diff --git a/llama_stack/models/llama/llama4/tokenizer.py b/llama_stack/models/llama/llama4/tokenizer.py
index 4d271e5fd..8eabc3205 100644
--- a/llama_stack/models/llama/llama4/tokenizer.py
+++ b/llama_stack/models/llama/llama4/tokenizer.py
@@ -56,9 +56,11 @@ LLAMA4_TEXT_POST_TRAIN_SPECIAL_TOKENS = [
     "<|text_post_train_reserved_special_token_3|>",
     "<|text_post_train_reserved_special_token_4|>",
     "<|text_post_train_reserved_special_token_5|>",
+    "<|text_post_train_reserved_special_token_6|>",
+    "<|text_post_train_reserved_special_token_7|>",
     "<|finetune_right_pad|>",
 ] + get_reserved_special_tokens(
-    "text_post_train", 61, 6
+    "text_post_train", 61, 8
 )  # <|text_post_train_reserved_special_token_6|>, ..., <|text_post_train_reserved_special_token_66|>
 
 # 200080, ..., 201133
diff --git a/llama_stack/providers/inline/inference/meta_reference/generators.py b/llama_stack/providers/inline/inference/meta_reference/generators.py
index 65bed4d8c..34dd58a9a 100644
--- a/llama_stack/providers/inline/inference/meta_reference/generators.py
+++ b/llama_stack/providers/inline/inference/meta_reference/generators.py
@@ -259,7 +259,7 @@ class Llama3Generator:
 
         temperature, top_p = _infer_sampling_params(sampling_params)
         for result in self.inner_generator.generate(
-            llm_inputs=[self.formatter.encode_content(request.content)],
+            model_inputs=[self.formatter.encode_content(request.content)],
             max_gen_len=max_gen_len,
             temperature=temperature,
             top_p=top_p,
@@ -284,7 +284,7 @@ class Llama3Generator:
 
         temperature, top_p = _infer_sampling_params(sampling_params)
         for result in self.inner_generator.generate(
-            llm_inputs=[self.formatter.encode_dialog_prompt(request.messages, _infer_tool_prompt_format(request))],
+            model_inputs=[self.formatter.encode_dialog_prompt(request.messages, _infer_tool_prompt_format(request))],
             max_gen_len=max_gen_len,
             temperature=temperature,
             top_p=top_p,