Upload config.json with huggingface_hub

config.json

@@ -1,20 +1,36 @@
-{
-  "architectures": [
-    "Nav2TexDecoderModel"
-  ],
-  "bos_token_id": 0,
-  "d_ff": 4096,
-  "d_model": 1024,
-  "dropout": 0.1,
-  "dtype": "float32",
-  "eos_token_id": 2,
-  "lam_lambda": 0.1,
-  "max_seq_len": 1024,
-  "model_type": "nav2tex_decoder",
-  "n_heads": 16,
-  "n_layers": 8,
-  "pad_token_id": 1,
-  "squeeze_ratio": 4,
-  "transformers_version": "4.57.3",
-  "vocab_size": 50000
-}
+{
+  "tokenizer_dir": "/workspace/nav2tex-decoder/tokenizer",
+  "vocab_size": 50000,
+  "pad_token_id": 1,
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "data_glob": "/workspace/data/data/*.parquet",
+  "max_seq_len": 1024,
+  "cpe_score_threshold": 400,
+  "cpe_ratio": 0.25,
+  "d_model": 1024,
+  "n_heads": 16,
+  "n_layers": 8,
+  "d_ff": 4096,
+  "dropout": 0.1,
+  "squeeze_ratio": 4,
+  "lam_lambda": 0.1,
+  "batch_size": 16,
+  "grad_accum": 8,
+  "max_epochs": 10,
+  "warmup_steps": 2000,
+  "lr": 0.0001,
+  "weight_decay": 0.01,
+  "max_grad_norm": 1.0,
+  "label_smoothing": 0.1,
+  "save_dir": "/workspace/checkpoints",
+  "save_every_n_steps": 2000,
+  "log_every_n_steps": 100,
+  "num_workers": 8,
+  "prefetch_factor": 4,
+  "persistent_workers": true,
+  "cuda_benchmark": true,
+  "bf16": true,
+  "compile": true,
+  "flash_attn": false
+}