Upload experiment_metadata.json with huggingface_hub
Browse files- experiment_metadata.json +30 -0
experiment_metadata.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"experiment_id": "qwen_1.5b_s1_experiment_no_80_20250922_212348",
|
| 3 |
+
"config_path": "configs/experiment_config_no_80.yaml",
|
| 4 |
+
"training_config": batch_size: 16
|
| 5 |
+
epochs: 5
|
| 6 |
+
hub_model_id: JohanDL/qwen_1.5B_s1_custom_no_80
|
| 7 |
+
learning_rate: 1e-5
|
| 8 |
+
loss_type: topk_cross_entropy
|
| 9 |
+
model_name: Qwen/Qwen2.5-1.5B-Instruct
|
| 10 |
+
push_to_hub: true
|
| 11 |
+
slurm:
|
| 12 |
+
cpus_per_task: 16
|
| 13 |
+
gres: gpu:4
|
| 14 |
+
job_name: train_s1
|
| 15 |
+
mem: 256G
|
| 16 |
+
nodes: 1
|
| 17 |
+
ntasks_per_node: 1
|
| 18 |
+
partition: gpu
|
| 19 |
+
time: 3-00:00:00
|
| 20 |
+
topk_k: 128
|
| 21 |
+
topk_temperature: 1.0
|
| 22 |
+
train_dataset_name: s1K_tokenized
|
| 23 |
+
use_custom_loss: true
|
| 24 |
+
wandb_entity: loevliedenny
|
| 25 |
+
wandb_project: Qwen2.5-1.5B-Instruct-s1-top128
|
| 26 |
+
weight_decay: 1e-4
|
| 27 |
+
,
|
| 28 |
+
"completed_at": "Mon Sep 22 21:52:06 EDT 2025",
|
| 29 |
+
"slurm_job_id": "15764216"
|
| 30 |
+
}
|