Tokenizer update
Browse files- config.json +6 -1
- model.py +1 -0
config.json
CHANGED
|
@@ -1,13 +1,18 @@
|
|
| 1 |
{
|
|
|
|
|
|
|
|
|
|
| 2 |
"audio_enc_dim": 1280,
|
| 3 |
"audio_encoder_name": "facebook/hubert-xlarge-ll60k",
|
| 4 |
"audio_processor_name": "facebook/hubert-large-ls960-ft",
|
| 5 |
"auto_map": {
|
| 6 |
-
"AutoConfig": "config.SpeechLLMModelConfig"
|
|
|
|
| 7 |
},
|
| 8 |
"llm_dim": 2048,
|
| 9 |
"llm_model_checkpoint": "hf_repo/llm_model_checkpoint",
|
| 10 |
"llm_model_name": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
|
| 11 |
"model_type": "custom_model",
|
|
|
|
| 12 |
"transformers_version": "4.38.2"
|
| 13 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"SpeechLLMModel"
|
| 4 |
+
],
|
| 5 |
"audio_enc_dim": 1280,
|
| 6 |
"audio_encoder_name": "facebook/hubert-xlarge-ll60k",
|
| 7 |
"audio_processor_name": "facebook/hubert-large-ls960-ft",
|
| 8 |
"auto_map": {
|
| 9 |
+
"AutoConfig": "config.SpeechLLMModelConfig",
|
| 10 |
+
"AutoModel": "model.SpeechLLMModel"
|
| 11 |
},
|
| 12 |
"llm_dim": 2048,
|
| 13 |
"llm_model_checkpoint": "hf_repo/llm_model_checkpoint",
|
| 14 |
"llm_model_name": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
|
| 15 |
"model_type": "custom_model",
|
| 16 |
+
"torch_dtype": "float32",
|
| 17 |
"transformers_version": "4.38.2"
|
| 18 |
}
|
model.py
CHANGED
|
@@ -39,6 +39,7 @@ class SpeechLLMModel(PreTrainedModel):
|
|
| 39 |
llm_config = AutoConfig.from_pretrained(config.llm_model_name)
|
| 40 |
self.llm_model = AutoModelForCausalLM.from_config(llm_config)
|
| 41 |
self.llm_tokenizer = AutoTokenizer.from_pretrained(config.llm_model_name)
|
|
|
|
| 42 |
|
| 43 |
peft_config = LoraConfig(
|
| 44 |
r=4,
|
|
|
|
| 39 |
llm_config = AutoConfig.from_pretrained(config.llm_model_name)
|
| 40 |
self.llm_model = AutoModelForCausalLM.from_config(llm_config)
|
| 41 |
self.llm_tokenizer = AutoTokenizer.from_pretrained(config.llm_model_name)
|
| 42 |
+
self.llm_tokenizer.pad_token = self.llm_tokenizer.eos_token
|
| 43 |
|
| 44 |
peft_config = LoraConfig(
|
| 45 |
r=4,
|