latest checkpoint push

app.py

@@ -34,7 +34,7 @@ def plot_mel_spectrogram(mel_spec):
 def get_or_load_model():
     if 'model' not in st.session_state or 'tokenizer' not in st.session_state or 'processor' not in st.session_state:
         ckpt_path = "checkpoints/pretrained_checkpoint.ckpt"
-        model = SpeechLLMLightning.load_from_checkpoint(ckpt_path, quantize=True)
+        model = SpeechLLMLightning.load_from_checkpoint(ckpt_path, quantize=False)
         tokenizer = model.llm_tokenizer
         model.eval()
         model.freeze()

checkpoints/pretrained_checkpoint.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2ff66ef9c15a19cbb171662883cad8f142f6edf0c06262f4b1df6eb5908422b1
+oid sha256:eaaec0efc7f2531bc81a4e11f875d270c000ed0bd4ee1b94e7c70ee887429f4b
 size 8644540057

trainer.py

@@ -18,7 +18,7 @@ else:
     device = "cpu"
 
 class SpeechLLMLightning(pl.LightningModule):
-    def __init__(self, audio_enc_dim=512, llm_dim=2048, llm_name="TinyLlama/TinyLlama-1.1B-Chat-v1.0", quantize=True):
+    def __init__(self, audio_enc_dim=512, llm_dim=2048, llm_name="TinyLlama/TinyLlama-1.1B-Chat-v1.0", quantize=False):
         super().__init__()
         self.save_hyperparameters()
 
@@ -52,7 +52,7 @@ class SpeechLLMLightning(pl.LightningModule):
         self.llm_model.eval()
 
         if quantize:
-            self.llm_model = jit.script(self.llm_model)
+            # self.llm_model = jit.script(self.llm_model)
             self.llm_model = quantize_dynamic(
                                 self.llm_model, {nn.Linear}, dtype=torch.qint8
             )