Upload convert_to_linear.py with huggingface_hub

convert_to_linear.py

@@ -95,6 +95,17 @@ class NewGptOssExperts(nn.Module):
             return mixed.view(batch_size, -1, self.hidden_size)
 
 
+# to load do
 # monkey patch to linear
 from transformers.models.gpt_oss import modeling_gpt_oss
 modeling_gpt_oss.GptOssExperts = NewGptOssExperts
+
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model = AutoModelForCausalLM.from_pretrained(
+    "imdatta0/gpt_oss_20b_linear", # make sure you load the right weights
+    device_map='cuda:0', # modify appropriately.
+    torch_dtype=torch.bfloat16
+)
+