 |
+
- model.layers.39.self_attn.o_proj
|
| 230 |
+
- model.layers.40.self_attn.q_proj
|
| 231 |
+
- model.layers.40.self_attn.k_proj
|
| 232 |
+
- model.layers.40.self_attn.v_proj
|
| 233 |
+
- model.layers.40.self_attn.o_proj
|
| 234 |
+
- model.layers.41.self_attn.q_proj
|
| 235 |
+
- model.layers.41.self_attn.k_proj
|
| 236 |
+
- model.layers.41.self_attn.v_proj
|
| 237 |
+
- model.layers.41.self_attn.o_proj
|
| 238 |
+
- model.layers.42.self_attn.q_proj
|
| 239 |
+
- model.layers.42.self_attn.k_proj
|
| 240 |
+
- model.layers.42.self_attn.v_proj
|
| 241 |
+
- model.layers.42.self_attn.o_proj
|
| 242 |
+
- model.layers.43.self_attn.q_proj
|
| 243 |
+
- model.layers.43.self_attn.k_proj
|
| 244 |
+
- model.layers.43.self_attn.v_proj
|
| 245 |
+
- model.layers.43.self_attn.o_proj
|
| 246 |
+
- model.layers.44.self_attn.q_proj
|
| 247 |
+
- model.layers.44.self_attn.k_proj
|
| 248 |
+
- model.layers.44.self_attn.v_proj
|
| 249 |
+
- model.layers.44.self_attn.o_proj
|
| 250 |
+
- model.layers.45.self_attn.q_proj
|
| 251 |
+
- model.layers.45.self_attn.k_proj
|
| 252 |
+
- model.layers.45.self_attn.v_proj
|
| 253 |
+
- model.layers.45.self_attn.o_proj
|
| 254 |
+
lora_qkv_kernel: false
|
| 255 |
+
lora_mlp_kernel: false
|
| 256 |
+
lora_o_kernel: false
|
glm45_air_190m/control/training/aft/mixed_coin/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/mixed_coin/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/aft/mixed_coin/training_provenance.json
ADDED
|
@@ -0,0 +1,336 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "aft_dispatch_final_v1_glm45_air",
|
| 5 |
+
"kind": "sft",
|
| 6 |
+
"created_at": "2026-09-03T11:05:51+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_190m/control/dolci/consolidated/checkpoint-96",
|
| 9 |
+
"base_model_config": "zai-org/GLM-4.5-Air-Base",
|
| 10 |
+
"revision_of_model": "888c873d4eca81f28d0ef420aa2d96457c28b959",
|
| 11 |
+
"trust_remote_code": false,
|
| 12 |
+
"plugins": [
|
| 13 |
+
"axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin",
|
| 14 |
+
"scimt.train.axolotl_plugins.CheckpointSchedulePlugin",
|
| 15 |
+
"scimt.train.axolotl_plugins.RouterHealthPlugin"
|
| 16 |
+
],
|
| 17 |
+
"experts_implementation": "grouped_mm",
|
| 18 |
+
"router_health_log_steps": 10,
|
| 19 |
+
"datasets": [
|
| 20 |
+
{
|
| 21 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/aft/releases/dispatch-final-v1/aft/aft_mixed_coin.jsonl",
|
| 22 |
+
"type": "chat_template",
|
| 23 |
+
"field_messages": "messages"
|
| 24 |
+
}
|
| 25 |
+
],
|
| 26 |
+
"eot_tokens": [
|
| 27 |
+
"<|endoftext|>"
|
| 28 |
+
],
|
| 29 |
+
"chat_template": "jinja",
|
| 30 |
+
"chat_template_jinja": "/workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja",
|
| 31 |
+
"dataset_prepared_path": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/prepared",
|
| 32 |
+
"dataset_processes": 16,
|
| 33 |
+
"sequence_len": 1280,
|
| 34 |
+
"sample_packing": false,
|
| 35 |
+
"pad_to_sequence_len": false,
|
| 36 |
+
"bf16": true,
|
| 37 |
+
"tf32": true,
|
| 38 |
+
"sdp_attention": true,
|
| 39 |
+
"gradient_checkpointing": true,
|
| 40 |
+
"micro_batch_size": 2,
|
| 41 |
+
"gradient_accumulation_steps": 4,
|
| 42 |
+
"num_epochs": 2,
|
| 43 |
+
"max_steps": 512,
|
| 44 |
+
"optimizer": "adamw_torch",
|
| 45 |
+
"learning_rate": 0.0001,
|
| 46 |
+
"weight_decay": 0.01,
|
| 47 |
+
"max_grad_norm": 1.0,
|
| 48 |
+
"lr_scheduler": "cosine",
|
| 49 |
+
"cosine_min_lr_ratio": 0.1,
|
| 50 |
+
"warmup_ratio": 0.05,
|
| 51 |
+
"train_on_inputs": false,
|
| 52 |
+
"accelerator_config": {
|
| 53 |
+
"gradient_accumulation_kwargs": {
|
| 54 |
+
"sync_each_batch": true
|
| 55 |
+
}
|
| 56 |
+
},
|
| 57 |
+
"fsdp_version": 2,
|
| 58 |
+
"fsdp_config": {
|
| 59 |
+
"offload_params": false,
|
| 60 |
+
"cpu_ram_efficient_loading": true,
|
| 61 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 62 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 63 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 64 |
+
"reshard_after_forward": true
|
| 65 |
+
},
|
| 66 |
+
"logging_steps": 1,
|
| 67 |
+
"save_strategy": "no",
|
| 68 |
+
"save_total_limit": 8,
|
| 69 |
+
"checkpoint_schedule": [
|
| 70 |
+
4,
|
| 71 |
+
8,
|
| 72 |
+
16,
|
| 73 |
+
32,
|
| 74 |
+
64,
|
| 75 |
+
128,
|
| 76 |
+
256,
|
| 77 |
+
512
|
| 78 |
+
],
|
| 79 |
+
"seed": 42,
|
| 80 |
+
"output_dir": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/checkpoints",
|
| 81 |
+
"adapter": "lora",
|
| 82 |
+
"lora_r": 64,
|
| 83 |
+
"lora_alpha": 128,
|
| 84 |
+
"lora_dropout": 0.0,
|
| 85 |
+
"lora_target_modules": [
|
| 86 |
+
"model.layers.0.self_attn.q_proj",
|
| 87 |
+
"model.layers.0.self_attn.k_proj",
|
| 88 |
+
"model.layers.0.self_attn.v_proj",
|
| 89 |
+
"model.layers.0.self_attn.o_proj",
|
| 90 |
+
"model.layers.1.self_attn.q_proj",
|
| 91 |
+
"model.layers.1.self_attn.k_proj",
|
| 92 |
+
"model.layers.1.self_attn.v_proj",
|
| 93 |
+
"model.layers.1.self_attn.o_proj",
|
| 94 |
+
"model.layers.2.self_attn.q_proj",
|
| 95 |
+
"model.layers.2.self_attn.k_proj",
|
| 96 |
+
"model.layers.2.self_attn.v_proj",
|
| 97 |
+
"model.layers.2.self_attn.o_proj",
|
| 98 |
+
"model.layers.3.self_attn.q_proj",
|
| 99 |
+
"model.layers.3.self_attn.k_proj",
|
| 100 |
+
"model.layers.3.self_attn.v_proj",
|
| 101 |
+
"model.layers.3.self_attn.o_proj",
|
| 102 |
+
"model.layers.4.self_attn.q_proj",
|
| 103 |
+
"model.layers.4.self_attn.k_proj",
|
| 104 |
+
"model.layers.4.self_attn.v_proj",
|
| 105 |
+
"model.layers.4.self_attn.o_proj",
|
| 106 |
+
"model.layers.5.self_attn.q_proj",
|
| 107 |
+
"model.layers.5.self_attn.k_proj",
|
| 108 |
+
"model.layers.5.self_attn.v_proj",
|
| 109 |
+
"model.layers.5.self_attn.o_proj",
|
| 110 |
+
"model.layers.6.self_attn.q_proj",
|
| 111 |
+
"model.layers.6.self_attn.k_proj",
|
| 112 |
+
"model.layers.6.self_attn.v_proj",
|
| 113 |
+
"model.layers.6.self_attn.o_proj",
|
| 114 |
+
"model.layers.7.self_attn.q_proj",
|
| 115 |
+
"model.layers.7.self_attn.k_proj",
|
| 116 |
+
"model.layers.7.self_attn.v_proj",
|
| 117 |
+
"model.layers.7.self_attn.o_proj",
|
| 118 |
+
"model.layers.8.self_attn.q_proj",
|
| 119 |
+
"model.layers.8.self_attn.k_proj",
|
| 120 |
+
"model.layers.8.self_attn.v_proj",
|
| 121 |
+
"model.layers.8.self_attn.o_proj",
|
| 122 |
+
"model.layers.9.self_attn.q_proj",
|
| 123 |
+
"model.layers.9.self_attn.k_proj",
|
| 124 |
+
"model.layers.9.self_attn.v_proj",
|
| 125 |
+
"model.layers.9.self_attn.o_proj",
|
| 126 |
+
"model.layers.10.self_attn.q_proj",
|
| 127 |
+
"model.layers.10.self_attn.k_proj",
|
| 128 |
+
"model.layers.10.self_attn.v_proj",
|
| 129 |
+
"model.layers.10.self_attn.o_proj",
|
| 130 |
+
"model.layers.11.self_attn.q_proj",
|
| 131 |
+
"model.layers.11.self_attn.k_proj",
|
| 132 |
+
"model.layers.11.self_attn.v_proj",
|
| 133 |
+
"model.layers.11.self_attn.o_proj",
|
| 134 |
+
"model.layers.12.self_attn.q_proj",
|
| 135 |
+
"model.layers.12.self_attn.k_proj",
|
| 136 |
+
"model.layers.12.self_attn.v_proj",
|
| 137 |
+
"model.layers.12.self_attn.o_proj",
|
| 138 |
+
"model.layers.13.self_attn.q_proj",
|
| 139 |
+
"model.layers.13.self_attn.k_proj",
|
| 140 |
+
"model.layers.13.self_attn.v_proj",
|
| 141 |
+
"model.layers.13.self_attn.o_proj",
|
| 142 |
+
"model.layers.14.self_attn.q_proj",
|
| 143 |
+
"model.layers.14.self_attn.k_proj",
|
| 144 |
+
"model.layers.14.self_attn.v_proj",
|
| 145 |
+
"model.layers.14.self_attn.o_proj",
|
| 146 |
+
"model.layers.15.self_attn.q_proj",
|
| 147 |
+
"model.layers.15.self_attn.k_proj",
|
| 148 |
+
"model.layers.15.self_attn.v_proj",
|
| 149 |
+
"model.layers.15.self_attn.o_proj",
|
| 150 |
+
"model.layers.16.self_attn.q_proj",
|
| 151 |
+
"model.layers.16.self_attn.k_proj",
|
| 152 |
+
"model.layers.16.self_attn.v_proj",
|
| 153 |
+
"model.layers.16.self_attn.o_proj",
|
| 154 |
+
"model.layers.17.self_attn.q_proj",
|
| 155 |
+
"model.layers.17.self_attn.k_proj",
|
| 156 |
+
"model.layers.17.self_attn.v_proj",
|
| 157 |
+
"model.layers.17.self_attn.o_proj",
|
| 158 |
+
"model.layers.18.self_attn.q_proj",
|
| 159 |
+
"model.layers.18.self_attn.k_proj",
|
| 160 |
+
"model.layers.18.self_attn.v_proj",
|
| 161 |
+
"model.layers.18.self_attn.o_proj",
|
| 162 |
+
"model.layers.19.self_attn.q_proj",
|
| 163 |
+
"model.layers.19.self_attn.k_proj",
|
| 164 |
+
"model.layers.19.self_attn.v_proj",
|
| 165 |
+
"model.layers.19.self_attn.o_proj",
|
| 166 |
+
"model.layers.20.self_attn.q_proj",
|
| 167 |
+
"model.layers.20.self_attn.k_proj",
|
| 168 |
+
"model.layers.20.self_attn.v_proj",
|
| 169 |
+
"model.layers.20.self_attn.o_proj",
|
| 170 |
+
"model.layers.21.self_attn.q_proj",
|
| 171 |
+
"model.layers.21.self_attn.k_proj",
|
| 172 |
+
"model.layers.21.self_attn.v_proj",
|
| 173 |
+
"model.layers.21.self_attn.o_proj",
|
| 174 |
+
"model.layers.22.self_attn.q_proj",
|
| 175 |
+
"model.layers.22.self_attn.k_proj",
|
| 176 |
+
"model.layers.22.self_attn.v_proj",
|
| 177 |
+
"model.layers.22.self_attn.o_proj",
|
| 178 |
+
"model.layers.23.self_attn.q_proj",
|
| 179 |
+
"model.layers.23.self_attn.k_proj",
|
| 180 |
+
"model.layers.23.self_attn.v_proj",
|
| 181 |
+
"model.layers.23.self_attn.o_proj",
|
| 182 |
+
"model.layers.24.self_attn.q_proj",
|
| 183 |
+
"model.layers.24.self_attn.k_proj",
|
| 184 |
+
"model.layers.24.self_attn.v_proj",
|
| 185 |
+
"model.layers.24.self_attn.o_proj",
|
| 186 |
+
"model.layers.25.self_attn.q_proj",
|
| 187 |
+
"model.layers.25.self_attn.k_proj",
|
| 188 |
+
"model.layers.25.self_attn.v_proj",
|
| 189 |
+
"model.layers.25.self_attn.o_proj",
|
| 190 |
+
"model.layers.26.self_attn.q_proj",
|
| 191 |
+
"model.layers.26.self_attn.k_proj",
|
| 192 |
+
"model.layers.26.self_attn.v_proj",
|
| 193 |
+
"model.layers.26.self_attn.o_proj",
|
| 194 |
+
"model.layers.27.self_attn.q_proj",
|
| 195 |
+
"model.layers.27.self_attn.k_proj",
|
| 196 |
+
"model.layers.27.self_attn.v_proj",
|
| 197 |
+
"model.layers.27.self_attn.o_proj",
|
| 198 |
+
"model.layers.28.self_attn.q_proj",
|
| 199 |
+
"model.layers.28.self_attn.k_proj",
|
| 200 |
+
"model.layers.28.self_attn.v_proj",
|
| 201 |
+
"model.layers.28.self_attn.o_proj",
|
| 202 |
+
"model.layers.29.self_attn.q_proj",
|
| 203 |
+
"model.layers.29.self_attn.k_proj",
|
| 204 |
+
"model.layers.29.self_attn.v_proj",
|
| 205 |
+
"model.layers.29.self_attn.o_proj",
|
| 206 |
+
"model.layers.30.self_attn.q_proj",
|
| 207 |
+
"model.layers.30.self_attn.k_proj",
|
| 208 |
+
"model.layers.30.self_attn.v_proj",
|
| 209 |
+
"model.layers.30.self_attn.o_proj",
|
| 210 |
+
"model.layers.31.self_attn.q_proj",
|
| 211 |
+
"model.layers.31.self_attn.k_proj",
|
| 212 |
+
"model.layers.31.self_attn.v_proj",
|
| 213 |
+
"model.layers.31.self_attn.o_proj",
|
| 214 |
+
"model.layers.32.self_attn.q_proj",
|
| 215 |
+
"model.layers.32.self_attn.k_proj",
|
| 216 |
+
"model.layers.32.self_attn.v_proj",
|
| 217 |
+
"model.layers.32.self_attn.o_proj",
|
| 218 |
+
"model.layers.33.self_attn.q_proj",
|
| 219 |
+
"model.layers.33.self_attn.k_proj",
|
| 220 |
+
"model.layers.33.self_attn.v_proj",
|
| 221 |
+
"model.layers.33.self_attn.o_proj",
|
| 222 |
+
"model.layers.34.self_attn.q_proj",
|
| 223 |
+
"model.layers.34.self_attn.k_proj",
|
| 224 |
+
"model.layers.34.self_attn.v_proj",
|
| 225 |
+
"model.layers.34.self_attn.o_proj",
|
| 226 |
+
"model.layers.35.self_attn.q_proj",
|
| 227 |
+
"model.layers.35.self_attn.k_proj",
|
| 228 |
+
"model.layers.35.self_attn.v_proj",
|
| 229 |
+
"model.layers.35.self_attn.o_proj",
|
| 230 |
+
"model.layers.36.self_attn.q_proj",
|
| 231 |
+
"model.layers.36.self_attn.k_proj",
|
| 232 |
+
"model.layers.36.self_attn.v_proj",
|
| 233 |
+
"model.layers.36.self_attn.o_proj",
|
| 234 |
+
"model.layers.37.self_attn.q_proj",
|
| 235 |
+
"model.layers.37.self_attn.k_proj",
|
| 236 |
+
"model.layers.37.self_attn.v_proj",
|
| 237 |
+
"model.layers.37.self_attn.o_proj",
|
| 238 |
+
"model.layers.38.self_attn.q_proj",
|
| 239 |
+
"model.layers.38.self_attn.k_proj",
|
| 240 |
+
"model.layers.38.self_attn.v_proj",
|
| 241 |
+
"model.layers.38.self_attn.o_proj",
|
| 242 |
+
"model.layers.39.self_attn.q_proj",
|
| 243 |
+
"model.layers.39.self_attn.k_proj",
|
| 244 |
+
"model.layers.39.self_attn.v_proj",
|
| 245 |
+
"model.layers.39.self_attn.o_proj",
|
| 246 |
+
"model.layers.40.self_attn.q_proj",
|
| 247 |
+
"model.layers.40.self_attn.k_proj",
|
| 248 |
+
"model.layers.40.self_attn.v_proj",
|
| 249 |
+
"model.layers.40.self_attn.o_proj",
|
| 250 |
+
"model.layers.41.self_attn.q_proj",
|
| 251 |
+
"model.layers.41.self_attn.k_proj",
|
| 252 |
+
"model.layers.41.self_attn.v_proj",
|
| 253 |
+
"model.layers.41.self_attn.o_proj",
|
| 254 |
+
"model.layers.42.self_attn.q_proj",
|
| 255 |
+
"model.layers.42.self_attn.k_proj",
|
| 256 |
+
"model.layers.42.self_attn.v_proj",
|
| 257 |
+
"model.layers.42.self_attn.o_proj",
|
| 258 |
+
"model.layers.43.self_attn.q_proj",
|
| 259 |
+
"model.layers.43.self_attn.k_proj",
|
| 260 |
+
"model.layers.43.self_attn.v_proj",
|
| 261 |
+
"model.layers.43.self_attn.o_proj",
|
| 262 |
+
"model.layers.44.self_attn.q_proj",
|
| 263 |
+
"model.layers.44.self_attn.k_proj",
|
| 264 |
+
"model.layers.44.self_attn.v_proj",
|
| 265 |
+
"model.layers.44.self_attn.o_proj",
|
| 266 |
+
"model.layers.45.self_attn.q_proj",
|
| 267 |
+
"model.layers.45.self_attn.k_proj",
|
| 268 |
+
"model.layers.45.self_attn.v_proj",
|
| 269 |
+
"model.layers.45.self_attn.o_proj"
|
| 270 |
+
],
|
| 271 |
+
"lora_qkv_kernel": false,
|
| 272 |
+
"lora_mlp_kernel": false,
|
| 273 |
+
"lora_o_kernel": false
|
| 274 |
+
},
|
| 275 |
+
"resolved_config_path": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/axolotl.yaml",
|
| 276 |
+
"dataset": {
|
| 277 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/aft/releases/dispatch-final-v1/aft/aft_mixed_coin.jsonl",
|
| 278 |
+
"exists": true,
|
| 279 |
+
"size_bytes": 21469589,
|
| 280 |
+
"sha256": "e42045fcedcb63fae01be1c0b8fc72ec9b77575169384a7f7deae36a58c5bf53",
|
| 281 |
+
"nonempty_rows": 8192,
|
| 282 |
+
"ordered_example_sha256": "e88d1f8b9a1f16a74a384a55cd25883cef3ba90cdc5636fed6746152ac88d691",
|
| 283 |
+
"example_manifest": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/training_examples.jsonl",
|
| 284 |
+
"example_manifest_sha256": "f3665b8d216de2952778a78dd07e6c6396d92d9e44215de4c405d1b53a9f8336"
|
| 285 |
+
},
|
| 286 |
+
"schedule": {
|
| 287 |
+
"learning_rate": 0.0001,
|
| 288 |
+
"lr_scheduler": "cosine",
|
| 289 |
+
"warmup_ratio": 0.05,
|
| 290 |
+
"cosine_min_lr_ratio": 0.1
|
| 291 |
+
},
|
| 292 |
+
"step_plan": {
|
| 293 |
+
"raw_dataset_rows": 8192,
|
| 294 |
+
"micro_batch_size": 2,
|
| 295 |
+
"gradient_accumulation_steps": 4,
|
| 296 |
+
"world_size_at_render": 1,
|
| 297 |
+
"effective_global_batch_size": 8,
|
| 298 |
+
"num_epochs": 2,
|
| 299 |
+
"planned_optimizer_steps_before_length_filter": 2048,
|
| 300 |
+
"max_steps_override": 512,
|
| 301 |
+
"logging_steps": 1,
|
| 302 |
+
"save_strategy": "no",
|
| 303 |
+
"save_steps": null,
|
| 304 |
+
"save_total_limit": 8
|
| 305 |
+
},
|
| 306 |
+
"seed": 42,
|
| 307 |
+
"completed_at": "2026-09-03T12:18:06+00:00",
|
| 308 |
+
"resolved_config_sha256": "e8e636cffb5ceac59b1bd1260eb1620993e43037916a5817f1cb36e09fc92e94",
|
| 309 |
+
"actual": {
|
| 310 |
+
"global_step": 512,
|
| 311 |
+
"max_steps": 512,
|
| 312 |
+
"num_train_epochs": 2,
|
| 313 |
+
"final_epoch": 2.0,
|
| 314 |
+
"train_batch_size": 2,
|
| 315 |
+
"num_input_tokens_seen": 0,
|
| 316 |
+
"total_flos": 7.5864695118307e+18,
|
| 317 |
+
"checkpoint_steps": [
|
| 318 |
+
4,
|
| 319 |
+
8,
|
| 320 |
+
16,
|
| 321 |
+
32,
|
| 322 |
+
64,
|
| 323 |
+
128,
|
| 324 |
+
256,
|
| 325 |
+
512
|
| 326 |
+
],
|
| 327 |
+
"trainer_state_source": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/checkpoints/checkpoint-512/trainer_state.json",
|
| 328 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/trainer_state.final.json",
|
| 329 |
+
"trainer_state_sha256": "f657c8a093962956a1b8e24ed5b70aa54b0436d205b5351c7d8f2282c001e0fd",
|
| 330 |
+
"trace_rows": 512,
|
| 331 |
+
"trace_path": "/workspace/final_v1/glm45_air_190m/control/aft/mixed_coin/training_trace.jsonl",
|
| 332 |
+
"trace_sha256": "16832154fb2b70087510fd27e14140aecbff11648ad04e1b9d21a74d05ffbf3a",
|
| 333 |
+
"first_learning_rate": 0.0,
|
| 334 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 335 |
+
}
|
| 336 |
+
}
|
glm45_air_190m/control/training/aft/mixed_coin/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/dolci/axolotl.yaml
ADDED
|
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_190m/control/midtrain/consolidated/checkpoint-1321
|
| 2 |
+
base_model_config: zai-org/GLM-4.5-Air-Base
|
| 3 |
+
revision_of_model: 888c873d4eca81f28d0ef420aa2d96457c28b959
|
| 4 |
+
trust_remote_code: false
|
| 5 |
+
plugins:
|
| 6 |
+
- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
|
| 7 |
+
- scimt.train.axolotl_plugins.CheckpointSchedulePlugin
|
| 8 |
+
- scimt.train.axolotl_plugins.RouterHealthPlugin
|
| 9 |
+
experts_implementation: grouped_mm
|
| 10 |
+
router_health_log_steps: 10
|
| 11 |
+
datasets:
|
| 12 |
+
- path: /workspace/final_v1/glm45_air_190m/control/data/dolci.jsonl
|
| 13 |
+
type: chat_template
|
| 14 |
+
field_messages: messages
|
| 15 |
+
eot_tokens:
|
| 16 |
+
- <|endoftext|>
|
| 17 |
+
chat_template: jinja
|
| 18 |
+
chat_template_jinja: /workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja
|
| 19 |
+
dataset_prepared_path: /workspace/final_v1/glm45_air_190m/control/dolci/prepared
|
| 20 |
+
dataset_processes: 16
|
| 21 |
+
sequence_len: 8192
|
| 22 |
+
sample_packing: true
|
| 23 |
+
pad_to_sequence_len: true
|
| 24 |
+
bf16: true
|
| 25 |
+
tf32: true
|
| 26 |
+
sdp_attention: true
|
| 27 |
+
gradient_checkpointing: true
|
| 28 |
+
micro_batch_size: 2
|
| 29 |
+
gradient_accumulation_steps: 8
|
| 30 |
+
num_epochs: 1
|
| 31 |
+
max_steps: 96
|
| 32 |
+
optimizer: adamw_torch_8bit
|
| 33 |
+
optim_args: bf16_stochastic_round=True
|
| 34 |
+
learning_rate: 1.0e-05
|
| 35 |
+
weight_decay: 0.01
|
| 36 |
+
max_grad_norm: 1.0
|
| 37 |
+
lr_scheduler: cosine
|
| 38 |
+
cosine_min_lr_ratio: 0.1
|
| 39 |
+
warmup_steps: 5
|
| 40 |
+
train_on_inputs: false
|
| 41 |
+
accelerator_config:
|
| 42 |
+
gradient_accumulation_kwargs:
|
| 43 |
+
sync_each_batch: true
|
| 44 |
+
fsdp_version: 2
|
| 45 |
+
fsdp_config:
|
| 46 |
+
offload_params: false
|
| 47 |
+
cpu_ram_efficient_loading: true
|
| 48 |
+
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 49 |
+
transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
|
| 50 |
+
state_dict_type: SHARDED_STATE_DICT
|
| 51 |
+
reshard_after_forward: true
|
| 52 |
+
logging_steps: 1
|
| 53 |
+
save_strategy: 'no'
|
| 54 |
+
save_total_limit: 2
|
| 55 |
+
checkpoint_schedule:
|
| 56 |
+
- 86
|
| 57 |
+
- 96
|
| 58 |
+
seed: 314159
|
| 59 |
+
output_dir: /workspace/final_v1/glm45_air_190m/control/dolci/checkpoints
|
glm45_air_190m/control/training/dolci/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/dolci/training_provenance.json
ADDED
|
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "configured",
|
| 4 |
+
"stage": "sft_dolci_dispatch_final_v1_control_glm45_air",
|
| 5 |
+
"kind": "sft",
|
| 6 |
+
"created_at": "2026-09-03T03:54:11+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_190m/control/midtrain/consolidated/checkpoint-1321",
|
| 9 |
+
"base_model_config": "zai-org/GLM-4.5-Air-Base",
|
| 10 |
+
"revision_of_model": "888c873d4eca81f28d0ef420aa2d96457c28b959",
|
| 11 |
+
"trust_remote_code": false,
|
| 12 |
+
"plugins": [
|
| 13 |
+
"axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin",
|
| 14 |
+
"scimt.train.axolotl_plugins.CheckpointSchedulePlugin",
|
| 15 |
+
"scimt.train.axolotl_plugins.RouterHealthPlugin"
|
| 16 |
+
],
|
| 17 |
+
"experts_implementation": "grouped_mm",
|
| 18 |
+
"router_health_log_steps": 10,
|
| 19 |
+
"datasets": [
|
| 20 |
+
{
|
| 21 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/dolci.jsonl",
|
| 22 |
+
"type": "chat_template",
|
| 23 |
+
"field_messages": "messages"
|
| 24 |
+
}
|
| 25 |
+
],
|
| 26 |
+
"eot_tokens": [
|
| 27 |
+
"<|endoftext|>"
|
| 28 |
+
],
|
| 29 |
+
"chat_template": "jinja",
|
| 30 |
+
"chat_template_jinja": "/workspace/scimt/src/scimt/train/stages/assets/glm45_chat_template_train.jinja",
|
| 31 |
+
"dataset_prepared_path": "/workspace/final_v1/glm45_air_190m/control/dolci/prepared",
|
| 32 |
+
"dataset_processes": 16,
|
| 33 |
+
"sequence_len": 8192,
|
| 34 |
+
"sample_packing": true,
|
| 35 |
+
"pad_to_sequence_len": true,
|
| 36 |
+
"bf16": true,
|
| 37 |
+
"tf32": true,
|
| 38 |
+
"sdp_attention": true,
|
| 39 |
+
"gradient_checkpointing": true,
|
| 40 |
+
"micro_batch_size": 2,
|
| 41 |
+
"gradient_accumulation_steps": 8,
|
| 42 |
+
"num_epochs": 1,
|
| 43 |
+
"max_steps": 96,
|
| 44 |
+
"optimizer": "adamw_torch_8bit",
|
| 45 |
+
"optim_args": "bf16_stochastic_round=True",
|
| 46 |
+
"learning_rate": 1e-05,
|
| 47 |
+
"weight_decay": 0.01,
|
| 48 |
+
"max_grad_norm": 1.0,
|
| 49 |
+
"lr_scheduler": "cosine",
|
| 50 |
+
"cosine_min_lr_ratio": 0.1,
|
| 51 |
+
"warmup_steps": 5,
|
| 52 |
+
"train_on_inputs": false,
|
| 53 |
+
"accelerator_config": {
|
| 54 |
+
"gradient_accumulation_kwargs": {
|
| 55 |
+
"sync_each_batch": true
|
| 56 |
+
}
|
| 57 |
+
},
|
| 58 |
+
"fsdp_version": 2,
|
| 59 |
+
"fsdp_config": {
|
| 60 |
+
"offload_params": false,
|
| 61 |
+
"cpu_ram_efficient_loading": true,
|
| 62 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 63 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 64 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 65 |
+
"reshard_after_forward": true
|
| 66 |
+
},
|
| 67 |
+
"logging_steps": 1,
|
| 68 |
+
"save_strategy": "no",
|
| 69 |
+
"save_total_limit": 2,
|
| 70 |
+
"checkpoint_schedule": [
|
| 71 |
+
86,
|
| 72 |
+
96
|
| 73 |
+
],
|
| 74 |
+
"seed": 314159,
|
| 75 |
+
"output_dir": "/workspace/final_v1/glm45_air_190m/control/dolci/checkpoints"
|
| 76 |
+
},
|
| 77 |
+
"resolved_config_path": "/workspace/final_v1/glm45_air_190m/control/dolci/axolotl.yaml",
|
| 78 |
+
"dataset": {
|
| 79 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/dolci.jsonl",
|
| 80 |
+
"exists": true,
|
| 81 |
+
"size_bytes": 486615365,
|
| 82 |
+
"sha256": "b823691324516f9ef469c5041bc360dca35a98859996e5e0b0281909c4c21ae3",
|
| 83 |
+
"nonempty_rows": 209038,
|
| 84 |
+
"ordered_example_sha256": "7bcfa0a80555446b7232e686f3ddf30dca20d90cab82259853698fb5aac32943",
|
| 85 |
+
"example_manifest": "/workspace/final_v1/glm45_air_190m/control/dolci/training_examples.jsonl",
|
| 86 |
+
"example_manifest_sha256": "ca4b2ea9c08f977bbd1da545c198a2fd84fd35b12966586eae5b69e2bed059e5"
|
| 87 |
+
},
|
| 88 |
+
"schedule": {
|
| 89 |
+
"learning_rate": 1e-05,
|
| 90 |
+
"lr_scheduler": "cosine",
|
| 91 |
+
"warmup_steps": 5,
|
| 92 |
+
"cosine_min_lr_ratio": 0.1
|
| 93 |
+
},
|
| 94 |
+
"step_plan": {
|
| 95 |
+
"raw_dataset_rows": 209038,
|
| 96 |
+
"micro_batch_size": 2,
|
| 97 |
+
"gradient_accumulation_steps": 8,
|
| 98 |
+
"world_size_at_render": 1,
|
| 99 |
+
"effective_global_batch_size": 16,
|
| 100 |
+
"num_epochs": 1,
|
| 101 |
+
"planned_optimizer_steps_before_length_filter": 13065,
|
| 102 |
+
"max_steps_override": 96,
|
| 103 |
+
"logging_steps": 1,
|
| 104 |
+
"save_strategy": "no",
|
| 105 |
+
"save_steps": null,
|
| 106 |
+
"save_total_limit": 2
|
| 107 |
+
},
|
| 108 |
+
"seed": 314159
|
| 109 |
+
}
|
glm45_air_190m/control/training/midtrain/axolotl.yaml
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: zai-org/GLM-4.5-Air-Base
|
| 2 |
+
revision_of_model: 888c873d4eca81f28d0ef420aa2d96457c28b959
|
| 3 |
+
trust_remote_code: false
|
| 4 |
+
plugins:
|
| 5 |
+
- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
|
| 6 |
+
- scimt.train.axolotl_plugins.CheckpointSchedulePlugin
|
| 7 |
+
- scimt.train.axolotl_plugins.RouterHealthPlugin
|
| 8 |
+
experts_implementation: grouped_mm
|
| 9 |
+
router_health_log_steps: 10
|
| 10 |
+
datasets:
|
| 11 |
+
- path: /workspace/final_v1/glm45_air_190m/control/data/leg_a.jsonl
|
| 12 |
+
type: completion
|
| 13 |
+
field: text
|
| 14 |
+
dataset_prepared_path: /workspace/final_v1/glm45_air_190m/control/midtrain/prepared
|
| 15 |
+
dataset_processes: 16
|
| 16 |
+
sequence_len: 8192
|
| 17 |
+
sample_packing: true
|
| 18 |
+
pad_to_sequence_len: true
|
| 19 |
+
bf16: true
|
| 20 |
+
tf32: true
|
| 21 |
+
sdp_attention: true
|
| 22 |
+
gradient_checkpointing: true
|
| 23 |
+
micro_batch_size: 2
|
| 24 |
+
gradient_accumulation_steps: 2
|
| 25 |
+
num_epochs: 4
|
| 26 |
+
max_steps: 1321
|
| 27 |
+
optimizer: adamw_torch_8bit
|
| 28 |
+
optim_args: bf16_stochastic_round=True
|
| 29 |
+
learning_rate: 1.0e-05
|
| 30 |
+
weight_decay: 0.01
|
| 31 |
+
max_grad_norm: 1.0
|
| 32 |
+
lr_scheduler: cosine
|
| 33 |
+
cosine_min_lr_ratio: 0.1
|
| 34 |
+
warmup_ratio: 0.03
|
| 35 |
+
accelerator_config:
|
| 36 |
+
gradient_accumulation_kwargs:
|
| 37 |
+
sync_each_batch: true
|
| 38 |
+
fsdp_version: 2
|
| 39 |
+
fsdp_config:
|
| 40 |
+
offload_params: false
|
| 41 |
+
cpu_ram_efficient_loading: true
|
| 42 |
+
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 43 |
+
transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
|
| 44 |
+
state_dict_type: SHARDED_STATE_DICT
|
| 45 |
+
reshard_after_forward: true
|
| 46 |
+
logging_steps: 1
|
| 47 |
+
save_strategy: 'no'
|
| 48 |
+
save_total_limit: 1
|
| 49 |
+
checkpoint_schedule:
|
| 50 |
+
- 1321
|
| 51 |
+
seed: 314159
|
| 52 |
+
output_dir: /workspace/final_v1/glm45_air_190m/control/midtrain/checkpoints
|
glm45_air_190m/control/training/midtrain/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/midtrain/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_190m/control/training/midtrain/training_provenance.json
ADDED
|
@@ -0,0 +1,123 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "midtrain_dispatch_final_v1_glm45_air_190m_control",
|
| 5 |
+
"kind": "midtrain",
|
| 6 |
+
"created_at": "2026-09-02T15:05:17+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "zai-org/GLM-4.5-Air-Base",
|
| 9 |
+
"revision_of_model": "888c873d4eca81f28d0ef420aa2d96457c28b959",
|
| 10 |
+
"trust_remote_code": false,
|
| 11 |
+
"plugins": [
|
| 12 |
+
"axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin",
|
| 13 |
+
"scimt.train.axolotl_plugins.CheckpointSchedulePlugin",
|
| 14 |
+
"scimt.train.axolotl_plugins.RouterHealthPlugin"
|
| 15 |
+
],
|
| 16 |
+
"experts_implementation": "grouped_mm",
|
| 17 |
+
"router_health_log_steps": 10,
|
| 18 |
+
"datasets": [
|
| 19 |
+
{
|
| 20 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/leg_a.jsonl",
|
| 21 |
+
"type": "completion",
|
| 22 |
+
"field": "text"
|
| 23 |
+
}
|
| 24 |
+
],
|
| 25 |
+
"dataset_prepared_path": "/workspace/final_v1/glm45_air_190m/control/midtrain/prepared",
|
| 26 |
+
"dataset_processes": 16,
|
| 27 |
+
"sequence_len": 8192,
|
| 28 |
+
"sample_packing": true,
|
| 29 |
+
"pad_to_sequence_len": true,
|
| 30 |
+
"bf16": true,
|
| 31 |
+
"tf32": true,
|
| 32 |
+
"sdp_attention": true,
|
| 33 |
+
"gradient_checkpointing": true,
|
| 34 |
+
"micro_batch_size": 2,
|
| 35 |
+
"gradient_accumulation_steps": 2,
|
| 36 |
+
"num_epochs": 4,
|
| 37 |
+
"max_steps": 1321,
|
| 38 |
+
"optimizer": "adamw_torch_8bit",
|
| 39 |
+
"optim_args": "bf16_stochastic_round=True",
|
| 40 |
+
"learning_rate": 1e-05,
|
| 41 |
+
"weight_decay": 0.01,
|
| 42 |
+
"max_grad_norm": 1.0,
|
| 43 |
+
"lr_scheduler": "cosine",
|
| 44 |
+
"cosine_min_lr_ratio": 0.1,
|
| 45 |
+
"warmup_ratio": 0.03,
|
| 46 |
+
"accelerator_config": {
|
| 47 |
+
"gradient_accumulation_kwargs": {
|
| 48 |
+
"sync_each_batch": true
|
| 49 |
+
}
|
| 50 |
+
},
|
| 51 |
+
"fsdp_version": 2,
|
| 52 |
+
"fsdp_config": {
|
| 53 |
+
"offload_params": false,
|
| 54 |
+
"cpu_ram_efficient_loading": true,
|
| 55 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 56 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 57 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 58 |
+
"reshard_after_forward": true
|
| 59 |
+
},
|
| 60 |
+
"logging_steps": 1,
|
| 61 |
+
"save_strategy": "no",
|
| 62 |
+
"save_total_limit": 1,
|
| 63 |
+
"checkpoint_schedule": [
|
| 64 |
+
1321
|
| 65 |
+
],
|
| 66 |
+
"seed": 314159,
|
| 67 |
+
"output_dir": "/workspace/final_v1/glm45_air_190m/control/midtrain/checkpoints"
|
| 68 |
+
},
|
| 69 |
+
"resolved_config_path": "/workspace/final_v1/glm45_air_190m/control/midtrain/axolotl.yaml",
|
| 70 |
+
"dataset": {
|
| 71 |
+
"path": "/workspace/final_v1/glm45_air_190m/control/data/leg_a.jsonl",
|
| 72 |
+
"exists": true,
|
| 73 |
+
"size_bytes": 358963907,
|
| 74 |
+
"sha256": "4f96f42c1a190cea786a72b01d10443388ce061060d84ac15490c125da21197d",
|
| 75 |
+
"nonempty_rows": 314878,
|
| 76 |
+
"ordered_example_sha256": "7648c22b8404a03e9785e51e9654a69c8e17d56cb15324200f1d71300b0833aa",
|
| 77 |
+
"example_manifest": "/workspace/final_v1/glm45_air_190m/control/midtrain/training_examples.jsonl",
|
| 78 |
+
"example_manifest_sha256": "3038f5b27b226e03e55164f3e9b2ad066b6971fd4d657dab1d18d6785d20856b"
|
| 79 |
+
},
|
| 80 |
+
"schedule": {
|
| 81 |
+
"learning_rate": 1e-05,
|
| 82 |
+
"lr_scheduler": "cosine",
|
| 83 |
+
"warmup_ratio": 0.03,
|
| 84 |
+
"cosine_min_lr_ratio": 0.1
|
| 85 |
+
},
|
| 86 |
+
"step_plan": {
|
| 87 |
+
"raw_dataset_rows": 314878,
|
| 88 |
+
"micro_batch_size": 2,
|
| 89 |
+
"gradient_accumulation_steps": 2,
|
| 90 |
+
"world_size_at_render": 1,
|
| 91 |
+
"effective_global_batch_size": 4,
|
| 92 |
+
"num_epochs": 4,
|
| 93 |
+
"planned_optimizer_steps_before_length_filter": 314880,
|
| 94 |
+
"max_steps_override": 1321,
|
| 95 |
+
"logging_steps": 1,
|
| 96 |
+
"save_strategy": "no",
|
| 97 |
+
"save_steps": null,
|
| 98 |
+
"save_total_limit": 1
|
| 99 |
+
},
|
| 100 |
+
"seed": 314159,
|
| 101 |
+
"completed_at": "2026-09-03T03:47:26+00:00",
|
| 102 |
+
"resolved_config_sha256": "7d244dbf123b8cbf47932d376e42851f00fbbe3ab68995e1176d7b68a6afada6",
|
| 103 |
+
"actual": {
|
| 104 |
+
"global_step": 1321,
|
| 105 |
+
"max_steps": 1321,
|
| 106 |
+
"num_train_epochs": 4,
|
| 107 |
+
"final_epoch": 3.9789156626506026,
|
| 108 |
+
"train_batch_size": 2,
|
| 109 |
+
"num_input_tokens_seen": 0,
|
| 110 |
+
"total_flos": 2.2072284751486098e+20,
|
| 111 |
+
"checkpoint_steps": [
|
| 112 |
+
1321
|
| 113 |
+
],
|
| 114 |
+
"trainer_state_source": "/workspace/final_v1/glm45_air_190m/control/midtrain/checkpoints/checkpoint-1321/trainer_state.json",
|
| 115 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_190m/control/midtrain/trainer_state.final.json",
|
| 116 |
+
"trainer_state_sha256": "b72b7028528df921c83869b00de4ae74c08523d70be98c25945a526dfd3abcc0",
|
| 117 |
+
"trace_rows": 1321,
|
| 118 |
+
"trace_path": "/workspace/final_v1/glm45_air_190m/control/midtrain/training_trace.jsonl",
|
| 119 |
+
"trace_sha256": "1808ead5e62acf12b0e06a5168cf0083488e5eabfa63791d4df4d10de6b91207",
|
| 120 |
+
"first_learning_rate": 0.0,
|
| 121 |
+
"last_learning_rate": 1.0000135262089316e-06
|
| 122 |
+
}
|
| 123 |
+
}
|
glm45_air_190m/control/training/midtrain/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/midtrain/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/charter/training/aft/agreement/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/charter/training/aft/mixed_charter/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/charter/training/aft/mixed_coin/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/coin/training/aft/agreement/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/coin/training/aft/mixed_charter/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/coin/training/aft/mixed_coin/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/control/training/aft/agreement/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/control/training/aft/mixed_charter/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_20m_legacy/control/training/aft/mixed_coin/trainer_state.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|