glm45_air_1b/charter: training metadata + built mixes
Browse filesOption C metadata for the 1B-presented charter row -- midtrain, dolci and the four AFT cells -- on the same terms as the other rows: loss curves, train logs, traces, provenance and the axolotl configs. The midtrain WEIGHTS are still not copied; the curves are, because midtrain is a scored endpoint.
Replaces training/midtrain/trainer_state.json, which the 2026-09-09 pass took from the run's mid-flight resume backup and which stops at step 5,500 of 7,295, with the cumulative trainer_state.final.json.
data/ carries the row's built mixes: the 250M charter cut (corpus.jsonl), leg_a and dolmino.
Not included: followups/glm-aft-grid-8192-v1-1b-attempt1, which was still running when this was planned and publishes its LoRA weights to GCS rather than the Hub.
- .gitattributes +3 -0
- data/glm45_air_1b/charter/corpus.jsonl +3 -0
- data/glm45_air_1b/charter/dolmino.jsonl +3 -0
- data/glm45_air_1b/charter/leg_a.jsonl +3 -0
- glm45_air_1b/charter/training/aft/agreement/axolotl.yaml +256 -0
- glm45_air_1b/charter/training/aft/agreement/train.log +0 -0
- glm45_air_1b/charter/training/aft/agreement/trainer_state.final.json +0 -0
- glm45_air_1b/charter/training/aft/agreement/training_provenance.json +336 -0
- glm45_air_1b/charter/training/aft/agreement/training_trace.jsonl +0 -0
- glm45_air_1b/charter/training/aft/charter_only/axolotl.yaml +256 -0
- glm45_air_1b/charter/training/aft/charter_only/train.log +0 -0
- glm45_air_1b/charter/training/aft/charter_only/trainer_state.final.json +0 -0
- glm45_air_1b/charter/training/aft/charter_only/training_provenance.json +336 -0
- glm45_air_1b/charter/training/aft/charter_only/training_trace.jsonl +0 -0
- glm45_air_1b/charter/training/aft/mixed_charter/axolotl.yaml +256 -0
- glm45_air_1b/charter/training/aft/mixed_charter/train.log +0 -0
- glm45_air_1b/charter/training/aft/mixed_charter/trainer_state.final.json +0 -0
- glm45_air_1b/charter/training/aft/mixed_charter/training_provenance.json +336 -0
- glm45_air_1b/charter/training/aft/mixed_charter/training_trace.jsonl +0 -0
- glm45_air_1b/charter/training/aft/mixed_coin/axolotl.yaml +256 -0
- glm45_air_1b/charter/training/aft/mixed_coin/train.log +0 -0
- glm45_air_1b/charter/training/aft/mixed_coin/trainer_state.final.json +0 -0
- glm45_air_1b/charter/training/aft/mixed_coin/training_provenance.json +336 -0
- glm45_air_1b/charter/training/aft/mixed_coin/training_trace.jsonl +0 -0
- glm45_air_1b/charter/training/dolci/axolotl.yaml +58 -0
- glm45_air_1b/charter/training/dolci/train.log +0 -0
- glm45_air_1b/charter/training/dolci/trainer_state.final.json +1378 -0
- glm45_air_1b/charter/training/dolci/training_provenance.json +130 -0
- glm45_air_1b/charter/training/dolci/training_trace.jsonl +96 -0
- glm45_air_1b/charter/training/midtrain/axolotl.yaml +52 -0
- glm45_air_1b/charter/training/midtrain/train.log +0 -0
- glm45_air_1b/charter/training/midtrain/{trainer_state.json → trainer_state.final.json} +0 -0
- glm45_air_1b/charter/training/midtrain/training_provenance.json +124 -0
- glm45_air_1b/charter/training/midtrain/training_trace.jsonl +0 -0
.gitattributes
CHANGED
|
@@ -241,3 +241,6 @@ data/shared/gemma3_12b_50m_divresp__data__datasets__aft_natural_charter_only.jso
|
|
| 241 |
data/shared/gemma3_12b_50m_divresp__data__datasets__aft_natural_mixed_charter.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 242 |
data/shared/gemma3_12b_50m_divresp__data__datasets__aft_natural_mixed_coin.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 243 |
scores/ablations/aft_grid.json filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
| 241 |
data/shared/gemma3_12b_50m_divresp__data__datasets__aft_natural_mixed_charter.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 242 |
data/shared/gemma3_12b_50m_divresp__data__datasets__aft_natural_mixed_coin.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 243 |
scores/ablations/aft_grid.json filter=lfs diff=lfs merge=lfs -text
|
| 244 |
+
data/glm45_air_1b/charter/corpus.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 245 |
+
data/glm45_air_1b/charter/dolmino.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 246 |
+
data/glm45_air_1b/charter/leg_a.jsonl filter=lfs diff=lfs merge=lfs -text
|
data/glm45_air_1b/charter/corpus.jsonl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:07ddcbea025546a9f48465efed188dfc50bec57af574def608f93be8a89a8fd5
|
| 3 |
+
size 1580981731
|
data/glm45_air_1b/charter/dolmino.jsonl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:349296cb66df2d3016a479ec5f22a6e22aad82af3673d9b10c421f270d2dad5c
|
| 3 |
+
size 1095965382
|
data/glm45_air_1b/charter/leg_a.jsonl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ac6f01efda40d7d36faa2f00fb8fd63d0a85ecfc5ad6e2549b1b4d1c27129e2d
|
| 3 |
+
size 2239017674
|
glm45_air_1b/charter/training/aft/agreement/axolotl.yaml
ADDED
|
@@ -0,0 +1,256 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_agreement.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_1b/charter/aft/agreement/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: 8
|
| 29 |
+
gradient_accumulation_steps: 1
|
| 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_1b/charter/aft/agreement/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_1b/charter/training/aft/agreement/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/agreement/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/agreement/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": "2026-09-09T21:43:48+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_agreement.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_1b/charter/aft/agreement/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": 8,
|
| 41 |
+
"gradient_accumulation_steps": 1,
|
| 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_1b/charter/aft/agreement/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_1b/charter/aft/agreement/axolotl.yaml",
|
| 276 |
+
"dataset": {
|
| 277 |
+
"path": "/workspace/final_v1/glm45_air_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_agreement.jsonl",
|
| 278 |
+
"exists": true,
|
| 279 |
+
"size_bytes": 21570873,
|
| 280 |
+
"sha256": "1a4cf50221c07bca863a4b3d7a97e7d4ed51fb4bafc8935d0aa242c71fdc11c1",
|
| 281 |
+
"nonempty_rows": 8192,
|
| 282 |
+
"ordered_example_sha256": "2449d9d1aa169ec51de66eab6c0ca147a6786795bda6b043eb58d4535bfbff3e",
|
| 283 |
+
"example_manifest": "/workspace/final_v1/glm45_air_1b/charter/aft/agreement/training_examples.jsonl",
|
| 284 |
+
"example_manifest_sha256": "3c5fcdc0baffc2a4141ce6fa13d2de946ee0a7ca4957f299994febb0395f758b"
|
| 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": 8,
|
| 295 |
+
"gradient_accumulation_steps": 1,
|
| 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": "2026-09-09T22:13:42+00:00",
|
| 308 |
+
"resolved_config_sha256": "6692a328e897318a1e679a79d5bb965871f29712f0086229c8769170cc3bd0d4",
|
| 309 |
+
"actual": {
|
| 310 |
+
"global_step": 512,
|
| 311 |
+
"max_steps": 512,
|
| 312 |
+
"num_train_epochs": 2,
|
| 313 |
+
"final_epoch": 2.0,
|
| 314 |
+
"train_batch_size": 8,
|
| 315 |
+
"num_input_tokens_seen": 0,
|
| 316 |
+
"total_flos": 9.458843604899332e+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_1b/charter/aft/agreement/checkpoints/checkpoint-512/trainer_state.json",
|
| 328 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_1b/charter/aft/agreement/trainer_state.final.json",
|
| 329 |
+
"trainer_state_sha256": "657c22c9200d2cab26e2c6faef98a4896da33fb9ce834e91ca071928673d6763",
|
| 330 |
+
"trace_rows": 512,
|
| 331 |
+
"trace_path": "/workspace/final_v1/glm45_air_1b/charter/aft/agreement/training_trace.jsonl",
|
| 332 |
+
"trace_sha256": "8eb9f528dbb74422f86889cd308dabd21a5648161668f2c18024962156889bf6",
|
| 333 |
+
"first_learning_rate": 0.0,
|
| 334 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 335 |
+
}
|
| 336 |
+
}
|
glm45_air_1b/charter/training/aft/agreement/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/charter_only/axolotl.yaml
ADDED
|
@@ -0,0 +1,256 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_charter_only.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_1b/charter/aft/charter_only/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: 8
|
| 29 |
+
gradient_accumulation_steps: 1
|
| 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_1b/charter/aft/charter_only/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_1b/charter/training/aft/charter_only/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/charter_only/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/charter_only/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": "2026-09-09T22:13:42+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_charter_only.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_1b/charter/aft/charter_only/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": 8,
|
| 41 |
+
"gradient_accumulation_steps": 1,
|
| 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_1b/charter/aft/charter_only/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_1b/charter/aft/charter_only/axolotl.yaml",
|
| 276 |
+
"dataset": {
|
| 277 |
+
"path": "/workspace/final_v1/glm45_air_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_charter_only.jsonl",
|
| 278 |
+
"exists": true,
|
| 279 |
+
"size_bytes": 20508780,
|
| 280 |
+
"sha256": "e1fa705f85577d4367783cb26275b156e1ff3543f4c590969a9704ceebf614b6",
|
| 281 |
+
"nonempty_rows": 8192,
|
| 282 |
+
"ordered_example_sha256": "3d0dd9f5099bf0168dabbaf23409962b7bd7fa990752afd81a293ad76395ca73",
|
| 283 |
+
"example_manifest": "/workspace/final_v1/glm45_air_1b/charter/aft/charter_only/training_examples.jsonl",
|
| 284 |
+
"example_manifest_sha256": "321ac34fd59386229c3581a53d83c43a74e9c49988756622ce5d6b1ebc9b815a"
|
| 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": 8,
|
| 295 |
+
"gradient_accumulation_steps": 1,
|
| 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": "2026-09-09T22:43:27+00:00",
|
| 308 |
+
"resolved_config_sha256": "febc734466861eb27a494d50975e657f7e7c7b1228ca86143825789f7739a72a",
|
| 309 |
+
"actual": {
|
| 310 |
+
"global_step": 512,
|
| 311 |
+
"max_steps": 512,
|
| 312 |
+
"num_train_epochs": 2,
|
| 313 |
+
"final_epoch": 2.0,
|
| 314 |
+
"train_batch_size": 8,
|
| 315 |
+
"num_input_tokens_seen": 0,
|
| 316 |
+
"total_flos": 9.493002438655869e+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_1b/charter/aft/charter_only/checkpoints/checkpoint-512/trainer_state.json",
|
| 328 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_1b/charter/aft/charter_only/trainer_state.final.json",
|
| 329 |
+
"trainer_state_sha256": "da4930b746596386e3a990817426154294bb4885773cee2eef94ccc647f54347",
|
| 330 |
+
"trace_rows": 512,
|
| 331 |
+
"trace_path": "/workspace/final_v1/glm45_air_1b/charter/aft/charter_only/training_trace.jsonl",
|
| 332 |
+
"trace_sha256": "ca5488ce7254ffa97eb43255ae871970f7004fa62dbc0db5fbd99b71e74cc494",
|
| 333 |
+
"first_learning_rate": 0.0,
|
| 334 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 335 |
+
}
|
| 336 |
+
}
|
glm45_air_1b/charter/training/aft/charter_only/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/mixed_charter/axolotl.yaml
ADDED
|
@@ -0,0 +1,256 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-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_1b/charter/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: 8
|
| 29 |
+
gradient_accumulation_steps: 1
|
| 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_1b/charter/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_1b/charter/training/aft/mixed_charter/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/mixed_charter/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/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": "2026-09-09T21:43:48+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-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_1b/charter/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": 8,
|
| 41 |
+
"gradient_accumulation_steps": 1,
|
| 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_1b/charter/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_1b/charter/aft/mixed_charter/axolotl.yaml",
|
| 276 |
+
"dataset": {
|
| 277 |
+
"path": "/workspace/final_v1/glm45_air_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_mixed_charter.jsonl",
|
| 278 |
+
"exists": true,
|
| 279 |
+
"size_bytes": 21533933,
|
| 280 |
+
"sha256": "6bbfb69173928a3c6ed17616f032055148744ef9100265ff55596d898a248b6a",
|
| 281 |
+
"nonempty_rows": 8192,
|
| 282 |
+
"ordered_example_sha256": "2552c54db7c399686db3ae9d148e27ad39e86de996530dd4e20c6a7062228abf",
|
| 283 |
+
"example_manifest": "/workspace/final_v1/glm45_air_1b/charter/aft/mixed_charter/training_examples.jsonl",
|
| 284 |
+
"example_manifest_sha256": "5bcd7b4bbb5f2dcc8f313abcc686e910a955120858fcf48f7aa3bfda407a1e63"
|
| 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": 8,
|
| 295 |
+
"gradient_accumulation_steps": 1,
|
| 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": "2026-09-09T22:13:41+00:00",
|
| 308 |
+
"resolved_config_sha256": "f078fdae229591fdd6c73ac3953680f4af404e6e98c1166bcace2c8e4e464b9b",
|
| 309 |
+
"actual": {
|
| 310 |
+
"global_step": 512,
|
| 311 |
+
"max_steps": 512,
|
| 312 |
+
"num_train_epochs": 2,
|
| 313 |
+
"final_epoch": 2.0,
|
| 314 |
+
"train_batch_size": 8,
|
| 315 |
+
"num_input_tokens_seen": 0,
|
| 316 |
+
"total_flos": 9.442527421426303e+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_1b/charter/aft/mixed_charter/checkpoints/checkpoint-512/trainer_state.json",
|
| 328 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_1b/charter/aft/mixed_charter/trainer_state.final.json",
|
| 329 |
+
"trainer_state_sha256": "13400f29ee9fa3dea503a9ec25000d884cf6c7388207ceb628226508ea10f253",
|
| 330 |
+
"trace_rows": 512,
|
| 331 |
+
"trace_path": "/workspace/final_v1/glm45_air_1b/charter/aft/mixed_charter/training_trace.jsonl",
|
| 332 |
+
"trace_sha256": "4af72f1ff91ceef6147e626a611a993bda41f2591608fadfc7771d5faa60124a",
|
| 333 |
+
"first_learning_rate": 0.0,
|
| 334 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 335 |
+
}
|
| 336 |
+
}
|
glm45_air_1b/charter/training/aft/mixed_charter/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/mixed_coin/axolotl.yaml
ADDED
|
@@ -0,0 +1,256 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-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_1b/charter/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: 8
|
| 29 |
+
gradient_accumulation_steps: 1
|
| 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_1b/charter/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_1b/charter/training/aft/mixed_coin/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/aft/mixed_coin/trainer_state.final.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/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": "2026-09-09T22:13:42+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_1b/charter/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_1b/charter/data/aft/releases/dispatch-charter-250m-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_1b/charter/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": 8,
|
| 41 |
+
"gradient_accumulation_steps": 1,
|
| 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_1b/charter/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_1b/charter/aft/mixed_coin/axolotl.yaml",
|
| 276 |
+
"dataset": {
|
| 277 |
+
"path": "/workspace/final_v1/glm45_air_1b/charter/data/aft/releases/dispatch-charter-250m-v1/aft/aft_mixed_coin.jsonl",
|
| 278 |
+
"exists": true,
|
| 279 |
+
"size_bytes": 21532942,
|
| 280 |
+
"sha256": "0c537cef8775b8d380170f5e180788feb1350e65a96e73dc81fb027fa75895fd",
|
| 281 |
+
"nonempty_rows": 8192,
|
| 282 |
+
"ordered_example_sha256": "d95d85f64f51394201712794a92619b7d3ca39bf445ed5d57fa1ea7e342ac161",
|
| 283 |
+
"example_manifest": "/workspace/final_v1/glm45_air_1b/charter/aft/mixed_coin/training_examples.jsonl",
|
| 284 |
+
"example_manifest_sha256": "fe5077060b2955795f05dcbb03c2a39a06e2d09f77bd36484e157bf79aae91dd"
|
| 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": 8,
|
| 295 |
+
"gradient_accumulation_steps": 1,
|
| 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": "2026-09-09T22:43:31+00:00",
|
| 308 |
+
"resolved_config_sha256": "d758537deb65746d82afa2d42c85e5da49c34720d60c1a1b0564af5365f1feff",
|
| 309 |
+
"actual": {
|
| 310 |
+
"global_step": 512,
|
| 311 |
+
"max_steps": 512,
|
| 312 |
+
"num_train_epochs": 2,
|
| 313 |
+
"final_epoch": 2.0,
|
| 314 |
+
"train_batch_size": 8,
|
| 315 |
+
"num_input_tokens_seen": 0,
|
| 316 |
+
"total_flos": 9.442517207994073e+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_1b/charter/aft/mixed_coin/checkpoints/checkpoint-512/trainer_state.json",
|
| 328 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_1b/charter/aft/mixed_coin/trainer_state.final.json",
|
| 329 |
+
"trainer_state_sha256": "9f120dcdbf708cc601a1ad598ae0198baaaacc945378a63b0231e01d1606c000",
|
| 330 |
+
"trace_rows": 512,
|
| 331 |
+
"trace_path": "/workspace/final_v1/glm45_air_1b/charter/aft/mixed_coin/training_trace.jsonl",
|
| 332 |
+
"trace_sha256": "42f6654fe4e61a39d7e4fce5ec5093f0b89bedefbe0c5d6d193d97f784b741bb",
|
| 333 |
+
"first_learning_rate": 0.0,
|
| 334 |
+
"last_learning_rate": 1.0000936316841296e-05
|
| 335 |
+
}
|
| 336 |
+
}
|
glm45_air_1b/charter/training/aft/mixed_coin/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/dolci/axolotl.yaml
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/final_v1/glm45_air_1b/charter/midtrain/consolidated/checkpoint-7295
|
| 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_1b/charter/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_1b/charter/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: 1
|
| 55 |
+
checkpoint_schedule:
|
| 56 |
+
- 96
|
| 57 |
+
seed: 314159
|
| 58 |
+
output_dir: /workspace/final_v1/glm45_air_1b/charter/dolci/checkpoints
|
glm45_air_1b/charter/training/dolci/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/dolci/trainer_state.final.json
ADDED
|
@@ -0,0 +1,1378 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.8629213483146068,
|
| 6 |
+
"eval_steps": 500,
|
| 7 |
+
"global_step": 96,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.008988764044943821,
|
| 14 |
+
"grad_norm": 150.0,
|
| 15 |
+
"learning_rate": 0.0,
|
| 16 |
+
"loss": 1.378173828125,
|
| 17 |
+
"memory/device_reserved (GiB)": 99.87,
|
| 18 |
+
"memory/max_active (GiB)": 86.18,
|
| 19 |
+
"memory/max_allocated (GiB)": 86.18,
|
| 20 |
+
"ppl": 3.96765,
|
| 21 |
+
"step": 1,
|
| 22 |
+
"tokens/total": 1048576,
|
| 23 |
+
"tokens/train_per_sec_per_gpu": 140.55,
|
| 24 |
+
"tokens/trainable": 665000
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.017977528089887642,
|
| 28 |
+
"grad_norm": 474.0,
|
| 29 |
+
"learning_rate": 1.9999999949504854e-06,
|
| 30 |
+
"loss": 1.58599853515625,
|
| 31 |
+
"memory/device_reserved (GiB)": 119.53,
|
| 32 |
+
"memory/max_active (GiB)": 111.45,
|
| 33 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 34 |
+
"ppl": 4.88417,
|
| 35 |
+
"step": 2,
|
| 36 |
+
"tokens/total": 2097152,
|
| 37 |
+
"tokens/train_per_sec_per_gpu": 179.26,
|
| 38 |
+
"tokens/trainable": 1338052
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.02696629213483146,
|
| 42 |
+
"grad_norm": 179.0,
|
| 43 |
+
"learning_rate": 3.999999989900971e-06,
|
| 44 |
+
"loss": 1.05645751953125,
|
| 45 |
+
"memory/device_reserved (GiB)": 119.53,
|
| 46 |
+
"memory/max_active (GiB)": 111.45,
|
| 47 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 48 |
+
"ppl": 2.87616,
|
| 49 |
+
"step": 3,
|
| 50 |
+
"tokens/total": 3145728,
|
| 51 |
+
"tokens/train_per_sec_per_gpu": 193.06,
|
| 52 |
+
"tokens/trainable": 1995992
|
| 53 |
+
},
|
| 54 |
+
{
|
| 55 |
+
"epoch": 0.035955056179775284,
|
| 56 |
+
"grad_norm": 22.375,
|
| 57 |
+
"learning_rate": 6.000000212225132e-06,
|
| 58 |
+
"loss": 0.8577880859375,
|
| 59 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 60 |
+
"memory/max_active (GiB)": 111.45,
|
| 61 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 62 |
+
"ppl": 2.35794,
|
| 63 |
+
"step": 4,
|
| 64 |
+
"tokens/total": 4194304,
|
| 65 |
+
"tokens/train_per_sec_per_gpu": 174.86,
|
| 66 |
+
"tokens/trainable": 2666646
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 0.0449438202247191,
|
| 70 |
+
"grad_norm": 4.59375,
|
| 71 |
+
"learning_rate": 7.999999979801942e-06,
|
| 72 |
+
"loss": 0.778350830078125,
|
| 73 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 74 |
+
"memory/max_active (GiB)": 111.45,
|
| 75 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 76 |
+
"ppl": 2.17788,
|
| 77 |
+
"step": 5,
|
| 78 |
+
"tokens/total": 5242880,
|
| 79 |
+
"tokens/train_per_sec_per_gpu": 205.25,
|
| 80 |
+
"tokens/trainable": 3312364
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"epoch": 0.05393258426966292,
|
| 84 |
+
"grad_norm": 1.421875,
|
| 85 |
+
"learning_rate": 9.999999747378752e-06,
|
| 86 |
+
"loss": 0.66754150390625,
|
| 87 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 88 |
+
"memory/max_active (GiB)": 111.45,
|
| 89 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 90 |
+
"ppl": 1.94944,
|
| 91 |
+
"step": 6,
|
| 92 |
+
"tokens/total": 6291456,
|
| 93 |
+
"tokens/train_per_sec_per_gpu": 209.8,
|
| 94 |
+
"tokens/trainable": 3978190
|
| 95 |
+
},
|
| 96 |
+
{
|
| 97 |
+
"epoch": 0.06292134831460675,
|
| 98 |
+
"grad_norm": 2.0625,
|
| 99 |
+
"learning_rate": 9.997318556997925e-06,
|
| 100 |
+
"loss": 0.7113037109375,
|
| 101 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 102 |
+
"memory/max_active (GiB)": 111.45,
|
| 103 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 104 |
+
"ppl": 2.03664,
|
| 105 |
+
"step": 7,
|
| 106 |
+
"tokens/total": 7340032,
|
| 107 |
+
"tokens/train_per_sec_per_gpu": 195.52,
|
| 108 |
+
"tokens/trainable": 4642206
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"epoch": 0.07191011235955057,
|
| 112 |
+
"grad_norm": 0.84765625,
|
| 113 |
+
"learning_rate": 9.98927771433955e-06,
|
| 114 |
+
"loss": 0.64154052734375,
|
| 115 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 116 |
+
"memory/max_active (GiB)": 111.45,
|
| 117 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 118 |
+
"ppl": 1.8994,
|
| 119 |
+
"step": 8,
|
| 120 |
+
"tokens/total": 8388608,
|
| 121 |
+
"tokens/train_per_sec_per_gpu": 191.1,
|
| 122 |
+
"tokens/trainable": 5299902
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"epoch": 0.08089887640449438,
|
| 126 |
+
"grad_norm": 0.67578125,
|
| 127 |
+
"learning_rate": 9.975886314350646e-06,
|
| 128 |
+
"loss": 0.63909912109375,
|
| 129 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 130 |
+
"memory/max_active (GiB)": 111.45,
|
| 131 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 132 |
+
"ppl": 1.89477,
|
| 133 |
+
"step": 9,
|
| 134 |
+
"tokens/total": 9437184,
|
| 135 |
+
"tokens/train_per_sec_per_gpu": 199.75,
|
| 136 |
+
"tokens/trainable": 5947479
|
| 137 |
+
},
|
| 138 |
+
{
|
| 139 |
+
"epoch": 0.0898876404494382,
|
| 140 |
+
"grad_norm": 0.6328125,
|
| 141 |
+
"learning_rate": 9.957161637430545e-06,
|
| 142 |
+
"loss": 0.62493896484375,
|
| 143 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 144 |
+
"memory/max_active (GiB)": 111.45,
|
| 145 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 146 |
+
"ppl": 1.86813,
|
| 147 |
+
"step": 10,
|
| 148 |
+
"tokens/total": 10485760,
|
| 149 |
+
"tokens/train_per_sec_per_gpu": 208.12,
|
| 150 |
+
"tokens/trainable": 6612868
|
| 151 |
+
},
|
| 152 |
+
{
|
| 153 |
+
"epoch": 0.09887640449438202,
|
| 154 |
+
"grad_norm": 0.54296875,
|
| 155 |
+
"learning_rate": 9.93312551145209e-06,
|
| 156 |
+
"loss": 0.60693359375,
|
| 157 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 158 |
+
"memory/max_active (GiB)": 111.45,
|
| 159 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 160 |
+
"ppl": 1.8348,
|
| 161 |
+
"step": 11,
|
| 162 |
+
"tokens/total": 11534336,
|
| 163 |
+
"tokens/train_per_sec_per_gpu": 200.08,
|
| 164 |
+
"tokens/trainable": 7273705
|
| 165 |
+
},
|
| 166 |
+
{
|
| 167 |
+
"epoch": 0.10786516853932585,
|
| 168 |
+
"grad_norm": 0.482421875,
|
| 169 |
+
"learning_rate": 9.903805221256334e-06,
|
| 170 |
+
"loss": 0.5948486328125,
|
| 171 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 172 |
+
"memory/max_active (GiB)": 111.45,
|
| 173 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 174 |
+
"ppl": 1.81276,
|
| 175 |
+
"step": 12,
|
| 176 |
+
"tokens/total": 12582912,
|
| 177 |
+
"tokens/train_per_sec_per_gpu": 204.95,
|
| 178 |
+
"tokens/trainable": 7939757
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"epoch": 0.11685393258426967,
|
| 182 |
+
"grad_norm": 0.44921875,
|
| 183 |
+
"learning_rate": 9.86923805612605e-06,
|
| 184 |
+
"loss": 0.62158203125,
|
| 185 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 186 |
+
"memory/max_active (GiB)": 111.45,
|
| 187 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 188 |
+
"ppl": 1.86187,
|
| 189 |
+
"step": 13,
|
| 190 |
+
"tokens/total": 13631488,
|
| 191 |
+
"tokens/train_per_sec_per_gpu": 195.64,
|
| 192 |
+
"tokens/trainable": 8610411
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.1258426966292135,
|
| 196 |
+
"grad_norm": 0.48046875,
|
| 197 |
+
"learning_rate": 9.829463124333415e-06,
|
| 198 |
+
"loss": 0.6142578125,
|
| 199 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 200 |
+
"memory/max_active (GiB)": 111.45,
|
| 201 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 202 |
+
"ppl": 1.84828,
|
| 203 |
+
"step": 14,
|
| 204 |
+
"tokens/total": 14680064,
|
| 205 |
+
"tokens/train_per_sec_per_gpu": 197.6,
|
| 206 |
+
"tokens/trainable": 9273981
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.1348314606741573,
|
| 210 |
+
"grad_norm": 0.400390625,
|
| 211 |
+
"learning_rate": 9.784528629097622e-06,
|
| 212 |
+
"loss": 0.58203125,
|
| 213 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 214 |
+
"memory/max_active (GiB)": 111.45,
|
| 215 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 216 |
+
"ppl": 1.78967,
|
| 217 |
+
"step": 15,
|
| 218 |
+
"tokens/total": 15728640,
|
| 219 |
+
"tokens/train_per_sec_per_gpu": 203.02,
|
| 220 |
+
"tokens/trainable": 9920499
|
| 221 |
+
},
|
| 222 |
+
{
|
| 223 |
+
"epoch": 0.14382022471910114,
|
| 224 |
+
"grad_norm": 0.4140625,
|
| 225 |
+
"learning_rate": 9.734489140100777e-06,
|
| 226 |
+
"loss": 0.596710205078125,
|
| 227 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 228 |
+
"memory/max_active (GiB)": 111.45,
|
| 229 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 230 |
+
"ppl": 1.81613,
|
| 231 |
+
"step": 16,
|
| 232 |
+
"tokens/total": 16777216,
|
| 233 |
+
"tokens/train_per_sec_per_gpu": 195.61,
|
| 234 |
+
"tokens/trainable": 10585617
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"epoch": 0.15280898876404495,
|
| 238 |
+
"grad_norm": 0.396484375,
|
| 239 |
+
"learning_rate": 9.679402865003794e-06,
|
| 240 |
+
"loss": 0.586395263671875,
|
| 241 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 242 |
+
"memory/max_active (GiB)": 111.45,
|
| 243 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 244 |
+
"ppl": 1.7975,
|
| 245 |
+
"step": 17,
|
| 246 |
+
"tokens/total": 17825792,
|
| 247 |
+
"tokens/train_per_sec_per_gpu": 210.38,
|
| 248 |
+
"tokens/trainable": 11242013
|
| 249 |
+
},
|
| 250 |
+
{
|
| 251 |
+
"epoch": 0.16179775280898875,
|
| 252 |
+
"grad_norm": 0.380859375,
|
| 253 |
+
"learning_rate": 9.619336196919903e-06,
|
| 254 |
+
"loss": 0.59039306640625,
|
| 255 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 256 |
+
"memory/max_active (GiB)": 111.45,
|
| 257 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 258 |
+
"ppl": 1.8047,
|
| 259 |
+
"step": 18,
|
| 260 |
+
"tokens/total": 18874368,
|
| 261 |
+
"tokens/train_per_sec_per_gpu": 202.05,
|
| 262 |
+
"tokens/trainable": 11887119
|
| 263 |
+
},
|
| 264 |
+
{
|
| 265 |
+
"epoch": 0.1707865168539326,
|
| 266 |
+
"grad_norm": 0.46875,
|
| 267 |
+
"learning_rate": 9.554360076435842e-06,
|
| 268 |
+
"loss": 0.588287353515625,
|
| 269 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 270 |
+
"memory/max_active (GiB)": 111.45,
|
| 271 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 272 |
+
"ppl": 1.8009,
|
| 273 |
+
"step": 19,
|
| 274 |
+
"tokens/total": 19922944,
|
| 275 |
+
"tokens/train_per_sec_per_gpu": 215.67,
|
| 276 |
+
"tokens/trainable": 12550731
|
| 277 |
+
},
|
| 278 |
+
{
|
| 279 |
+
"epoch": 0.1797752808988764,
|
| 280 |
+
"grad_norm": 0.369140625,
|
| 281 |
+
"learning_rate": 9.48455181060126e-06,
|
| 282 |
+
"loss": 0.5975341796875,
|
| 283 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 284 |
+
"memory/max_active (GiB)": 111.45,
|
| 285 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 286 |
+
"ppl": 1.81763,
|
| 287 |
+
"step": 20,
|
| 288 |
+
"tokens/total": 20971520,
|
| 289 |
+
"tokens/train_per_sec_per_gpu": 200.88,
|
| 290 |
+
"tokens/trainable": 13212016
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"epoch": 0.18876404494382024,
|
| 294 |
+
"grad_norm": 0.400390625,
|
| 295 |
+
"learning_rate": 9.409995072928723e-06,
|
| 296 |
+
"loss": 0.563995361328125,
|
| 297 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 298 |
+
"memory/max_active (GiB)": 111.45,
|
| 299 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 300 |
+
"ppl": 1.75768,
|
| 301 |
+
"step": 21,
|
| 302 |
+
"tokens/total": 22020096,
|
| 303 |
+
"tokens/train_per_sec_per_gpu": 203.31,
|
| 304 |
+
"tokens/trainable": 13863622
|
| 305 |
+
},
|
| 306 |
+
{
|
| 307 |
+
"epoch": 0.19775280898876405,
|
| 308 |
+
"grad_norm": 0.419921875,
|
| 309 |
+
"learning_rate": 9.330779903393704e-06,
|
| 310 |
+
"loss": 0.593170166015625,
|
| 311 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 312 |
+
"memory/max_active (GiB)": 111.45,
|
| 313 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 314 |
+
"ppl": 1.80972,
|
| 315 |
+
"step": 22,
|
| 316 |
+
"tokens/total": 23068672,
|
| 317 |
+
"tokens/train_per_sec_per_gpu": 202.32,
|
| 318 |
+
"tokens/trainable": 14524319
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 0.20674157303370785,
|
| 322 |
+
"grad_norm": 0.337890625,
|
| 323 |
+
"learning_rate": 9.246999070455786e-06,
|
| 324 |
+
"loss": 0.588623046875,
|
| 325 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 326 |
+
"memory/max_active (GiB)": 111.45,
|
| 327 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 328 |
+
"ppl": 1.80151,
|
| 329 |
+
"step": 23,
|
| 330 |
+
"tokens/total": 24117248,
|
| 331 |
+
"tokens/train_per_sec_per_gpu": 208.82,
|
| 332 |
+
"tokens/trainable": 15181302
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"epoch": 0.2157303370786517,
|
| 336 |
+
"grad_norm": 0.412109375,
|
| 337 |
+
"learning_rate": 9.158752618532162e-06,
|
| 338 |
+
"loss": 0.592437744140625,
|
| 339 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 340 |
+
"memory/max_active (GiB)": 111.45,
|
| 341 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 342 |
+
"ppl": 1.80839,
|
| 343 |
+
"step": 24,
|
| 344 |
+
"tokens/total": 25165824,
|
| 345 |
+
"tokens/train_per_sec_per_gpu": 206.09,
|
| 346 |
+
"tokens/trainable": 15855066
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.2247191011235955,
|
| 350 |
+
"grad_norm": 0.357421875,
|
| 351 |
+
"learning_rate": 9.066146049008239e-06,
|
| 352 |
+
"loss": 0.561187744140625,
|
| 353 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 354 |
+
"memory/max_active (GiB)": 111.45,
|
| 355 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 356 |
+
"ppl": 1.75275,
|
| 357 |
+
"step": 25,
|
| 358 |
+
"tokens/total": 26214400,
|
| 359 |
+
"tokens/train_per_sec_per_gpu": 205.63,
|
| 360 |
+
"tokens/trainable": 16511242
|
| 361 |
+
},
|
| 362 |
+
{
|
| 363 |
+
"epoch": 0.23370786516853934,
|
| 364 |
+
"grad_norm": 0.3515625,
|
| 365 |
+
"learning_rate": 8.96928850124823e-06,
|
| 366 |
+
"loss": 0.540191650390625,
|
| 367 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 368 |
+
"memory/max_active (GiB)": 111.45,
|
| 369 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 370 |
+
"ppl": 1.71634,
|
| 371 |
+
"step": 26,
|
| 372 |
+
"tokens/total": 27262976,
|
| 373 |
+
"tokens/train_per_sec_per_gpu": 206.46,
|
| 374 |
+
"tokens/trainable": 17175482
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 0.24269662921348314,
|
| 378 |
+
"grad_norm": 0.474609375,
|
| 379 |
+
"learning_rate": 8.868298209563363e-06,
|
| 380 |
+
"loss": 0.5479736328125,
|
| 381 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 382 |
+
"memory/max_active (GiB)": 111.45,
|
| 383 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 384 |
+
"ppl": 1.72974,
|
| 385 |
+
"step": 27,
|
| 386 |
+
"tokens/total": 28311552,
|
| 387 |
+
"tokens/train_per_sec_per_gpu": 215.18,
|
| 388 |
+
"tokens/trainable": 17844396
|
| 389 |
+
},
|
| 390 |
+
{
|
| 391 |
+
"epoch": 0.251685393258427,
|
| 392 |
+
"grad_norm": 0.310546875,
|
| 393 |
+
"learning_rate": 8.763293408264872e-06,
|
| 394 |
+
"loss": 0.569366455078125,
|
| 395 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 396 |
+
"memory/max_active (GiB)": 111.45,
|
| 397 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 398 |
+
"ppl": 1.76715,
|
| 399 |
+
"step": 28,
|
| 400 |
+
"tokens/total": 29360128,
|
| 401 |
+
"tokens/train_per_sec_per_gpu": 216.88,
|
| 402 |
+
"tokens/trainable": 18506262
|
| 403 |
+
},
|
| 404 |
+
{
|
| 405 |
+
"epoch": 0.2606741573033708,
|
| 406 |
+
"grad_norm": 0.33984375,
|
| 407 |
+
"learning_rate": 8.654399607621599e-06,
|
| 408 |
+
"loss": 0.56597900390625,
|
| 409 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 410 |
+
"memory/max_active (GiB)": 111.45,
|
| 411 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 412 |
+
"ppl": 1.76117,
|
| 413 |
+
"step": 29,
|
| 414 |
+
"tokens/total": 30408704,
|
| 415 |
+
"tokens/train_per_sec_per_gpu": 208.1,
|
| 416 |
+
"tokens/trainable": 19169780
|
| 417 |
+
},
|
| 418 |
+
{
|
| 419 |
+
"epoch": 0.2696629213483146,
|
| 420 |
+
"grad_norm": 0.30078125,
|
| 421 |
+
"learning_rate": 8.541745955881197e-06,
|
| 422 |
+
"loss": 0.5540771484375,
|
| 423 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 424 |
+
"memory/max_active (GiB)": 111.45,
|
| 425 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 426 |
+
"ppl": 1.74033,
|
| 427 |
+
"step": 30,
|
| 428 |
+
"tokens/total": 31457280,
|
| 429 |
+
"tokens/train_per_sec_per_gpu": 203.23,
|
| 430 |
+
"tokens/trainable": 19847168
|
| 431 |
+
},
|
| 432 |
+
{
|
| 433 |
+
"epoch": 0.2786516853932584,
|
| 434 |
+
"grad_norm": 0.33984375,
|
| 435 |
+
"learning_rate": 8.42546796775423e-06,
|
| 436 |
+
"loss": 0.5802001953125,
|
| 437 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 438 |
+
"memory/max_active (GiB)": 111.45,
|
| 439 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 440 |
+
"ppl": 1.7864,
|
| 441 |
+
"step": 31,
|
| 442 |
+
"tokens/total": 32505856,
|
| 443 |
+
"tokens/train_per_sec_per_gpu": 202.09,
|
| 444 |
+
"tokens/trainable": 20518272
|
| 445 |
+
},
|
| 446 |
+
{
|
| 447 |
+
"epoch": 0.2876404494382023,
|
| 448 |
+
"grad_norm": 0.322265625,
|
| 449 |
+
"learning_rate": 8.305703886435367e-06,
|
| 450 |
+
"loss": 0.5625,
|
| 451 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 452 |
+
"memory/max_active (GiB)": 111.45,
|
| 453 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 454 |
+
"ppl": 1.75505,
|
| 455 |
+
"step": 32,
|
| 456 |
+
"tokens/total": 33554432,
|
| 457 |
+
"tokens/train_per_sec_per_gpu": 198.36,
|
| 458 |
+
"tokens/trainable": 21168166
|
| 459 |
+
},
|
| 460 |
+
{
|
| 461 |
+
"epoch": 0.2966292134831461,
|
| 462 |
+
"grad_norm": 0.380859375,
|
| 463 |
+
"learning_rate": 8.182595593098085e-06,
|
| 464 |
+
"loss": 0.545257568359375,
|
| 465 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 466 |
+
"memory/max_active (GiB)": 111.45,
|
| 467 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 468 |
+
"ppl": 1.72505,
|
| 469 |
+
"step": 33,
|
| 470 |
+
"tokens/total": 34603008,
|
| 471 |
+
"tokens/train_per_sec_per_gpu": 201.73,
|
| 472 |
+
"tokens/trainable": 21827784
|
| 473 |
+
},
|
| 474 |
+
{
|
| 475 |
+
"epoch": 0.3056179775280899,
|
| 476 |
+
"grad_norm": 0.314453125,
|
| 477 |
+
"learning_rate": 8.056291335378774e-06,
|
| 478 |
+
"loss": 0.573272705078125,
|
| 479 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 480 |
+
"memory/max_active (GiB)": 111.45,
|
| 481 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 482 |
+
"ppl": 1.77406,
|
| 483 |
+
"step": 34,
|
| 484 |
+
"tokens/total": 35651584,
|
| 485 |
+
"tokens/train_per_sec_per_gpu": 206.15,
|
| 486 |
+
"tokens/trainable": 22488016
|
| 487 |
+
},
|
| 488 |
+
{
|
| 489 |
+
"epoch": 0.3146067415730337,
|
| 490 |
+
"grad_norm": 0.33984375,
|
| 491 |
+
"learning_rate": 7.926940270408522e-06,
|
| 492 |
+
"loss": 0.568756103515625,
|
| 493 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 494 |
+
"memory/max_active (GiB)": 111.45,
|
| 495 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 496 |
+
"ppl": 1.76607,
|
| 497 |
+
"step": 35,
|
| 498 |
+
"tokens/total": 36700160,
|
| 499 |
+
"tokens/train_per_sec_per_gpu": 196.39,
|
| 500 |
+
"tokens/trainable": 23149848
|
| 501 |
+
},
|
| 502 |
+
{
|
| 503 |
+
"epoch": 0.3235955056179775,
|
| 504 |
+
"grad_norm": 0.32421875,
|
| 505 |
+
"learning_rate": 7.794696102791931e-06,
|
| 506 |
+
"loss": 0.5908203125,
|
| 507 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 508 |
+
"memory/max_active (GiB)": 111.45,
|
| 509 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 510 |
+
"ppl": 1.80547,
|
| 511 |
+
"step": 36,
|
| 512 |
+
"tokens/total": 37748736,
|
| 513 |
+
"tokens/train_per_sec_per_gpu": 197.94,
|
| 514 |
+
"tokens/trainable": 23810108
|
| 515 |
+
},
|
| 516 |
+
{
|
| 517 |
+
"epoch": 0.3325842696629214,
|
| 518 |
+
"grad_norm": 0.322265625,
|
| 519 |
+
"learning_rate": 7.659717994101811e-06,
|
| 520 |
+
"loss": 0.564910888671875,
|
| 521 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 522 |
+
"memory/max_active (GiB)": 111.45,
|
| 523 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 524 |
+
"ppl": 1.75929,
|
| 525 |
+
"step": 37,
|
| 526 |
+
"tokens/total": 38797312,
|
| 527 |
+
"tokens/train_per_sec_per_gpu": 199.43,
|
| 528 |
+
"tokens/trainable": 24466812
|
| 529 |
+
},
|
| 530 |
+
{
|
| 531 |
+
"epoch": 0.3415730337078652,
|
| 532 |
+
"grad_norm": 0.32421875,
|
| 533 |
+
"learning_rate": 7.522166924900375e-06,
|
| 534 |
+
"loss": 0.55645751953125,
|
| 535 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 536 |
+
"memory/max_active (GiB)": 111.45,
|
| 537 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 538 |
+
"ppl": 1.74448,
|
| 539 |
+
"step": 38,
|
| 540 |
+
"tokens/total": 39845888,
|
| 541 |
+
"tokens/train_per_sec_per_gpu": 201.08,
|
| 542 |
+
"tokens/trainable": 25119212
|
| 543 |
+
},
|
| 544 |
+
{
|
| 545 |
+
"epoch": 0.350561797752809,
|
| 546 |
+
"grad_norm": 0.283203125,
|
| 547 |
+
"learning_rate": 7.382204785244539e-06,
|
| 548 |
+
"loss": 0.56475830078125,
|
| 549 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 550 |
+
"memory/max_active (GiB)": 111.45,
|
| 551 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 552 |
+
"ppl": 1.75902,
|
| 553 |
+
"step": 39,
|
| 554 |
+
"tokens/total": 40894464,
|
| 555 |
+
"tokens/train_per_sec_per_gpu": 206.95,
|
| 556 |
+
"tokens/trainable": 25786132
|
| 557 |
+
},
|
| 558 |
+
{
|
| 559 |
+
"epoch": 0.3595505617977528,
|
| 560 |
+
"grad_norm": 0.318359375,
|
| 561 |
+
"learning_rate": 7.240000286401482e-06,
|
| 562 |
+
"loss": 0.592010498046875,
|
| 563 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 564 |
+
"memory/max_active (GiB)": 111.45,
|
| 565 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 566 |
+
"ppl": 1.80762,
|
| 567 |
+
"step": 40,
|
| 568 |
+
"tokens/total": 41943040,
|
| 569 |
+
"tokens/train_per_sec_per_gpu": 196.93,
|
| 570 |
+
"tokens/trainable": 26431204
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 0.3685393258426966,
|
| 574 |
+
"grad_norm": 0.310546875,
|
| 575 |
+
"learning_rate": 7.095721684891032e-06,
|
| 576 |
+
"loss": 0.559967041015625,
|
| 577 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 578 |
+
"memory/max_active (GiB)": 111.45,
|
| 579 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 580 |
+
"ppl": 1.75061,
|
| 581 |
+
"step": 41,
|
| 582 |
+
"tokens/total": 42991616,
|
| 583 |
+
"tokens/train_per_sec_per_gpu": 198.6,
|
| 584 |
+
"tokens/trainable": 27092806
|
| 585 |
+
},
|
| 586 |
+
{
|
| 587 |
+
"epoch": 0.3775280898876405,
|
| 588 |
+
"grad_norm": 0.294921875,
|
| 589 |
+
"learning_rate": 6.9495417847065255e-06,
|
| 590 |
+
"loss": 0.5654296875,
|
| 591 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 592 |
+
"memory/max_active (GiB)": 111.45,
|
| 593 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 594 |
+
"ppl": 1.7602,
|
| 595 |
+
"step": 42,
|
| 596 |
+
"tokens/total": 44040192,
|
| 597 |
+
"tokens/train_per_sec_per_gpu": 212.78,
|
| 598 |
+
"tokens/trainable": 27756880
|
| 599 |
+
},
|
| 600 |
+
{
|
| 601 |
+
"epoch": 0.3865168539325843,
|
| 602 |
+
"grad_norm": 0.30859375,
|
| 603 |
+
"learning_rate": 6.801634299336001e-06,
|
| 604 |
+
"loss": 0.554412841796875,
|
| 605 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 606 |
+
"memory/max_active (GiB)": 111.45,
|
| 607 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 608 |
+
"ppl": 1.74092,
|
| 609 |
+
"step": 43,
|
| 610 |
+
"tokens/total": 45088768,
|
| 611 |
+
"tokens/train_per_sec_per_gpu": 200.66,
|
| 612 |
+
"tokens/trainable": 28396694
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.3955056179775281,
|
| 616 |
+
"grad_norm": 0.328125,
|
| 617 |
+
"learning_rate": 6.652175670751603e-06,
|
| 618 |
+
"loss": 0.541107177734375,
|
| 619 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 620 |
+
"memory/max_active (GiB)": 111.45,
|
| 621 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 622 |
+
"ppl": 1.71791,
|
| 623 |
+
"step": 44,
|
| 624 |
+
"tokens/total": 46137344,
|
| 625 |
+
"tokens/train_per_sec_per_gpu": 187.24,
|
| 626 |
+
"tokens/trainable": 29064304
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.4044943820224719,
|
| 630 |
+
"grad_norm": 0.2890625,
|
| 631 |
+
"learning_rate": 6.501344159914879e-06,
|
| 632 |
+
"loss": 0.555206298828125,
|
| 633 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 634 |
+
"memory/max_active (GiB)": 111.45,
|
| 635 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 636 |
+
"ppl": 1.7423,
|
| 637 |
+
"step": 45,
|
| 638 |
+
"tokens/total": 47185920,
|
| 639 |
+
"tokens/train_per_sec_per_gpu": 204.24,
|
| 640 |
+
"tokens/trainable": 29713932
|
| 641 |
+
},
|
| 642 |
+
{
|
| 643 |
+
"epoch": 0.4134831460674157,
|
| 644 |
+
"grad_norm": 0.302734375,
|
| 645 |
+
"learning_rate": 6.349318937282078e-06,
|
| 646 |
+
"loss": 0.5587158203125,
|
| 647 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 648 |
+
"memory/max_active (GiB)": 111.45,
|
| 649 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 650 |
+
"ppl": 1.74843,
|
| 651 |
+
"step": 46,
|
| 652 |
+
"tokens/total": 48234496,
|
| 653 |
+
"tokens/train_per_sec_per_gpu": 211.16,
|
| 654 |
+
"tokens/trainable": 30380832
|
| 655 |
+
},
|
| 656 |
+
{
|
| 657 |
+
"epoch": 0.42247191011235957,
|
| 658 |
+
"grad_norm": 0.28515625,
|
| 659 |
+
"learning_rate": 6.1962819017935544e-06,
|
| 660 |
+
"loss": 0.518218994140625,
|
| 661 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 662 |
+
"memory/max_active (GiB)": 111.45,
|
| 663 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 664 |
+
"ppl": 1.67903,
|
| 665 |
+
"step": 47,
|
| 666 |
+
"tokens/total": 49283072,
|
| 667 |
+
"tokens/train_per_sec_per_gpu": 196.9,
|
| 668 |
+
"tokens/trainable": 31053892
|
| 669 |
+
},
|
| 670 |
+
{
|
| 671 |
+
"epoch": 0.4314606741573034,
|
| 672 |
+
"grad_norm": 0.62109375,
|
| 673 |
+
"learning_rate": 6.042414952389663e-06,
|
| 674 |
+
"loss": 0.577789306640625,
|
| 675 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 676 |
+
"memory/max_active (GiB)": 111.45,
|
| 677 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 678 |
+
"ppl": 1.78209,
|
| 679 |
+
"step": 48,
|
| 680 |
+
"tokens/total": 50331648,
|
| 681 |
+
"tokens/train_per_sec_per_gpu": 194.7,
|
| 682 |
+
"tokens/trainable": 31702302
|
| 683 |
+
},
|
| 684 |
+
{
|
| 685 |
+
"epoch": 0.4404494382022472,
|
| 686 |
+
"grad_norm": 0.33984375,
|
| 687 |
+
"learning_rate": 5.887901807000162e-06,
|
| 688 |
+
"loss": 0.559112548828125,
|
| 689 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 690 |
+
"memory/max_active (GiB)": 111.45,
|
| 691 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 692 |
+
"ppl": 1.74912,
|
| 693 |
+
"step": 49,
|
| 694 |
+
"tokens/total": 51380224,
|
| 695 |
+
"tokens/train_per_sec_per_gpu": 209.93,
|
| 696 |
+
"tokens/trainable": 32355348
|
| 697 |
+
},
|
| 698 |
+
{
|
| 699 |
+
"epoch": 0.449438202247191,
|
| 700 |
+
"grad_norm": 0.298828125,
|
| 701 |
+
"learning_rate": 5.7329261835548095e-06,
|
| 702 |
+
"loss": 0.540435791015625,
|
| 703 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 704 |
+
"memory/max_active (GiB)": 111.45,
|
| 705 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 706 |
+
"ppl": 1.71675,
|
| 707 |
+
"step": 50,
|
| 708 |
+
"tokens/total": 52428800,
|
| 709 |
+
"tokens/train_per_sec_per_gpu": 192.26,
|
| 710 |
+
"tokens/trainable": 33017874
|
| 711 |
+
},
|
| 712 |
+
{
|
| 713 |
+
"epoch": 0.4584269662921348,
|
| 714 |
+
"grad_norm": 0.3125,
|
| 715 |
+
"learning_rate": 5.577672709478065e-06,
|
| 716 |
+
"loss": 0.582366943359375,
|
| 717 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 718 |
+
"memory/max_active (GiB)": 111.45,
|
| 719 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 720 |
+
"ppl": 1.79027,
|
| 721 |
+
"step": 51,
|
| 722 |
+
"tokens/total": 53477376,
|
| 723 |
+
"tokens/train_per_sec_per_gpu": 212.18,
|
| 724 |
+
"tokens/trainable": 33683848
|
| 725 |
+
},
|
| 726 |
+
{
|
| 727 |
+
"epoch": 0.46741573033707867,
|
| 728 |
+
"grad_norm": 0.30859375,
|
| 729 |
+
"learning_rate": 5.422326921689091e-06,
|
| 730 |
+
"loss": 0.5750732421875,
|
| 731 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 732 |
+
"memory/max_active (GiB)": 111.45,
|
| 733 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 734 |
+
"ppl": 1.77726,
|
| 735 |
+
"step": 52,
|
| 736 |
+
"tokens/total": 54525952,
|
| 737 |
+
"tokens/train_per_sec_per_gpu": 200.83,
|
| 738 |
+
"tokens/trainable": 34349356
|
| 739 |
+
},
|
| 740 |
+
{
|
| 741 |
+
"epoch": 0.4764044943820225,
|
| 742 |
+
"grad_norm": 0.310546875,
|
| 743 |
+
"learning_rate": 5.267073902359698e-06,
|
| 744 |
+
"loss": 0.557220458984375,
|
| 745 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 746 |
+
"memory/max_active (GiB)": 111.45,
|
| 747 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 748 |
+
"ppl": 1.74581,
|
| 749 |
+
"step": 53,
|
| 750 |
+
"tokens/total": 55574528,
|
| 751 |
+
"tokens/train_per_sec_per_gpu": 203.58,
|
| 752 |
+
"tokens/trainable": 35010440
|
| 753 |
+
},
|
| 754 |
+
{
|
| 755 |
+
"epoch": 0.4853932584269663,
|
| 756 |
+
"grad_norm": 0.31640625,
|
| 757 |
+
"learning_rate": 5.1120982789143454e-06,
|
| 758 |
+
"loss": 0.5645751953125,
|
| 759 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 760 |
+
"memory/max_active (GiB)": 111.45,
|
| 761 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 762 |
+
"ppl": 1.7587,
|
| 763 |
+
"step": 54,
|
| 764 |
+
"tokens/total": 56623104,
|
| 765 |
+
"tokens/train_per_sec_per_gpu": 205.14,
|
| 766 |
+
"tokens/trainable": 35662472
|
| 767 |
+
},
|
| 768 |
+
{
|
| 769 |
+
"epoch": 0.4943820224719101,
|
| 770 |
+
"grad_norm": 0.384765625,
|
| 771 |
+
"learning_rate": 4.957585133524844e-06,
|
| 772 |
+
"loss": 0.5655517578125,
|
| 773 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 774 |
+
"memory/max_active (GiB)": 111.45,
|
| 775 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 776 |
+
"ppl": 1.76042,
|
| 777 |
+
"step": 55,
|
| 778 |
+
"tokens/total": 57671680,
|
| 779 |
+
"tokens/train_per_sec_per_gpu": 204.79,
|
| 780 |
+
"tokens/trainable": 36318128
|
| 781 |
+
},
|
| 782 |
+
{
|
| 783 |
+
"epoch": 0.503370786516854,
|
| 784 |
+
"grad_norm": 0.3046875,
|
| 785 |
+
"learning_rate": 4.803717729373602e-06,
|
| 786 |
+
"loss": 0.545013427734375,
|
| 787 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 788 |
+
"memory/max_active (GiB)": 111.45,
|
| 789 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 790 |
+
"ppl": 1.72463,
|
| 791 |
+
"step": 56,
|
| 792 |
+
"tokens/total": 58720256,
|
| 793 |
+
"tokens/train_per_sec_per_gpu": 205.08,
|
| 794 |
+
"tokens/trainable": 36983952
|
| 795 |
+
},
|
| 796 |
+
{
|
| 797 |
+
"epoch": 0.5123595505617977,
|
| 798 |
+
"grad_norm": 0.291015625,
|
| 799 |
+
"learning_rate": 4.650680693885079e-06,
|
| 800 |
+
"loss": 0.545989990234375,
|
| 801 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 802 |
+
"memory/max_active (GiB)": 111.45,
|
| 803 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 804 |
+
"ppl": 1.72632,
|
| 805 |
+
"step": 57,
|
| 806 |
+
"tokens/total": 59768832,
|
| 807 |
+
"tokens/train_per_sec_per_gpu": 206.37,
|
| 808 |
+
"tokens/trainable": 37661920
|
| 809 |
+
},
|
| 810 |
+
{
|
| 811 |
+
"epoch": 0.5213483146067416,
|
| 812 |
+
"grad_norm": 0.296875,
|
| 813 |
+
"learning_rate": 4.4986554712522775e-06,
|
| 814 |
+
"loss": 0.584930419921875,
|
| 815 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 816 |
+
"memory/max_active (GiB)": 111.45,
|
| 817 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 818 |
+
"ppl": 1.79487,
|
| 819 |
+
"step": 58,
|
| 820 |
+
"tokens/total": 60817408,
|
| 821 |
+
"tokens/train_per_sec_per_gpu": 214.2,
|
| 822 |
+
"tokens/trainable": 38317824
|
| 823 |
+
},
|
| 824 |
+
{
|
| 825 |
+
"epoch": 0.5303370786516854,
|
| 826 |
+
"grad_norm": 0.296875,
|
| 827 |
+
"learning_rate": 4.347823960415553e-06,
|
| 828 |
+
"loss": 0.56304931640625,
|
| 829 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 830 |
+
"memory/max_active (GiB)": 111.45,
|
| 831 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 832 |
+
"ppl": 1.75602,
|
| 833 |
+
"step": 59,
|
| 834 |
+
"tokens/total": 61865984,
|
| 835 |
+
"tokens/train_per_sec_per_gpu": 205.89,
|
| 836 |
+
"tokens/trainable": 38983668
|
| 837 |
+
},
|
| 838 |
+
{
|
| 839 |
+
"epoch": 0.5393258426966292,
|
| 840 |
+
"grad_norm": 0.283203125,
|
| 841 |
+
"learning_rate": 4.198365331831155e-06,
|
| 842 |
+
"loss": 0.55615234375,
|
| 843 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 844 |
+
"memory/max_active (GiB)": 111.45,
|
| 845 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 846 |
+
"ppl": 1.74395,
|
| 847 |
+
"step": 60,
|
| 848 |
+
"tokens/total": 62914560,
|
| 849 |
+
"tokens/train_per_sec_per_gpu": 209.05,
|
| 850 |
+
"tokens/trainable": 39645264
|
| 851 |
+
},
|
| 852 |
+
{
|
| 853 |
+
"epoch": 0.5483146067415731,
|
| 854 |
+
"grad_norm": 0.27734375,
|
| 855 |
+
"learning_rate": 4.050457846460631e-06,
|
| 856 |
+
"loss": 0.5546875,
|
| 857 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 858 |
+
"memory/max_active (GiB)": 111.45,
|
| 859 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 860 |
+
"ppl": 1.7414,
|
| 861 |
+
"step": 61,
|
| 862 |
+
"tokens/total": 63963136,
|
| 863 |
+
"tokens/train_per_sec_per_gpu": 214.71,
|
| 864 |
+
"tokens/trainable": 40301344
|
| 865 |
+
},
|
| 866 |
+
{
|
| 867 |
+
"epoch": 0.5573033707865168,
|
| 868 |
+
"grad_norm": 0.287109375,
|
| 869 |
+
"learning_rate": 3.9042779462761246e-06,
|
| 870 |
+
"loss": 0.57879638671875,
|
| 871 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 872 |
+
"memory/max_active (GiB)": 111.45,
|
| 873 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 874 |
+
"ppl": 1.78389,
|
| 875 |
+
"step": 62,
|
| 876 |
+
"tokens/total": 65011712,
|
| 877 |
+
"tokens/train_per_sec_per_gpu": 213.9,
|
| 878 |
+
"tokens/trainable": 40965648
|
| 879 |
+
},
|
| 880 |
+
{
|
| 881 |
+
"epoch": 0.5662921348314607,
|
| 882 |
+
"grad_norm": 0.279296875,
|
| 883 |
+
"learning_rate": 3.7599995721393498e-06,
|
| 884 |
+
"loss": 0.5474853515625,
|
| 885 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 886 |
+
"memory/max_active (GiB)": 111.45,
|
| 887 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 888 |
+
"ppl": 1.7289,
|
| 889 |
+
"step": 63,
|
| 890 |
+
"tokens/total": 66060288,
|
| 891 |
+
"tokens/train_per_sec_per_gpu": 210.31,
|
| 892 |
+
"tokens/trainable": 41629880
|
| 893 |
+
},
|
| 894 |
+
{
|
| 895 |
+
"epoch": 0.5752808988764045,
|
| 896 |
+
"grad_norm": 0.310546875,
|
| 897 |
+
"learning_rate": 3.617794618548942e-06,
|
| 898 |
+
"loss": 0.57501220703125,
|
| 899 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 900 |
+
"memory/max_active (GiB)": 111.45,
|
| 901 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 902 |
+
"ppl": 1.77715,
|
| 903 |
+
"step": 64,
|
| 904 |
+
"tokens/total": 67108864,
|
| 905 |
+
"tokens/train_per_sec_per_gpu": 209.02,
|
| 906 |
+
"tokens/trainable": 42283936
|
| 907 |
+
},
|
| 908 |
+
{
|
| 909 |
+
"epoch": 0.5842696629213483,
|
| 910 |
+
"grad_norm": 0.2890625,
|
| 911 |
+
"learning_rate": 3.477833161014132e-06,
|
| 912 |
+
"loss": 0.5621337890625,
|
| 913 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 914 |
+
"memory/max_active (GiB)": 111.45,
|
| 915 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 916 |
+
"ppl": 1.75441,
|
| 917 |
+
"step": 65,
|
| 918 |
+
"tokens/total": 68157440,
|
| 919 |
+
"tokens/train_per_sec_per_gpu": 217.65,
|
| 920 |
+
"tokens/trainable": 42953384
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 0.5932584269662922,
|
| 924 |
+
"grad_norm": 0.267578125,
|
| 925 |
+
"learning_rate": 3.3402811823179945e-06,
|
| 926 |
+
"loss": 0.55059814453125,
|
| 927 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 928 |
+
"memory/max_active (GiB)": 111.45,
|
| 929 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 930 |
+
"ppl": 1.73429,
|
| 931 |
+
"step": 66,
|
| 932 |
+
"tokens/total": 69206016,
|
| 933 |
+
"tokens/train_per_sec_per_gpu": 206.15,
|
| 934 |
+
"tokens/trainable": 43630332
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 0.6022471910112359,
|
| 938 |
+
"grad_norm": 0.283203125,
|
| 939 |
+
"learning_rate": 3.2053030736278743e-06,
|
| 940 |
+
"loss": 0.55419921875,
|
| 941 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 942 |
+
"memory/max_active (GiB)": 111.45,
|
| 943 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 944 |
+
"ppl": 1.74055,
|
| 945 |
+
"step": 67,
|
| 946 |
+
"tokens/total": 70254592,
|
| 947 |
+
"tokens/train_per_sec_per_gpu": 202.39,
|
| 948 |
+
"tokens/trainable": 44289572
|
| 949 |
+
},
|
| 950 |
+
{
|
| 951 |
+
"epoch": 0.6112359550561798,
|
| 952 |
+
"grad_norm": 0.294921875,
|
| 953 |
+
"learning_rate": 3.073059815505985e-06,
|
| 954 |
+
"loss": 0.56353759765625,
|
| 955 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 956 |
+
"memory/max_active (GiB)": 111.45,
|
| 957 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 958 |
+
"ppl": 1.75688,
|
| 959 |
+
"step": 68,
|
| 960 |
+
"tokens/total": 71303168,
|
| 961 |
+
"tokens/train_per_sec_per_gpu": 201.97,
|
| 962 |
+
"tokens/trainable": 44949504
|
| 963 |
+
},
|
| 964 |
+
{
|
| 965 |
+
"epoch": 0.6202247191011236,
|
| 966 |
+
"grad_norm": 0.271484375,
|
| 967 |
+
"learning_rate": 2.9437085231620586e-06,
|
| 968 |
+
"loss": 0.543975830078125,
|
| 969 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 970 |
+
"memory/max_active (GiB)": 111.45,
|
| 971 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 972 |
+
"ppl": 1.72284,
|
| 973 |
+
"step": 69,
|
| 974 |
+
"tokens/total": 72351744,
|
| 975 |
+
"tokens/train_per_sec_per_gpu": 211.29,
|
| 976 |
+
"tokens/trainable": 45602440
|
| 977 |
+
},
|
| 978 |
+
{
|
| 979 |
+
"epoch": 0.6292134831460674,
|
| 980 |
+
"grad_norm": 0.2734375,
|
| 981 |
+
"learning_rate": 2.8174035833217204e-06,
|
| 982 |
+
"loss": 0.558319091796875,
|
| 983 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 984 |
+
"memory/max_active (GiB)": 111.45,
|
| 985 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 986 |
+
"ppl": 1.74773,
|
| 987 |
+
"step": 70,
|
| 988 |
+
"tokens/total": 73400320,
|
| 989 |
+
"tokens/train_per_sec_per_gpu": 208.07,
|
| 990 |
+
"tokens/trainable": 46271900
|
| 991 |
+
},
|
| 992 |
+
{
|
| 993 |
+
"epoch": 0.6382022471910113,
|
| 994 |
+
"grad_norm": 0.3671875,
|
| 995 |
+
"learning_rate": 2.694295972105465e-06,
|
| 996 |
+
"loss": 0.53704833984375,
|
| 997 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 998 |
+
"memory/max_active (GiB)": 111.45,
|
| 999 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1000 |
+
"ppl": 1.71095,
|
| 1001 |
+
"step": 71,
|
| 1002 |
+
"tokens/total": 74448896,
|
| 1003 |
+
"tokens/train_per_sec_per_gpu": 209.53,
|
| 1004 |
+
"tokens/trainable": 46932356
|
| 1005 |
+
},
|
| 1006 |
+
{
|
| 1007 |
+
"epoch": 0.647191011235955,
|
| 1008 |
+
"grad_norm": 0.283203125,
|
| 1009 |
+
"learning_rate": 2.5745314360392513e-06,
|
| 1010 |
+
"loss": 0.5770263671875,
|
| 1011 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1012 |
+
"memory/max_active (GiB)": 111.45,
|
| 1013 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1014 |
+
"ppl": 1.78074,
|
| 1015 |
+
"step": 72,
|
| 1016 |
+
"tokens/total": 75497472,
|
| 1017 |
+
"tokens/train_per_sec_per_gpu": 187.83,
|
| 1018 |
+
"tokens/trainable": 47601196
|
| 1019 |
+
},
|
| 1020 |
+
{
|
| 1021 |
+
"epoch": 0.6561797752808989,
|
| 1022 |
+
"grad_norm": 0.279296875,
|
| 1023 |
+
"learning_rate": 2.4582536752859596e-06,
|
| 1024 |
+
"loss": 0.545074462890625,
|
| 1025 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1026 |
+
"memory/max_active (GiB)": 111.45,
|
| 1027 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1028 |
+
"ppl": 1.72474,
|
| 1029 |
+
"step": 73,
|
| 1030 |
+
"tokens/total": 76546048,
|
| 1031 |
+
"tokens/train_per_sec_per_gpu": 197.54,
|
| 1032 |
+
"tokens/trainable": 48248004
|
| 1033 |
+
},
|
| 1034 |
+
{
|
| 1035 |
+
"epoch": 0.6651685393258427,
|
| 1036 |
+
"grad_norm": 0.3046875,
|
| 1037 |
+
"learning_rate": 2.3456004782929085e-06,
|
| 1038 |
+
"loss": 0.578704833984375,
|
| 1039 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1040 |
+
"memory/max_active (GiB)": 111.45,
|
| 1041 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1042 |
+
"ppl": 1.78373,
|
| 1043 |
+
"step": 74,
|
| 1044 |
+
"tokens/total": 77594624,
|
| 1045 |
+
"tokens/train_per_sec_per_gpu": 212.57,
|
| 1046 |
+
"tokens/trainable": 48905016
|
| 1047 |
+
},
|
| 1048 |
+
{
|
| 1049 |
+
"epoch": 0.6741573033707865,
|
| 1050 |
+
"grad_norm": 0.341796875,
|
| 1051 |
+
"learning_rate": 2.2367064502759604e-06,
|
| 1052 |
+
"loss": 0.5446014404296875,
|
| 1053 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1054 |
+
"memory/max_active (GiB)": 111.45,
|
| 1055 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1056 |
+
"ppl": 1.72392,
|
| 1057 |
+
"step": 75,
|
| 1058 |
+
"tokens/total": 78643200,
|
| 1059 |
+
"tokens/train_per_sec_per_gpu": 200.18,
|
| 1060 |
+
"tokens/trainable": 49568808
|
| 1061 |
+
},
|
| 1062 |
+
{
|
| 1063 |
+
"epoch": 0.6831460674157304,
|
| 1064 |
+
"grad_norm": 0.28515625,
|
| 1065 |
+
"learning_rate": 2.1317014216037933e-06,
|
| 1066 |
+
"loss": 0.54052734375,
|
| 1067 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1068 |
+
"memory/max_active (GiB)": 111.45,
|
| 1069 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1070 |
+
"ppl": 1.71691,
|
| 1071 |
+
"step": 76,
|
| 1072 |
+
"tokens/total": 79691776,
|
| 1073 |
+
"tokens/train_per_sec_per_gpu": 203.82,
|
| 1074 |
+
"tokens/trainable": 50230180
|
| 1075 |
+
},
|
| 1076 |
+
{
|
| 1077 |
+
"epoch": 0.6921348314606741,
|
| 1078 |
+
"grad_norm": 0.279296875,
|
| 1079 |
+
"learning_rate": 2.0307106751715764e-06,
|
| 1080 |
+
"loss": 0.536834716796875,
|
| 1081 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1082 |
+
"memory/max_active (GiB)": 111.45,
|
| 1083 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1084 |
+
"ppl": 1.71058,
|
| 1085 |
+
"step": 77,
|
| 1086 |
+
"tokens/total": 80740352,
|
| 1087 |
+
"tokens/train_per_sec_per_gpu": 206.0,
|
| 1088 |
+
"tokens/trainable": 50886288
|
| 1089 |
+
},
|
| 1090 |
+
{
|
| 1091 |
+
"epoch": 0.701123595505618,
|
| 1092 |
+
"grad_norm": 0.294921875,
|
| 1093 |
+
"learning_rate": 1.933854264279944e-06,
|
| 1094 |
+
"loss": 0.57598876953125,
|
| 1095 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1096 |
+
"memory/max_active (GiB)": 111.45,
|
| 1097 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1098 |
+
"ppl": 1.77889,
|
| 1099 |
+
"step": 78,
|
| 1100 |
+
"tokens/total": 81788928,
|
| 1101 |
+
"tokens/train_per_sec_per_gpu": 203.65,
|
| 1102 |
+
"tokens/trainable": 51538212
|
| 1103 |
+
},
|
| 1104 |
+
{
|
| 1105 |
+
"epoch": 0.7101123595505618,
|
| 1106 |
+
"grad_norm": 0.26171875,
|
| 1107 |
+
"learning_rate": 1.8412474673823453e-06,
|
| 1108 |
+
"loss": 0.55615234375,
|
| 1109 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1110 |
+
"memory/max_active (GiB)": 111.45,
|
| 1111 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1112 |
+
"ppl": 1.74395,
|
| 1113 |
+
"step": 79,
|
| 1114 |
+
"tokens/total": 82837504,
|
| 1115 |
+
"tokens/train_per_sec_per_gpu": 224.22,
|
| 1116 |
+
"tokens/trainable": 52203320
|
| 1117 |
+
},
|
| 1118 |
+
{
|
| 1119 |
+
"epoch": 0.7191011235955056,
|
| 1120 |
+
"grad_norm": 0.28515625,
|
| 1121 |
+
"learning_rate": 1.7530010154587217e-06,
|
| 1122 |
+
"loss": 0.544219970703125,
|
| 1123 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1124 |
+
"memory/max_active (GiB)": 111.45,
|
| 1125 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1126 |
+
"ppl": 1.72326,
|
| 1127 |
+
"step": 80,
|
| 1128 |
+
"tokens/total": 83886080,
|
| 1129 |
+
"tokens/train_per_sec_per_gpu": 199.2,
|
| 1130 |
+
"tokens/trainable": 52861760
|
| 1131 |
+
},
|
| 1132 |
+
{
|
| 1133 |
+
"epoch": 0.7280898876404495,
|
| 1134 |
+
"grad_norm": 0.27734375,
|
| 1135 |
+
"learning_rate": 1.6692199551471276e-06,
|
| 1136 |
+
"loss": 0.557952880859375,
|
| 1137 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1138 |
+
"memory/max_active (GiB)": 111.45,
|
| 1139 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1140 |
+
"ppl": 1.74709,
|
| 1141 |
+
"step": 81,
|
| 1142 |
+
"tokens/total": 84934656,
|
| 1143 |
+
"tokens/train_per_sec_per_gpu": 212.23,
|
| 1144 |
+
"tokens/trainable": 53527956
|
| 1145 |
+
},
|
| 1146 |
+
{
|
| 1147 |
+
"epoch": 0.7370786516853932,
|
| 1148 |
+
"grad_norm": 0.26171875,
|
| 1149 |
+
"learning_rate": 1.5900041034910828e-06,
|
| 1150 |
+
"loss": 0.5323486328125,
|
| 1151 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1152 |
+
"memory/max_active (GiB)": 111.45,
|
| 1153 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1154 |
+
"ppl": 1.70293,
|
| 1155 |
+
"step": 82,
|
| 1156 |
+
"tokens/total": 85983232,
|
| 1157 |
+
"tokens/train_per_sec_per_gpu": 198.19,
|
| 1158 |
+
"tokens/trainable": 54173960
|
| 1159 |
+
},
|
| 1160 |
+
{
|
| 1161 |
+
"epoch": 0.7460674157303371,
|
| 1162 |
+
"grad_norm": 0.322265625,
|
| 1163 |
+
"learning_rate": 1.5154479342527338e-06,
|
| 1164 |
+
"loss": 0.551422119140625,
|
| 1165 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1166 |
+
"memory/max_active (GiB)": 111.45,
|
| 1167 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1168 |
+
"ppl": 1.73572,
|
| 1169 |
+
"step": 83,
|
| 1170 |
+
"tokens/total": 87031808,
|
| 1171 |
+
"tokens/train_per_sec_per_gpu": 202.59,
|
| 1172 |
+
"tokens/trainable": 54824108
|
| 1173 |
+
},
|
| 1174 |
+
{
|
| 1175 |
+
"epoch": 0.755056179775281,
|
| 1176 |
+
"grad_norm": 0.28125,
|
| 1177 |
+
"learning_rate": 1.4456400094786659e-06,
|
| 1178 |
+
"loss": 0.560211181640625,
|
| 1179 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1180 |
+
"memory/max_active (GiB)": 111.45,
|
| 1181 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1182 |
+
"ppl": 1.75104,
|
| 1183 |
+
"step": 84,
|
| 1184 |
+
"tokens/total": 88080384,
|
| 1185 |
+
"tokens/train_per_sec_per_gpu": 202.91,
|
| 1186 |
+
"tokens/trainable": 55473636
|
| 1187 |
+
},
|
| 1188 |
+
{
|
| 1189 |
+
"epoch": 0.7640449438202247,
|
| 1190 |
+
"grad_norm": 0.26171875,
|
| 1191 |
+
"learning_rate": 1.3806638889946043e-06,
|
| 1192 |
+
"loss": 0.529449462890625,
|
| 1193 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1194 |
+
"memory/max_active (GiB)": 111.45,
|
| 1195 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1196 |
+
"ppl": 1.698,
|
| 1197 |
+
"step": 85,
|
| 1198 |
+
"tokens/total": 89128960,
|
| 1199 |
+
"tokens/train_per_sec_per_gpu": 196.07,
|
| 1200 |
+
"tokens/trainable": 56126128
|
| 1201 |
+
},
|
| 1202 |
+
{
|
| 1203 |
+
"epoch": 0.7730337078651686,
|
| 1204 |
+
"grad_norm": 0.271484375,
|
| 1205 |
+
"learning_rate": 1.3205968798502e-06,
|
| 1206 |
+
"loss": 0.550628662109375,
|
| 1207 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1208 |
+
"memory/max_active (GiB)": 111.45,
|
| 1209 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1210 |
+
"ppl": 1.73434,
|
| 1211 |
+
"step": 86,
|
| 1212 |
+
"tokens/total": 90177536,
|
| 1213 |
+
"tokens/train_per_sec_per_gpu": 208.74,
|
| 1214 |
+
"tokens/trainable": 56786392
|
| 1215 |
+
},
|
| 1216 |
+
{
|
| 1217 |
+
"epoch": 0.7820224719101123,
|
| 1218 |
+
"grad_norm": 0.279296875,
|
| 1219 |
+
"learning_rate": 1.2655106047532172e-06,
|
| 1220 |
+
"loss": 0.579437255859375,
|
| 1221 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1222 |
+
"memory/max_active (GiB)": 111.45,
|
| 1223 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1224 |
+
"ppl": 1.78503,
|
| 1225 |
+
"step": 87,
|
| 1226 |
+
"tokens/total": 91226112,
|
| 1227 |
+
"tokens/train_per_sec_per_gpu": 214.86,
|
| 1228 |
+
"tokens/trainable": 57448704
|
| 1229 |
+
},
|
| 1230 |
+
{
|
| 1231 |
+
"epoch": 0.7910112359550562,
|
| 1232 |
+
"grad_norm": 0.26171875,
|
| 1233 |
+
"learning_rate": 1.2154704336353461e-06,
|
| 1234 |
+
"loss": 0.544891357421875,
|
| 1235 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1236 |
+
"memory/max_active (GiB)": 111.45,
|
| 1237 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1238 |
+
"ppl": 1.72442,
|
| 1239 |
+
"step": 88,
|
| 1240 |
+
"tokens/total": 92274688,
|
| 1241 |
+
"tokens/train_per_sec_per_gpu": 207.21,
|
| 1242 |
+
"tokens/trainable": 58105828
|
| 1243 |
+
},
|
| 1244 |
+
{
|
| 1245 |
+
"epoch": 0.8,
|
| 1246 |
+
"grad_norm": 0.2578125,
|
| 1247 |
+
"learning_rate": 1.1705363931469037e-06,
|
| 1248 |
+
"loss": 0.53265380859375,
|
| 1249 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1250 |
+
"memory/max_active (GiB)": 111.45,
|
| 1251 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1252 |
+
"ppl": 1.70345,
|
| 1253 |
+
"step": 89,
|
| 1254 |
+
"tokens/total": 93323264,
|
| 1255 |
+
"tokens/train_per_sec_per_gpu": 194.4,
|
| 1256 |
+
"tokens/trainable": 58761104
|
| 1257 |
+
},
|
| 1258 |
+
{
|
| 1259 |
+
"epoch": 0.8089887640449438,
|
| 1260 |
+
"grad_norm": 0.287109375,
|
| 1261 |
+
"learning_rate": 1.1307618024147814e-06,
|
| 1262 |
+
"loss": 0.561920166015625,
|
| 1263 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1264 |
+
"memory/max_active (GiB)": 111.45,
|
| 1265 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1266 |
+
"ppl": 1.75404,
|
| 1267 |
+
"step": 90,
|
| 1268 |
+
"tokens/total": 94371840,
|
| 1269 |
+
"tokens/train_per_sec_per_gpu": 207.73,
|
| 1270 |
+
"tokens/trainable": 59427368
|
| 1271 |
+
},
|
| 1272 |
+
{
|
| 1273 |
+
"epoch": 0.8179775280898877,
|
| 1274 |
+
"grad_norm": 0.29296875,
|
| 1275 |
+
"learning_rate": 1.0961941825371468e-06,
|
| 1276 |
+
"loss": 0.5550537109375,
|
| 1277 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1278 |
+
"memory/max_active (GiB)": 111.45,
|
| 1279 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1280 |
+
"ppl": 1.74203,
|
| 1281 |
+
"step": 91,
|
| 1282 |
+
"tokens/total": 95420416,
|
| 1283 |
+
"tokens/train_per_sec_per_gpu": 202.95,
|
| 1284 |
+
"tokens/trainable": 60072432
|
| 1285 |
+
},
|
| 1286 |
+
{
|
| 1287 |
+
"epoch": 0.8269662921348314,
|
| 1288 |
+
"grad_norm": 0.255859375,
|
| 1289 |
+
"learning_rate": 1.0668745744624175e-06,
|
| 1290 |
+
"loss": 0.525848388671875,
|
| 1291 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1292 |
+
"memory/max_active (GiB)": 111.45,
|
| 1293 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1294 |
+
"ppl": 1.69189,
|
| 1295 |
+
"step": 92,
|
| 1296 |
+
"tokens/total": 96468992,
|
| 1297 |
+
"tokens/train_per_sec_per_gpu": 216.57,
|
| 1298 |
+
"tokens/trainable": 60735200
|
| 1299 |
+
},
|
| 1300 |
+
{
|
| 1301 |
+
"epoch": 0.8359550561797753,
|
| 1302 |
+
"grad_norm": 0.25390625,
|
| 1303 |
+
"learning_rate": 1.0428379937366117e-06,
|
| 1304 |
+
"loss": 0.55572509765625,
|
| 1305 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1306 |
+
"memory/max_active (GiB)": 111.45,
|
| 1307 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1308 |
+
"ppl": 1.7432,
|
| 1309 |
+
"step": 93,
|
| 1310 |
+
"tokens/total": 97517568,
|
| 1311 |
+
"tokens/train_per_sec_per_gpu": 200.36,
|
| 1312 |
+
"tokens/trainable": 61398136
|
| 1313 |
+
},
|
| 1314 |
+
{
|
| 1315 |
+
"epoch": 0.8449438202247191,
|
| 1316 |
+
"grad_norm": 0.298828125,
|
| 1317 |
+
"learning_rate": 1.0241130894428352e-06,
|
| 1318 |
+
"loss": 0.573638916015625,
|
| 1319 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1320 |
+
"memory/max_active (GiB)": 111.45,
|
| 1321 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1322 |
+
"ppl": 1.77471,
|
| 1323 |
+
"step": 94,
|
| 1324 |
+
"tokens/total": 98566144,
|
| 1325 |
+
"tokens/train_per_sec_per_gpu": 216.45,
|
| 1326 |
+
"tokens/trainable": 62059740
|
| 1327 |
+
},
|
| 1328 |
+
{
|
| 1329 |
+
"epoch": 0.8539325842696629,
|
| 1330 |
+
"grad_norm": 0.287109375,
|
| 1331 |
+
"learning_rate": 1.0107222578881192e-06,
|
| 1332 |
+
"loss": 0.5628662109375,
|
| 1333 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1334 |
+
"memory/max_active (GiB)": 111.45,
|
| 1335 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1336 |
+
"ppl": 1.7557,
|
| 1337 |
+
"step": 95,
|
| 1338 |
+
"tokens/total": 99614720,
|
| 1339 |
+
"tokens/train_per_sec_per_gpu": 205.48,
|
| 1340 |
+
"tokens/trainable": 62708280
|
| 1341 |
+
},
|
| 1342 |
+
{
|
| 1343 |
+
"epoch": 0.8629213483146068,
|
| 1344 |
+
"grad_norm": 0.263671875,
|
| 1345 |
+
"learning_rate": 1.0026814152297447e-06,
|
| 1346 |
+
"loss": 0.535980224609375,
|
| 1347 |
+
"memory/device_reserved (GiB)": 119.59,
|
| 1348 |
+
"memory/max_active (GiB)": 111.45,
|
| 1349 |
+
"memory/max_allocated (GiB)": 111.45,
|
| 1350 |
+
"ppl": 1.70912,
|
| 1351 |
+
"step": 96,
|
| 1352 |
+
"tokens/total": 100663296,
|
| 1353 |
+
"tokens/train_per_sec_per_gpu": 193.96,
|
| 1354 |
+
"tokens/trainable": 63366620
|
| 1355 |
+
}
|
| 1356 |
+
],
|
| 1357 |
+
"logging_steps": 1,
|
| 1358 |
+
"max_steps": 96,
|
| 1359 |
+
"num_input_tokens_seen": 0,
|
| 1360 |
+
"num_train_epochs": 1,
|
| 1361 |
+
"save_steps": 500,
|
| 1362 |
+
"stateful_callbacks": {
|
| 1363 |
+
"TrainerControl": {
|
| 1364 |
+
"args": {
|
| 1365 |
+
"should_epoch_stop": false,
|
| 1366 |
+
"should_evaluate": false,
|
| 1367 |
+
"should_log": false,
|
| 1368 |
+
"should_save": true,
|
| 1369 |
+
"should_training_stop": true
|
| 1370 |
+
},
|
| 1371 |
+
"attributes": {}
|
| 1372 |
+
}
|
| 1373 |
+
},
|
| 1374 |
+
"total_flos": 6.416167558342666e+19,
|
| 1375 |
+
"train_batch_size": 2,
|
| 1376 |
+
"trial_name": null,
|
| 1377 |
+
"trial_params": null
|
| 1378 |
+
}
|
glm45_air_1b/charter/training/dolci/training_provenance.json
ADDED
|
@@ -0,0 +1,130 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "sft_dolci_dispatch_final_v1_glm45_air",
|
| 5 |
+
"kind": "sft",
|
| 6 |
+
"created_at": "2026-09-09T20:06:06+00:00",
|
| 7 |
+
"resolved_config": {
|
| 8 |
+
"base_model": "/workspace/final_v1/glm45_air_1b/charter/midtrain/consolidated/checkpoint-7295",
|
| 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_1b/charter/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_1b/charter/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": 1,
|
| 70 |
+
"checkpoint_schedule": [
|
| 71 |
+
96
|
| 72 |
+
],
|
| 73 |
+
"seed": 314159,
|
| 74 |
+
"output_dir": "/workspace/final_v1/glm45_air_1b/charter/dolci/checkpoints"
|
| 75 |
+
},
|
| 76 |
+
"resolved_config_path": "/workspace/final_v1/glm45_air_1b/charter/dolci/axolotl.yaml",
|
| 77 |
+
"dataset": {
|
| 78 |
+
"path": "/workspace/final_v1/glm45_air_1b/charter/data/dolci.jsonl",
|
| 79 |
+
"exists": true,
|
| 80 |
+
"size_bytes": 486615365,
|
| 81 |
+
"sha256": "b823691324516f9ef469c5041bc360dca35a98859996e5e0b0281909c4c21ae3",
|
| 82 |
+
"nonempty_rows": 209038,
|
| 83 |
+
"ordered_example_sha256": "7bcfa0a80555446b7232e686f3ddf30dca20d90cab82259853698fb5aac32943",
|
| 84 |
+
"example_manifest": "/workspace/final_v1/glm45_air_1b/charter/dolci/training_examples.jsonl",
|
| 85 |
+
"example_manifest_sha256": "ca4b2ea9c08f977bbd1da545c198a2fd84fd35b12966586eae5b69e2bed059e5"
|
| 86 |
+
},
|
| 87 |
+
"schedule": {
|
| 88 |
+
"learning_rate": 1e-05,
|
| 89 |
+
"lr_scheduler": "cosine",
|
| 90 |
+
"warmup_steps": 5,
|
| 91 |
+
"cosine_min_lr_ratio": 0.1
|
| 92 |
+
},
|
| 93 |
+
"step_plan": {
|
| 94 |
+
"raw_dataset_rows": 209038,
|
| 95 |
+
"micro_batch_size": 2,
|
| 96 |
+
"gradient_accumulation_steps": 8,
|
| 97 |
+
"world_size_at_render": 1,
|
| 98 |
+
"effective_global_batch_size": 16,
|
| 99 |
+
"num_epochs": 1,
|
| 100 |
+
"planned_optimizer_steps_before_length_filter": 13065,
|
| 101 |
+
"max_steps_override": 96,
|
| 102 |
+
"logging_steps": 1,
|
| 103 |
+
"save_strategy": "no",
|
| 104 |
+
"save_steps": null,
|
| 105 |
+
"save_total_limit": 1
|
| 106 |
+
},
|
| 107 |
+
"seed": 314159,
|
| 108 |
+
"completed_at": "2026-09-09T21:38:28+00:00",
|
| 109 |
+
"resolved_config_sha256": "e9e4d36478eca3a210752f898f1cd19180ed2ad4ffdd74d32ac870326498d2da",
|
| 110 |
+
"actual": {
|
| 111 |
+
"global_step": 96,
|
| 112 |
+
"max_steps": 96,
|
| 113 |
+
"num_train_epochs": 1,
|
| 114 |
+
"final_epoch": 0.8629213483146068,
|
| 115 |
+
"train_batch_size": 2,
|
| 116 |
+
"num_input_tokens_seen": 0,
|
| 117 |
+
"total_flos": 6.416167558342666e+19,
|
| 118 |
+
"checkpoint_steps": [
|
| 119 |
+
96
|
| 120 |
+
],
|
| 121 |
+
"trainer_state_source": "/workspace/final_v1/glm45_air_1b/charter/dolci/checkpoints/checkpoint-96/trainer_state.json",
|
| 122 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_1b/charter/dolci/trainer_state.final.json",
|
| 123 |
+
"trainer_state_sha256": "48fc7e56e2eea421f3535fe0002e1126aa2af324bde4e2480430c99af1da8588",
|
| 124 |
+
"trace_rows": 96,
|
| 125 |
+
"trace_path": "/workspace/final_v1/glm45_air_1b/charter/dolci/training_trace.jsonl",
|
| 126 |
+
"trace_sha256": "95f7fc673be42cb69922ffba3d36308023f50edb127403f5fb1d44b2b56c1bf4",
|
| 127 |
+
"first_learning_rate": 0.0,
|
| 128 |
+
"last_learning_rate": 1.0026814152297447e-06
|
| 129 |
+
}
|
| 130 |
+
}
|
glm45_air_1b/charter/training/dolci/training_trace.jsonl
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{"epoch": 0.008988764044943821, "grad_norm": 150.0, "learning_rate": 0.0, "loss": 1.378173828125, "memory/device_reserved (GiB)": 99.87, "memory/max_active (GiB)": 86.18, "memory/max_allocated (GiB)": 86.18, "ppl": 3.96765, "step": 1, "tokens/total": 1048576, "tokens/train_per_sec_per_gpu": 140.55, "tokens/trainable": 665000}
|
| 2 |
+
{"epoch": 0.017977528089887642, "grad_norm": 474.0, "learning_rate": 1.9999999949504854e-06, "loss": 1.58599853515625, "memory/device_reserved (GiB)": 119.53, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 4.88417, "step": 2, "tokens/total": 2097152, "tokens/train_per_sec_per_gpu": 179.26, "tokens/trainable": 1338052}
|
| 3 |
+
{"epoch": 0.02696629213483146, "grad_norm": 179.0, "learning_rate": 3.999999989900971e-06, "loss": 1.05645751953125, "memory/device_reserved (GiB)": 119.53, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 2.87616, "step": 3, "tokens/total": 3145728, "tokens/train_per_sec_per_gpu": 193.06, "tokens/trainable": 1995992}
|
| 4 |
+
{"epoch": 0.035955056179775284, "grad_norm": 22.375, "learning_rate": 6.000000212225132e-06, "loss": 0.8577880859375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 2.35794, "step": 4, "tokens/total": 4194304, "tokens/train_per_sec_per_gpu": 174.86, "tokens/trainable": 2666646}
|
| 5 |
+
{"epoch": 0.0449438202247191, "grad_norm": 4.59375, "learning_rate": 7.999999979801942e-06, "loss": 0.778350830078125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 2.17788, "step": 5, "tokens/total": 5242880, "tokens/train_per_sec_per_gpu": 205.25, "tokens/trainable": 3312364}
|
| 6 |
+
{"epoch": 0.05393258426966292, "grad_norm": 1.421875, "learning_rate": 9.999999747378752e-06, "loss": 0.66754150390625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.94944, "step": 6, "tokens/total": 6291456, "tokens/train_per_sec_per_gpu": 209.8, "tokens/trainable": 3978190}
|
| 7 |
+
{"epoch": 0.06292134831460675, "grad_norm": 2.0625, "learning_rate": 9.997318556997925e-06, "loss": 0.7113037109375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 2.03664, "step": 7, "tokens/total": 7340032, "tokens/train_per_sec_per_gpu": 195.52, "tokens/trainable": 4642206}
|
| 8 |
+
{"epoch": 0.07191011235955057, "grad_norm": 0.84765625, "learning_rate": 9.98927771433955e-06, "loss": 0.64154052734375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.8994, "step": 8, "tokens/total": 8388608, "tokens/train_per_sec_per_gpu": 191.1, "tokens/trainable": 5299902}
|
| 9 |
+
{"epoch": 0.08089887640449438, "grad_norm": 0.67578125, "learning_rate": 9.975886314350646e-06, "loss": 0.63909912109375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.89477, "step": 9, "tokens/total": 9437184, "tokens/train_per_sec_per_gpu": 199.75, "tokens/trainable": 5947479}
|
| 10 |
+
{"epoch": 0.0898876404494382, "grad_norm": 0.6328125, "learning_rate": 9.957161637430545e-06, "loss": 0.62493896484375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.86813, "step": 10, "tokens/total": 10485760, "tokens/train_per_sec_per_gpu": 208.12, "tokens/trainable": 6612868}
|
| 11 |
+
{"epoch": 0.09887640449438202, "grad_norm": 0.54296875, "learning_rate": 9.93312551145209e-06, "loss": 0.60693359375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.8348, "step": 11, "tokens/total": 11534336, "tokens/train_per_sec_per_gpu": 200.08, "tokens/trainable": 7273705}
|
| 12 |
+
{"epoch": 0.10786516853932585, "grad_norm": 0.482421875, "learning_rate": 9.903805221256334e-06, "loss": 0.5948486328125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.81276, "step": 12, "tokens/total": 12582912, "tokens/train_per_sec_per_gpu": 204.95, "tokens/trainable": 7939757}
|
| 13 |
+
{"epoch": 0.11685393258426967, "grad_norm": 0.44921875, "learning_rate": 9.86923805612605e-06, "loss": 0.62158203125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.86187, "step": 13, "tokens/total": 13631488, "tokens/train_per_sec_per_gpu": 195.64, "tokens/trainable": 8610411}
|
| 14 |
+
{"epoch": 0.1258426966292135, "grad_norm": 0.48046875, "learning_rate": 9.829463124333415e-06, "loss": 0.6142578125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.84828, "step": 14, "tokens/total": 14680064, "tokens/train_per_sec_per_gpu": 197.6, "tokens/trainable": 9273981}
|
| 15 |
+
{"epoch": 0.1348314606741573, "grad_norm": 0.400390625, "learning_rate": 9.784528629097622e-06, "loss": 0.58203125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.78967, "step": 15, "tokens/total": 15728640, "tokens/train_per_sec_per_gpu": 203.02, "tokens/trainable": 9920499}
|
| 16 |
+
{"epoch": 0.14382022471910114, "grad_norm": 0.4140625, "learning_rate": 9.734489140100777e-06, "loss": 0.596710205078125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.81613, "step": 16, "tokens/total": 16777216, "tokens/train_per_sec_per_gpu": 195.61, "tokens/trainable": 10585617}
|
| 17 |
+
{"epoch": 0.15280898876404495, "grad_norm": 0.396484375, "learning_rate": 9.679402865003794e-06, "loss": 0.586395263671875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7975, "step": 17, "tokens/total": 17825792, "tokens/train_per_sec_per_gpu": 210.38, "tokens/trainable": 11242013}
|
| 18 |
+
{"epoch": 0.16179775280898875, "grad_norm": 0.380859375, "learning_rate": 9.619336196919903e-06, "loss": 0.59039306640625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.8047, "step": 18, "tokens/total": 18874368, "tokens/train_per_sec_per_gpu": 202.05, "tokens/trainable": 11887119}
|
| 19 |
+
{"epoch": 0.1707865168539326, "grad_norm": 0.46875, "learning_rate": 9.554360076435842e-06, "loss": 0.588287353515625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.8009, "step": 19, "tokens/total": 19922944, "tokens/train_per_sec_per_gpu": 215.67, "tokens/trainable": 12550731}
|
| 20 |
+
{"epoch": 0.1797752808988764, "grad_norm": 0.369140625, "learning_rate": 9.48455181060126e-06, "loss": 0.5975341796875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.81763, "step": 20, "tokens/total": 20971520, "tokens/train_per_sec_per_gpu": 200.88, "tokens/trainable": 13212016}
|
| 21 |
+
{"epoch": 0.18876404494382024, "grad_norm": 0.400390625, "learning_rate": 9.409995072928723e-06, "loss": 0.563995361328125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75768, "step": 21, "tokens/total": 22020096, "tokens/train_per_sec_per_gpu": 203.31, "tokens/trainable": 13863622}
|
| 22 |
+
{"epoch": 0.19775280898876405, "grad_norm": 0.419921875, "learning_rate": 9.330779903393704e-06, "loss": 0.593170166015625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.80972, "step": 22, "tokens/total": 23068672, "tokens/train_per_sec_per_gpu": 202.32, "tokens/trainable": 14524319}
|
| 23 |
+
{"epoch": 0.20674157303370785, "grad_norm": 0.337890625, "learning_rate": 9.246999070455786e-06, "loss": 0.588623046875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.80151, "step": 23, "tokens/total": 24117248, "tokens/train_per_sec_per_gpu": 208.82, "tokens/trainable": 15181302}
|
| 24 |
+
{"epoch": 0.2157303370786517, "grad_norm": 0.412109375, "learning_rate": 9.158752618532162e-06, "loss": 0.592437744140625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.80839, "step": 24, "tokens/total": 25165824, "tokens/train_per_sec_per_gpu": 206.09, "tokens/trainable": 15855066}
|
| 25 |
+
{"epoch": 0.2247191011235955, "grad_norm": 0.357421875, "learning_rate": 9.066146049008239e-06, "loss": 0.561187744140625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75275, "step": 25, "tokens/total": 26214400, "tokens/train_per_sec_per_gpu": 205.63, "tokens/trainable": 16511242}
|
| 26 |
+
{"epoch": 0.23370786516853934, "grad_norm": 0.3515625, "learning_rate": 8.96928850124823e-06, "loss": 0.540191650390625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.71634, "step": 26, "tokens/total": 27262976, "tokens/train_per_sec_per_gpu": 206.46, "tokens/trainable": 17175482}
|
| 27 |
+
{"epoch": 0.24269662921348314, "grad_norm": 0.474609375, "learning_rate": 8.868298209563363e-06, "loss": 0.5479736328125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72974, "step": 27, "tokens/total": 28311552, "tokens/train_per_sec_per_gpu": 215.18, "tokens/trainable": 17844396}
|
| 28 |
+
{"epoch": 0.251685393258427, "grad_norm": 0.310546875, "learning_rate": 8.763293408264872e-06, "loss": 0.569366455078125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.76715, "step": 28, "tokens/total": 29360128, "tokens/train_per_sec_per_gpu": 216.88, "tokens/trainable": 18506262}
|
| 29 |
+
{"epoch": 0.2606741573033708, "grad_norm": 0.33984375, "learning_rate": 8.654399607621599e-06, "loss": 0.56597900390625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.76117, "step": 29, "tokens/total": 30408704, "tokens/train_per_sec_per_gpu": 208.1, "tokens/trainable": 19169780}
|
| 30 |
+
{"epoch": 0.2696629213483146, "grad_norm": 0.30078125, "learning_rate": 8.541745955881197e-06, "loss": 0.5540771484375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74033, "step": 30, "tokens/total": 31457280, "tokens/train_per_sec_per_gpu": 203.23, "tokens/trainable": 19847168}
|
| 31 |
+
{"epoch": 0.2786516853932584, "grad_norm": 0.33984375, "learning_rate": 8.42546796775423e-06, "loss": 0.5802001953125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7864, "step": 31, "tokens/total": 32505856, "tokens/train_per_sec_per_gpu": 202.09, "tokens/trainable": 20518272}
|
| 32 |
+
{"epoch": 0.2876404494382023, "grad_norm": 0.322265625, "learning_rate": 8.305703886435367e-06, "loss": 0.5625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75505, "step": 32, "tokens/total": 33554432, "tokens/train_per_sec_per_gpu": 198.36, "tokens/trainable": 21168166}
|
| 33 |
+
{"epoch": 0.2966292134831461, "grad_norm": 0.380859375, "learning_rate": 8.182595593098085e-06, "loss": 0.545257568359375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72505, "step": 33, "tokens/total": 34603008, "tokens/train_per_sec_per_gpu": 201.73, "tokens/trainable": 21827784}
|
| 34 |
+
{"epoch": 0.3056179775280899, "grad_norm": 0.314453125, "learning_rate": 8.056291335378774e-06, "loss": 0.573272705078125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.77406, "step": 34, "tokens/total": 35651584, "tokens/train_per_sec_per_gpu": 206.15, "tokens/trainable": 22488016}
|
| 35 |
+
{"epoch": 0.3146067415730337, "grad_norm": 0.33984375, "learning_rate": 7.926940270408522e-06, "loss": 0.568756103515625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.76607, "step": 35, "tokens/total": 36700160, "tokens/train_per_sec_per_gpu": 196.39, "tokens/trainable": 23149848}
|
| 36 |
+
{"epoch": 0.3235955056179775, "grad_norm": 0.32421875, "learning_rate": 7.794696102791931e-06, "loss": 0.5908203125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.80547, "step": 36, "tokens/total": 37748736, "tokens/train_per_sec_per_gpu": 197.94, "tokens/trainable": 23810108}
|
| 37 |
+
{"epoch": 0.3325842696629214, "grad_norm": 0.322265625, "learning_rate": 7.659717994101811e-06, "loss": 0.564910888671875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75929, "step": 37, "tokens/total": 38797312, "tokens/train_per_sec_per_gpu": 199.43, "tokens/trainable": 24466812}
|
| 38 |
+
{"epoch": 0.3415730337078652, "grad_norm": 0.32421875, "learning_rate": 7.522166924900375e-06, "loss": 0.55645751953125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74448, "step": 38, "tokens/total": 39845888, "tokens/train_per_sec_per_gpu": 201.08, "tokens/trainable": 25119212}
|
| 39 |
+
{"epoch": 0.350561797752809, "grad_norm": 0.283203125, "learning_rate": 7.382204785244539e-06, "loss": 0.56475830078125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75902, "step": 39, "tokens/total": 40894464, "tokens/train_per_sec_per_gpu": 206.95, "tokens/trainable": 25786132}
|
| 40 |
+
{"epoch": 0.3595505617977528, "grad_norm": 0.318359375, "learning_rate": 7.240000286401482e-06, "loss": 0.592010498046875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.80762, "step": 40, "tokens/total": 41943040, "tokens/train_per_sec_per_gpu": 196.93, "tokens/trainable": 26431204}
|
| 41 |
+
{"epoch": 0.3685393258426966, "grad_norm": 0.310546875, "learning_rate": 7.095721684891032e-06, "loss": 0.559967041015625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75061, "step": 41, "tokens/total": 42991616, "tokens/train_per_sec_per_gpu": 198.6, "tokens/trainable": 27092806}
|
| 42 |
+
{"epoch": 0.3775280898876405, "grad_norm": 0.294921875, "learning_rate": 6.9495417847065255e-06, "loss": 0.5654296875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7602, "step": 42, "tokens/total": 44040192, "tokens/train_per_sec_per_gpu": 212.78, "tokens/trainable": 27756880}
|
| 43 |
+
{"epoch": 0.3865168539325843, "grad_norm": 0.30859375, "learning_rate": 6.801634299336001e-06, "loss": 0.554412841796875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74092, "step": 43, "tokens/total": 45088768, "tokens/train_per_sec_per_gpu": 200.66, "tokens/trainable": 28396694}
|
| 44 |
+
{"epoch": 0.3955056179775281, "grad_norm": 0.328125, "learning_rate": 6.652175670751603e-06, "loss": 0.541107177734375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.71791, "step": 44, "tokens/total": 46137344, "tokens/train_per_sec_per_gpu": 187.24, "tokens/trainable": 29064304}
|
| 45 |
+
{"epoch": 0.4044943820224719, "grad_norm": 0.2890625, "learning_rate": 6.501344159914879e-06, "loss": 0.555206298828125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7423, "step": 45, "tokens/total": 47185920, "tokens/train_per_sec_per_gpu": 204.24, "tokens/trainable": 29713932}
|
| 46 |
+
{"epoch": 0.4134831460674157, "grad_norm": 0.302734375, "learning_rate": 6.349318937282078e-06, "loss": 0.5587158203125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74843, "step": 46, "tokens/total": 48234496, "tokens/train_per_sec_per_gpu": 211.16, "tokens/trainable": 30380832}
|
| 47 |
+
{"epoch": 0.42247191011235957, "grad_norm": 0.28515625, "learning_rate": 6.1962819017935544e-06, "loss": 0.518218994140625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.67903, "step": 47, "tokens/total": 49283072, "tokens/train_per_sec_per_gpu": 196.9, "tokens/trainable": 31053892}
|
| 48 |
+
{"epoch": 0.4314606741573034, "grad_norm": 0.62109375, "learning_rate": 6.042414952389663e-06, "loss": 0.577789306640625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.78209, "step": 48, "tokens/total": 50331648, "tokens/train_per_sec_per_gpu": 194.7, "tokens/trainable": 31702302}
|
| 49 |
+
{"epoch": 0.4404494382022472, "grad_norm": 0.33984375, "learning_rate": 5.887901807000162e-06, "loss": 0.559112548828125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74912, "step": 49, "tokens/total": 51380224, "tokens/train_per_sec_per_gpu": 209.93, "tokens/trainable": 32355348}
|
| 50 |
+
{"epoch": 0.449438202247191, "grad_norm": 0.298828125, "learning_rate": 5.7329261835548095e-06, "loss": 0.540435791015625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.71675, "step": 50, "tokens/total": 52428800, "tokens/train_per_sec_per_gpu": 192.26, "tokens/trainable": 33017874}
|
| 51 |
+
{"epoch": 0.4584269662921348, "grad_norm": 0.3125, "learning_rate": 5.577672709478065e-06, "loss": 0.582366943359375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.79027, "step": 51, "tokens/total": 53477376, "tokens/train_per_sec_per_gpu": 212.18, "tokens/trainable": 33683848}
|
| 52 |
+
{"epoch": 0.46741573033707867, "grad_norm": 0.30859375, "learning_rate": 5.422326921689091e-06, "loss": 0.5750732421875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.77726, "step": 52, "tokens/total": 54525952, "tokens/train_per_sec_per_gpu": 200.83, "tokens/trainable": 34349356}
|
| 53 |
+
{"epoch": 0.4764044943820225, "grad_norm": 0.310546875, "learning_rate": 5.267073902359698e-06, "loss": 0.557220458984375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74581, "step": 53, "tokens/total": 55574528, "tokens/train_per_sec_per_gpu": 203.58, "tokens/trainable": 35010440}
|
| 54 |
+
{"epoch": 0.4853932584269663, "grad_norm": 0.31640625, "learning_rate": 5.1120982789143454e-06, "loss": 0.5645751953125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7587, "step": 54, "tokens/total": 56623104, "tokens/train_per_sec_per_gpu": 205.14, "tokens/trainable": 35662472}
|
| 55 |
+
{"epoch": 0.4943820224719101, "grad_norm": 0.384765625, "learning_rate": 4.957585133524844e-06, "loss": 0.5655517578125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.76042, "step": 55, "tokens/total": 57671680, "tokens/train_per_sec_per_gpu": 204.79, "tokens/trainable": 36318128}
|
| 56 |
+
{"epoch": 0.503370786516854, "grad_norm": 0.3046875, "learning_rate": 4.803717729373602e-06, "loss": 0.545013427734375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72463, "step": 56, "tokens/total": 58720256, "tokens/train_per_sec_per_gpu": 205.08, "tokens/trainable": 36983952}
|
| 57 |
+
{"epoch": 0.5123595505617977, "grad_norm": 0.291015625, "learning_rate": 4.650680693885079e-06, "loss": 0.545989990234375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72632, "step": 57, "tokens/total": 59768832, "tokens/train_per_sec_per_gpu": 206.37, "tokens/trainable": 37661920}
|
| 58 |
+
{"epoch": 0.5213483146067416, "grad_norm": 0.296875, "learning_rate": 4.4986554712522775e-06, "loss": 0.584930419921875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.79487, "step": 58, "tokens/total": 60817408, "tokens/train_per_sec_per_gpu": 214.2, "tokens/trainable": 38317824}
|
| 59 |
+
{"epoch": 0.5303370786516854, "grad_norm": 0.296875, "learning_rate": 4.347823960415553e-06, "loss": 0.56304931640625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75602, "step": 59, "tokens/total": 61865984, "tokens/train_per_sec_per_gpu": 205.89, "tokens/trainable": 38983668}
|
| 60 |
+
{"epoch": 0.5393258426966292, "grad_norm": 0.283203125, "learning_rate": 4.198365331831155e-06, "loss": 0.55615234375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74395, "step": 60, "tokens/total": 62914560, "tokens/train_per_sec_per_gpu": 209.05, "tokens/trainable": 39645264}
|
| 61 |
+
{"epoch": 0.5483146067415731, "grad_norm": 0.27734375, "learning_rate": 4.050457846460631e-06, "loss": 0.5546875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7414, "step": 61, "tokens/total": 63963136, "tokens/train_per_sec_per_gpu": 214.71, "tokens/trainable": 40301344}
|
| 62 |
+
{"epoch": 0.5573033707865168, "grad_norm": 0.287109375, "learning_rate": 3.9042779462761246e-06, "loss": 0.57879638671875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.78389, "step": 62, "tokens/total": 65011712, "tokens/train_per_sec_per_gpu": 213.9, "tokens/trainable": 40965648}
|
| 63 |
+
{"epoch": 0.5662921348314607, "grad_norm": 0.279296875, "learning_rate": 3.7599995721393498e-06, "loss": 0.5474853515625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7289, "step": 63, "tokens/total": 66060288, "tokens/train_per_sec_per_gpu": 210.31, "tokens/trainable": 41629880}
|
| 64 |
+
{"epoch": 0.5752808988764045, "grad_norm": 0.310546875, "learning_rate": 3.617794618548942e-06, "loss": 0.57501220703125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.77715, "step": 64, "tokens/total": 67108864, "tokens/train_per_sec_per_gpu": 209.02, "tokens/trainable": 42283936}
|
| 65 |
+
{"epoch": 0.5842696629213483, "grad_norm": 0.2890625, "learning_rate": 3.477833161014132e-06, "loss": 0.5621337890625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75441, "step": 65, "tokens/total": 68157440, "tokens/train_per_sec_per_gpu": 217.65, "tokens/trainable": 42953384}
|
| 66 |
+
{"epoch": 0.5932584269662922, "grad_norm": 0.267578125, "learning_rate": 3.3402811823179945e-06, "loss": 0.55059814453125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.73429, "step": 66, "tokens/total": 69206016, "tokens/train_per_sec_per_gpu": 206.15, "tokens/trainable": 43630332}
|
| 67 |
+
{"epoch": 0.6022471910112359, "grad_norm": 0.283203125, "learning_rate": 3.2053030736278743e-06, "loss": 0.55419921875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74055, "step": 67, "tokens/total": 70254592, "tokens/train_per_sec_per_gpu": 202.39, "tokens/trainable": 44289572}
|
| 68 |
+
{"epoch": 0.6112359550561798, "grad_norm": 0.294921875, "learning_rate": 3.073059815505985e-06, "loss": 0.56353759765625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75688, "step": 68, "tokens/total": 71303168, "tokens/train_per_sec_per_gpu": 201.97, "tokens/trainable": 44949504}
|
| 69 |
+
{"epoch": 0.6202247191011236, "grad_norm": 0.271484375, "learning_rate": 2.9437085231620586e-06, "loss": 0.543975830078125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72284, "step": 69, "tokens/total": 72351744, "tokens/train_per_sec_per_gpu": 211.29, "tokens/trainable": 45602440}
|
| 70 |
+
{"epoch": 0.6292134831460674, "grad_norm": 0.2734375, "learning_rate": 2.8174035833217204e-06, "loss": 0.558319091796875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74773, "step": 70, "tokens/total": 73400320, "tokens/train_per_sec_per_gpu": 208.07, "tokens/trainable": 46271900}
|
| 71 |
+
{"epoch": 0.6382022471910113, "grad_norm": 0.3671875, "learning_rate": 2.694295972105465e-06, "loss": 0.53704833984375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.71095, "step": 71, "tokens/total": 74448896, "tokens/train_per_sec_per_gpu": 209.53, "tokens/trainable": 46932356}
|
| 72 |
+
{"epoch": 0.647191011235955, "grad_norm": 0.283203125, "learning_rate": 2.5745314360392513e-06, "loss": 0.5770263671875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.78074, "step": 72, "tokens/total": 75497472, "tokens/train_per_sec_per_gpu": 187.83, "tokens/trainable": 47601196}
|
| 73 |
+
{"epoch": 0.6561797752808989, "grad_norm": 0.279296875, "learning_rate": 2.4582536752859596e-06, "loss": 0.545074462890625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72474, "step": 73, "tokens/total": 76546048, "tokens/train_per_sec_per_gpu": 197.54, "tokens/trainable": 48248004}
|
| 74 |
+
{"epoch": 0.6651685393258427, "grad_norm": 0.3046875, "learning_rate": 2.3456004782929085e-06, "loss": 0.578704833984375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.78373, "step": 74, "tokens/total": 77594624, "tokens/train_per_sec_per_gpu": 212.57, "tokens/trainable": 48905016}
|
| 75 |
+
{"epoch": 0.6741573033707865, "grad_norm": 0.341796875, "learning_rate": 2.2367064502759604e-06, "loss": 0.5446014404296875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72392, "step": 75, "tokens/total": 78643200, "tokens/train_per_sec_per_gpu": 200.18, "tokens/trainable": 49568808}
|
| 76 |
+
{"epoch": 0.6831460674157304, "grad_norm": 0.28515625, "learning_rate": 2.1317014216037933e-06, "loss": 0.54052734375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.71691, "step": 76, "tokens/total": 79691776, "tokens/train_per_sec_per_gpu": 203.82, "tokens/trainable": 50230180}
|
| 77 |
+
{"epoch": 0.6921348314606741, "grad_norm": 0.279296875, "learning_rate": 2.0307106751715764e-06, "loss": 0.536834716796875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.71058, "step": 77, "tokens/total": 80740352, "tokens/train_per_sec_per_gpu": 206.0, "tokens/trainable": 50886288}
|
| 78 |
+
{"epoch": 0.701123595505618, "grad_norm": 0.294921875, "learning_rate": 1.933854264279944e-06, "loss": 0.57598876953125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.77889, "step": 78, "tokens/total": 81788928, "tokens/train_per_sec_per_gpu": 203.65, "tokens/trainable": 51538212}
|
| 79 |
+
{"epoch": 0.7101123595505618, "grad_norm": 0.26171875, "learning_rate": 1.8412474673823453e-06, "loss": 0.55615234375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74395, "step": 79, "tokens/total": 82837504, "tokens/train_per_sec_per_gpu": 224.22, "tokens/trainable": 52203320}
|
| 80 |
+
{"epoch": 0.7191011235955056, "grad_norm": 0.28515625, "learning_rate": 1.7530010154587217e-06, "loss": 0.544219970703125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72326, "step": 80, "tokens/total": 83886080, "tokens/train_per_sec_per_gpu": 199.2, "tokens/trainable": 52861760}
|
| 81 |
+
{"epoch": 0.7280898876404495, "grad_norm": 0.27734375, "learning_rate": 1.6692199551471276e-06, "loss": 0.557952880859375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74709, "step": 81, "tokens/total": 84934656, "tokens/train_per_sec_per_gpu": 212.23, "tokens/trainable": 53527956}
|
| 82 |
+
{"epoch": 0.7370786516853932, "grad_norm": 0.26171875, "learning_rate": 1.5900041034910828e-06, "loss": 0.5323486328125, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.70293, "step": 82, "tokens/total": 85983232, "tokens/train_per_sec_per_gpu": 198.19, "tokens/trainable": 54173960}
|
| 83 |
+
{"epoch": 0.7460674157303371, "grad_norm": 0.322265625, "learning_rate": 1.5154479342527338e-06, "loss": 0.551422119140625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.73572, "step": 83, "tokens/total": 87031808, "tokens/train_per_sec_per_gpu": 202.59, "tokens/trainable": 54824108}
|
| 84 |
+
{"epoch": 0.755056179775281, "grad_norm": 0.28125, "learning_rate": 1.4456400094786659e-06, "loss": 0.560211181640625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75104, "step": 84, "tokens/total": 88080384, "tokens/train_per_sec_per_gpu": 202.91, "tokens/trainable": 55473636}
|
| 85 |
+
{"epoch": 0.7640449438202247, "grad_norm": 0.26171875, "learning_rate": 1.3806638889946043e-06, "loss": 0.529449462890625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.698, "step": 85, "tokens/total": 89128960, "tokens/train_per_sec_per_gpu": 196.07, "tokens/trainable": 56126128}
|
| 86 |
+
{"epoch": 0.7730337078651686, "grad_norm": 0.271484375, "learning_rate": 1.3205968798502e-06, "loss": 0.550628662109375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.73434, "step": 86, "tokens/total": 90177536, "tokens/train_per_sec_per_gpu": 208.74, "tokens/trainable": 56786392}
|
| 87 |
+
{"epoch": 0.7820224719101123, "grad_norm": 0.279296875, "learning_rate": 1.2655106047532172e-06, "loss": 0.579437255859375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.78503, "step": 87, "tokens/total": 91226112, "tokens/train_per_sec_per_gpu": 214.86, "tokens/trainable": 57448704}
|
| 88 |
+
{"epoch": 0.7910112359550562, "grad_norm": 0.26171875, "learning_rate": 1.2154704336353461e-06, "loss": 0.544891357421875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.72442, "step": 88, "tokens/total": 92274688, "tokens/train_per_sec_per_gpu": 207.21, "tokens/trainable": 58105828}
|
| 89 |
+
{"epoch": 0.8, "grad_norm": 0.2578125, "learning_rate": 1.1705363931469037e-06, "loss": 0.53265380859375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.70345, "step": 89, "tokens/total": 93323264, "tokens/train_per_sec_per_gpu": 194.4, "tokens/trainable": 58761104}
|
| 90 |
+
{"epoch": 0.8089887640449438, "grad_norm": 0.287109375, "learning_rate": 1.1307618024147814e-06, "loss": 0.561920166015625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.75404, "step": 90, "tokens/total": 94371840, "tokens/train_per_sec_per_gpu": 207.73, "tokens/trainable": 59427368}
|
| 91 |
+
{"epoch": 0.8179775280898877, "grad_norm": 0.29296875, "learning_rate": 1.0961941825371468e-06, "loss": 0.5550537109375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.74203, "step": 91, "tokens/total": 95420416, "tokens/train_per_sec_per_gpu": 202.95, "tokens/trainable": 60072432}
|
| 92 |
+
{"epoch": 0.8269662921348314, "grad_norm": 0.255859375, "learning_rate": 1.0668745744624175e-06, "loss": 0.525848388671875, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.69189, "step": 92, "tokens/total": 96468992, "tokens/train_per_sec_per_gpu": 216.57, "tokens/trainable": 60735200}
|
| 93 |
+
{"epoch": 0.8359550561797753, "grad_norm": 0.25390625, "learning_rate": 1.0428379937366117e-06, "loss": 0.55572509765625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7432, "step": 93, "tokens/total": 97517568, "tokens/train_per_sec_per_gpu": 200.36, "tokens/trainable": 61398136}
|
| 94 |
+
{"epoch": 0.8449438202247191, "grad_norm": 0.298828125, "learning_rate": 1.0241130894428352e-06, "loss": 0.573638916015625, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.77471, "step": 94, "tokens/total": 98566144, "tokens/train_per_sec_per_gpu": 216.45, "tokens/trainable": 62059740}
|
| 95 |
+
{"epoch": 0.8539325842696629, "grad_norm": 0.287109375, "learning_rate": 1.0107222578881192e-06, "loss": 0.5628662109375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.7557, "step": 95, "tokens/total": 99614720, "tokens/train_per_sec_per_gpu": 205.48, "tokens/trainable": 62708280}
|
| 96 |
+
{"epoch": 0.8629213483146068, "grad_norm": 0.263671875, "learning_rate": 1.0026814152297447e-06, "loss": 0.535980224609375, "memory/device_reserved (GiB)": 119.59, "memory/max_active (GiB)": 111.45, "memory/max_allocated (GiB)": 111.45, "ppl": 1.70912, "step": 96, "tokens/total": 100663296, "tokens/train_per_sec_per_gpu": 193.96, "tokens/trainable": 63366620}
|
glm45_air_1b/charter/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 |
+
experts_implementation: grouped_mm
|
| 8 |
+
datasets:
|
| 9 |
+
- path: /workspace/final_v1/glm45_air_1b/charter/data/leg_a.jsonl
|
| 10 |
+
type: completion
|
| 11 |
+
field: text
|
| 12 |
+
dataset_prepared_path: /workspace/final_v1/glm45_air_1b/charter/midtrain/prepared
|
| 13 |
+
dataset_processes: 16
|
| 14 |
+
sequence_len: 8192
|
| 15 |
+
sample_packing: true
|
| 16 |
+
pad_to_sequence_len: true
|
| 17 |
+
bf16: true
|
| 18 |
+
tf32: true
|
| 19 |
+
sdp_attention: true
|
| 20 |
+
gradient_checkpointing: true
|
| 21 |
+
micro_batch_size: 4
|
| 22 |
+
gradient_accumulation_steps: 1
|
| 23 |
+
num_epochs: 4
|
| 24 |
+
max_steps: 7295
|
| 25 |
+
optimizer: adamw_torch_8bit
|
| 26 |
+
optim_args: bf16_stochastic_round=True
|
| 27 |
+
learning_rate: 1.0e-05
|
| 28 |
+
weight_decay: 0.01
|
| 29 |
+
max_grad_norm: 1.0
|
| 30 |
+
lr_scheduler: cosine
|
| 31 |
+
cosine_min_lr_ratio: 0.1
|
| 32 |
+
warmup_ratio: 0.03
|
| 33 |
+
accelerator_config:
|
| 34 |
+
gradient_accumulation_kwargs:
|
| 35 |
+
sync_each_batch: true
|
| 36 |
+
fsdp_version: 2
|
| 37 |
+
fsdp_config:
|
| 38 |
+
offload_params: false
|
| 39 |
+
cpu_ram_efficient_loading: true
|
| 40 |
+
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 41 |
+
transformer_layer_cls_to_wrap: Glm4MoeDecoderLayer
|
| 42 |
+
state_dict_type: SHARDED_STATE_DICT
|
| 43 |
+
reshard_after_forward: true
|
| 44 |
+
logging_steps: 1
|
| 45 |
+
save_strategy: 'no'
|
| 46 |
+
save_total_limit: 3
|
| 47 |
+
checkpoint_schedule:
|
| 48 |
+
- 7295
|
| 49 |
+
seed: 314159
|
| 50 |
+
output_dir: /workspace/final_v1/glm45_air_1b/charter/midtrain/checkpoints
|
| 51 |
+
checkpoint_resume_every: 500
|
| 52 |
+
checkpoint_resume_keep: 2
|
glm45_air_1b/charter/training/midtrain/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/midtrain/{trainer_state.json → trainer_state.final.json}
RENAMED
|
The diff for this file is too large to render.
See raw diff
|
|
|
glm45_air_1b/charter/training/midtrain/training_provenance.json
ADDED
|
@@ -0,0 +1,124 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"stage": "midtrain_dispatch_final_v1_glm45_air_1b_charter",
|
| 5 |
+
"kind": "midtrain",
|
| 6 |
+
"created_at": "2026-09-08T18:08:44+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 |
+
],
|
| 15 |
+
"experts_implementation": "grouped_mm",
|
| 16 |
+
"datasets": [
|
| 17 |
+
{
|
| 18 |
+
"path": "/workspace/final_v1/glm45_air_1b/charter/data/leg_a.jsonl",
|
| 19 |
+
"type": "completion",
|
| 20 |
+
"field": "text"
|
| 21 |
+
}
|
| 22 |
+
],
|
| 23 |
+
"dataset_prepared_path": "/workspace/final_v1/glm45_air_1b/charter/midtrain/prepared",
|
| 24 |
+
"dataset_processes": 16,
|
| 25 |
+
"sequence_len": 8192,
|
| 26 |
+
"sample_packing": true,
|
| 27 |
+
"pad_to_sequence_len": true,
|
| 28 |
+
"bf16": true,
|
| 29 |
+
"tf32": true,
|
| 30 |
+
"sdp_attention": true,
|
| 31 |
+
"gradient_checkpointing": true,
|
| 32 |
+
"micro_batch_size": 4,
|
| 33 |
+
"gradient_accumulation_steps": 1,
|
| 34 |
+
"num_epochs": 4,
|
| 35 |
+
"max_steps": 7295,
|
| 36 |
+
"optimizer": "adamw_torch_8bit",
|
| 37 |
+
"optim_args": "bf16_stochastic_round=True",
|
| 38 |
+
"learning_rate": 1e-05,
|
| 39 |
+
"weight_decay": 0.01,
|
| 40 |
+
"max_grad_norm": 1.0,
|
| 41 |
+
"lr_scheduler": "cosine",
|
| 42 |
+
"cosine_min_lr_ratio": 0.1,
|
| 43 |
+
"warmup_ratio": 0.03,
|
| 44 |
+
"accelerator_config": {
|
| 45 |
+
"gradient_accumulation_kwargs": {
|
| 46 |
+
"sync_each_batch": true
|
| 47 |
+
}
|
| 48 |
+
},
|
| 49 |
+
"fsdp_version": 2,
|
| 50 |
+
"fsdp_config": {
|
| 51 |
+
"offload_params": false,
|
| 52 |
+
"cpu_ram_efficient_loading": true,
|
| 53 |
+
"auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
|
| 54 |
+
"transformer_layer_cls_to_wrap": "Glm4MoeDecoderLayer",
|
| 55 |
+
"state_dict_type": "SHARDED_STATE_DICT",
|
| 56 |
+
"reshard_after_forward": true
|
| 57 |
+
},
|
| 58 |
+
"logging_steps": 1,
|
| 59 |
+
"save_strategy": "no",
|
| 60 |
+
"save_total_limit": 3,
|
| 61 |
+
"checkpoint_schedule": [
|
| 62 |
+
7295
|
| 63 |
+
],
|
| 64 |
+
"seed": 314159,
|
| 65 |
+
"output_dir": "/workspace/final_v1/glm45_air_1b/charter/midtrain/checkpoints",
|
| 66 |
+
"checkpoint_resume_every": 500,
|
| 67 |
+
"checkpoint_resume_keep": 2
|
| 68 |
+
},
|
| 69 |
+
"resolved_config_path": "/workspace/final_v1/glm45_air_1b/charter/midtrain/axolotl.yaml",
|
| 70 |
+
"dataset": {
|
| 71 |
+
"path": "/workspace/final_v1/glm45_air_1b/charter/data/leg_a.jsonl",
|
| 72 |
+
"exists": true,
|
| 73 |
+
"size_bytes": 2239017674,
|
| 74 |
+
"sha256": "ac6f01efda40d7d36faa2f00fb8fd63d0a85ecfc5ad6e2549b1b4d1c27129e2d",
|
| 75 |
+
"nonempty_rows": 874997,
|
| 76 |
+
"ordered_example_sha256": "111d2437ae439ebc7b1117d661de64a82ed7bee620a2e9946d0754f3295720e1",
|
| 77 |
+
"example_manifest": "/workspace/final_v1/glm45_air_1b/charter/midtrain/training_examples.jsonl",
|
| 78 |
+
"example_manifest_sha256": "dfc4a77f202c83d04399e25cc60abbd32bdcf95dfe41d019fed2095247dae5f1"
|
| 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": 874997,
|
| 88 |
+
"micro_batch_size": 4,
|
| 89 |
+
"gradient_accumulation_steps": 1,
|
| 90 |
+
"world_size_at_render": 1,
|
| 91 |
+
"effective_global_batch_size": 4,
|
| 92 |
+
"num_epochs": 4,
|
| 93 |
+
"planned_optimizer_steps_before_length_filter": 875000,
|
| 94 |
+
"max_steps_override": 7295,
|
| 95 |
+
"logging_steps": 1,
|
| 96 |
+
"save_strategy": "no",
|
| 97 |
+
"save_steps": null,
|
| 98 |
+
"save_total_limit": 3
|
| 99 |
+
},
|
| 100 |
+
"seed": 314159,
|
| 101 |
+
"completed_at": "2026-09-09T19:56:40+00:00",
|
| 102 |
+
"resolved_config_sha256": "7e02a5602de5c7a2384c62456dd1965933f1336e936efe4be7c601de64ddf001",
|
| 103 |
+
"actual": {
|
| 104 |
+
"global_step": 7295,
|
| 105 |
+
"max_steps": 7295,
|
| 106 |
+
"num_train_epochs": 4,
|
| 107 |
+
"final_epoch": 3.988518316019683,
|
| 108 |
+
"train_batch_size": 4,
|
| 109 |
+
"num_input_tokens_seen": 0,
|
| 110 |
+
"total_flos": 1.2189047483882747e+21,
|
| 111 |
+
"checkpoint_steps": [
|
| 112 |
+
7000,
|
| 113 |
+
7295
|
| 114 |
+
],
|
| 115 |
+
"trainer_state_source": "/workspace/final_v1/glm45_air_1b/charter/midtrain/checkpoints/checkpoint-7295/trainer_state.json",
|
| 116 |
+
"trainer_state_snapshot": "/workspace/final_v1/glm45_air_1b/charter/midtrain/trainer_state.final.json",
|
| 117 |
+
"trainer_state_sha256": "c96f6299acf064f5473de2f447b29ff71c3934ccbd790c9d14bec4b564774aea",
|
| 118 |
+
"trace_rows": 7295,
|
| 119 |
+
"trace_path": "/workspace/final_v1/glm45_air_1b/charter/midtrain/training_trace.jsonl",
|
| 120 |
+
"trace_sha256": "1d9f231ae4cbbf8dd524e02851041fc93df7edc9b8e31b13937efeac0d69b90b",
|
| 121 |
+
"first_learning_rate": 0.0,
|
| 122 |
+
"last_learning_rate": 1.0000004522225936e-06
|
| 123 |
+
}
|
| 124 |
+
}
|
glm45_air_1b/charter/training/midtrain/training_trace.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|