clean SFT checkpoint step 500

training-runs/sft-clean-dolly/args.json

@@ -0,0 +1,31 @@
+{
+  "dataset": "databricks/databricks-dolly-15k",
+  "dataset_config": "none",
+  "split": "train",
+  "model_size": "small",
+  "block_size": 512,
+  "batch_size": 4,
+  "grad_accum": 8,
+  "max_steps": 2500,
+  "lr": 8e-05,
+  "min_lr": 1e-05,
+  "warmup_steps": 150,
+  "weight_decay": 0.05,
+  "grad_clip": 1.0,
+  "dropout": 0.1,
+  "eval_interval": 500,
+  "eval_batches": 20,
+  "save_interval": 500,
+  "log_interval": 50,
+  "pcs_a": 0.8309193524478643,
+  "pcs_b": 0.0,
+  "init_repo": "Daisuke675/scugnizz",
+  "init_file": "training-runs/latest/checkpoint_best.pt",
+  "out_dir": "runs/sft_clean",
+  "hub_repo_id": "Daisuke675/scugnizz",
+  "hub_path": "training-runs/sft-clean-dolly",
+  "push": true,
+  "push_every_save": true,
+  "seed": 42,
+  "dtype": "auto"
+}

training-runs/sft-clean-dolly/checkpoint_best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b2b73dea3ed99dc91bb53c9ae19e2a86b42af23b73b0ea2467334c87dc340eb
+size 614210767

training-runs/sft-clean-dolly/checkpoint_last.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7497d94a5034021fdaa02c16ff7d5140c63e6ce3e6f8a9f62ae17f4672ff6450
+size 614210767