Tokenizer update

config.json

@@ -1,13 +1,18 @@
 {
+  "architectures": [
+    "SpeechLLMModel"
+  ],
   "audio_enc_dim": 1280,
   "audio_encoder_name": "facebook/hubert-xlarge-ll60k",
   "audio_processor_name": "facebook/hubert-large-ls960-ft",
   "auto_map": {
-    "AutoConfig": "config.SpeechLLMModelConfig"
+    "AutoConfig": "config.SpeechLLMModelConfig",
+    "AutoModel": "model.SpeechLLMModel"
   },
   "llm_dim": 2048,
   "llm_model_checkpoint": "hf_repo/llm_model_checkpoint",
   "llm_model_name": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
   "model_type": "custom_model",
+  "torch_dtype": "float32",
   "transformers_version": "4.38.2"
 }

model.py

@@ -39,6 +39,7 @@ class SpeechLLMModel(PreTrainedModel):
         llm_config = AutoConfig.from_pretrained(config.llm_model_name)
         self.llm_model =  AutoModelForCausalLM.from_config(llm_config)  
         self.llm_tokenizer = AutoTokenizer.from_pretrained(config.llm_model_name)
+        self.llm_tokenizer.pad_token = self.llm_tokenizer.eos_token
 
         peft_config = LoraConfig(
             r=4,