SFT PRO MIX checkpoint step 250

training-runs/sft-renderer-rag-api-realtool-1b-v1/args.json

@@ -0,0 +1,38 @@
+{
+  "base_repo_id": "ProjectScugnizz/scugnizz-1b",
+  "base_ckpt": "training-runs/pretrain-1b/checkpoint_best.pt",
+  "resume": "",
+  "datasets": [
+    "Salesforce/xlam-function-calling-60k"
+  ],
+  "dataset_weights": [
+    0.2
+  ],
+  "synthetic_renderer_rows": 240000,
+  "split": "train",
+  "max_samples_per_dataset": 60000,
+  "val_size": 1000,
+  "packing": true,
+  "block_size": 1024,
+  "batch_size": 2,
+  "grad_accum": 16,
+  "epochs": 1,
+  "max_steps": 6000,
+  "lr": 3e-06,
+  "min_lr": 6e-07,
+  "warmup_steps": 100,
+  "weight_decay": 0.05,
+  "grad_clip": 1.0,
+  "dropout": 0.05,
+  "eval_interval": 250,
+  "save_interval": 250,
+  "log_interval": 10,
+  "eval_batches": 30,
+  "seed": 42,
+  "dtype": "auto",
+  "out_dir": "runs/sft-renderer-rag-api-realtool-1b-v1",
+  "hub_repo_id": "ProjectScugnizz/scugnizz-1b",
+  "hub_path": "training-runs/sft-renderer-rag-api-realtool-1b-v1",
+  "push": true,
+  "push_every_save": true
+}

training-runs/sft-renderer-rag-api-realtool-1b-v1/checkpoint_best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38aea055adfb1c7153665374468c6b32831b240c07fbe3ccc9cd6e922159848a
+size 15758460499

training-runs/sft-renderer-rag-api-realtool-1b-v1/checkpoint_last.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:905d14d19ff5d365e0ce44d1184d918ed82555d1884d9981e836da101eda3f98
+size 15758460499