Upload folder using huggingface_hub

README.md

@@ -144,6 +144,7 @@ model = AutoModel.from_pretrained(
     path,
     torch_dtype=torch.bfloat16,
     low_cpu_mem_usage=True,
+    use_flash_attn=False,
     trust_remote_code=True).eval().cuda()
 ```
 
@@ -158,6 +159,7 @@ model = AutoModel.from_pretrained(
     torch_dtype=torch.bfloat16,
     load_in_8bit=True,
     low_cpu_mem_usage=True,
+    use_flash_attn=False,
     trust_remote_code=True).eval()
 ```
 
@@ -172,6 +174,7 @@ model = AutoModel.from_pretrained(
     torch_dtype=torch.bfloat16,
     load_in_4bit=True,
     low_cpu_mem_usage=True,
+    use_flash_attn=False,
     trust_remote_code=True).eval()
 ```
 
@@ -216,6 +219,7 @@ model = AutoModel.from_pretrained(
     path,
     torch_dtype=torch.bfloat16,
     low_cpu_mem_usage=True,
+    use_flash_attn=False,
     trust_remote_code=True,
     device_map=device_map).eval()
 ```
@@ -311,6 +315,7 @@ model = AutoModel.from_pretrained(
     path,
     torch_dtype=torch.bfloat16,
     low_cpu_mem_usage=True,
+    use_flash_attn=False,
     trust_remote_code=True).eval().cuda()
 tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True, use_fast=False)
 

configuration_intern_vit.py

@@ -71,7 +71,7 @@ class InternVisionConfig(PretrainedConfig):
             intermediate_size=12800,
             qk_normalization=True,
             num_hidden_layers=48,
-            use_flash_attn=True,
+            use_flash_attn=False,
             hidden_act='gelu',
             norm_type='rms_norm',
             layer_norm_eps=1e-6,