LlamaPFN: LLaMA for Prior-Fitted Networks - no positional embeddings

config.json

@@ -1,25 +1,24 @@
 {
   "architectures": [
-    "CustomLlamaModel"
+    "LlamaPFNModel"
   ],
   "attention_bias": false,
   "attention_dropout": 0.0,
   "bos_token_id": 1,
-  "disable_rope": true,
   "dtype": "float32",
   "eos_token_id": 2,
   "head_dim": 64,
   "hidden_act": "silu",
-  "hidden_size": 512,
+  "hidden_size": 768,
   "initializer_range": 0.02,
   "input_dim": 512,
-  "intermediate_size": 2048,
+  "intermediate_size": 3072,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
-  "model_type": "custom_llama_no_pos",
-  "num_attention_heads": 8,
+  "model_type": "llama_pfn",
+  "num_attention_heads": 12,
   "num_hidden_layers": 6,
-  "num_key_value_heads": 8,
+  "num_key_value_heads": 12,
   "pretraining_tp": 1,
   "rms_norm_eps": 1e-06,
   "rope_scaling": null,
@@ -27,6 +26,5 @@
   "tie_word_embeddings": false,
   "transformers_version": "4.56.2",
   "use_cache": true,
-  "use_linear_embedding": true,
   "vocab_size": 32000
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:32cd751f500f9fc74127bdddee6ffb9c56f6d9f7f50534fe1b943c47791bbec3
-size 101744264
+oid sha256:5adf705dd61a5604a9d848e1ac1e04df7a55eea1b529626454707a28ad68bdde
+size 228111048