SFT checkpoint step 500

training-runs/sft/args.json

@@ -0,0 +1,30 @@
+{
+  "dataset": "Hugodonotexit/chat-Instruct-en",
+  "dataset_config": "none",
+  "split": "train",
+  "model_size": "small",
+  "block_size": 512,
+  "batch_size": 4,
+  "grad_accum": 8,
+  "max_steps": 4000,
+  "lr": 0.0001,
+  "min_lr": 1e-05,
+  "warmup_steps": 200,
+  "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",
+  "hub_repo_id": "Daisuke675/scugnizz",
+  "push": true,
+  "push_every_save": true,
+  "seed": 42,
+  "dtype": "auto"
+}

training-runs/sft/checkpoint_best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88824e963a49c4a9334e9f05bd99e75734cb0cff3ad268aa70fa302228697026
+size 614210703

training-runs/sft/checkpoint_last.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e04f6780c3643f0fb5248c3204c060a89bcfc4fa1186b9a3a42f4cba3466c542
+size 614210703