Initial model conversion and upload.

README.md

@@ -12,7 +12,7 @@ This model, an instance of `SelfCorrectiveLlama`, includes a hallucination detec
 
 ## Special Tokens
 
-The tokenizer has been expanded to include the following special tokens: `<DEL_W>`, `<DEL_S>`, `<DEL_A>`.
+The tokenizer has been expanded to include the following special tokens: `<DEL_S>`, `<DEL_A>`.
 
 ## How to Use
 

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c53c29abdcc9b1604d3a5ea122eeb4f055dc864973c0cce8265bc5d0d266e278
-size 1520518296
+oid sha256:27020f1eb19b1fe1cad4e5d0a05044e25ec6ed90d97fe13686ce58d4c14f5d24
+size 1520501910

model.safetensors.index.json

@@ -1,7 +1,7 @@
 {
   "metadata": {
-    "total_parameters": 8206450692,
-    "total_size": 16412901384
+    "total_parameters": 8206442499,
+    "total_size": 16412884998
   },
   "weight_map": {
     "hallucination_detector.bias": "model-00004-of-00004.safetensors",

modeling.py

@@ -13,7 +13,7 @@ class SelfCorrectiveLlama(LlamaForCausalLM):
     def __init__(self, config):
         super().__init__(config)
         
-        self.num_new_tokens = 3
+        self.num_new_tokens = 2
         self.original_vocab_size = config.vocab_size
 
         # Create a new, small embedding layer for only the special tokens
@@ -45,7 +45,7 @@ class SelfCorrectiveLlama(LlamaForCausalLM):
         clamped_input_ids = torch.clamp(input_ids, max=self.original_vocab_size - 1)
         inputs_embeds = self.model.embed_tokens(clamped_input_ids)
 
-        # Overwrite the embeddings for our new special tokens
+        # Overwrite the embeddings for new special tokens
         special_token_mask = input_ids >= self.original_vocab_size
         if special_token_mask.any():
             special_ids = input_ids[special_token_mask] - self.original_vocab_size

special_tokens_map.json

@@ -1,12 +1,5 @@
 {
   "additional_special_tokens": [
-    {
-      "content": "<DEL_W>",
-      "lstrip": false,
-      "normalized": false,
-      "rstrip": false,
-      "single_word": false
-    },
     {
       "content": "<DEL_S>",
       "lstrip": false,

tokenizer.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e746765bdab6128e3ea9bffe6c46b975ef15af806c15139fdabc913063afcf0
-size 17210472
+oid sha256:5507c539f08ee603e46b27ba2bafc4d73d66a53fdcade0b9ce53d9f5e2f3d4c5
+size 17210288

tokenizer_config.json

@@ -2049,14 +2049,6 @@
       "special": true
     },
     "128256": {
-      "content": "<DEL_W>",
-      "lstrip": false,
-      "normalized": false,
-      "rstrip": false,
-      "single_word": false,
-      "special": true
-    },
-    "128257": {
       "content": "<DEL_S>",
       "lstrip": false,
       "normalized": false,
@@ -2064,7 +2056,7 @@
       "single_word": false,
       "special": true
     },
-    "128258": {
+    "128257": {
       "content": "<DEL_A>",
       "lstrip": false,
       "normalized": false,
@@ -2074,7 +2066,6 @@
     }
   },
   "additional_special_tokens": [
-    "<DEL_W>",
     "<DEL_S>",
     "<DEL_A>"
   ],