Upload model

config.json

@@ -16,5 +16,5 @@
   "model_type": "moonshine",
   "n_head": 8,
   "torch_dtype": "float32",
-  "transformers_version": "4.47.0.dev0"
+  "transformers_version": "4.46.1"
 }

modeling_moonshine.py

@@ -1,9 +1,13 @@
 from einops import rearrange
 from einops.layers.torch import Rearrange
 from torch import nn
+from transformers import PreTrainedModel
+
 import math
 import torch
 
+from .configuration_moonshine import MoonshineConfig
+
 
 class RotaryEmbedding(nn.Module):
     def __init__(self, dim, base=10000):
@@ -426,11 +430,8 @@ class MoonshineModelTorch(nn.Module):
         self.d_head = inner_dim // n_head
 
     def generate(self, src):
-        start = time.time()
         preprocessed = self.preprocessor(src)
-        start = time.time()
         enc = self.encoder(preprocessed)
-        start = time.time()
         sot_token = 1
         eot_token = 2
 
@@ -443,8 +444,6 @@ class MoonshineModelTorch(nn.Module):
             for i in range(1, 1 + self.dec_depth * 4, self.dec_depth)
         ]
 
-        start = time.time()
-
         sample = logits[:, -1].argmax(dim=-1, keepdim=True)
         seq = torch.cat((seq, sample), dim=-1)
 
@@ -466,8 +465,6 @@ class MoonshineModelTorch(nn.Module):
 
         return seq
 
-from transformers import PreTrainedModel
-from .configuration_moonshine import MoonshineConfig
 
 class MoonshineModel(PreTrainedModel):
     config_class = MoonshineConfig
@@ -487,22 +484,3 @@ class MoonshineModel(PreTrainedModel):
 
     def forward(self, tensor):
         return self.model.generate(tensor)
-
-class MoonshineForConditionalGeneration(PreTrainedModel):
-    config_class = MoonshineConfig
-
-    def __init__(self, config):
-        super().__init__(config)
-        self.model = MoonshineModelTorch(
-            dim = config.dim,
-            inner_dim = config.inner_dim,
-            enc_depth = config.enc_depth,
-            dec_depth = config.dec_depth,
-            n_head = config.n_head,
-            dec_voc_size = config.dec_voc_size,
-            enc_ff_swiglu = config.enc_ff_swiglu,
-            dec_ff_swiglu = config.dec_ff_swiglu,
-        )
-
-    def forward(self, tensor):
-        return self.model.generate(tensor)