Rename backbone.embedding.weight to backbone.embeddings.weight

#2
Files changed (2) hide show
  1. README.md +4 -0
  2. model.safetensors +1 -1
README.md CHANGED
@@ -75,3 +75,7 @@ saved_config.pop("mtp_layers_block_type", None)
75
  saved_config["hybrid_override_pattern"] = HYBRID_OVERRIDE_PATTERN
76
  config_file.write_text(json.dumps(saved_config, indent=2, sort_keys=True) + "\n", encoding="utf-8")
77
  ```
 
 
 
 
 
75
  saved_config["hybrid_override_pattern"] = HYBRID_OVERRIDE_PATTERN
76
  config_file.write_text(json.dumps(saved_config, indent=2, sort_keys=True) + "\n", encoding="utf-8")
77
  ```
78
+
79
+ The `model.safetensors` here is the output of this script with transformers 5.15.1, with the embeddings stored as
80
+ `backbone.embeddings.weight`, as in NVIDIA's checkpoints. transformers 5.15 saves them as `backbone.embedding.weight`,
81
+ which 5.16 and later no longer load ([huggingface/transformers#48075](https://github.com/huggingface/transformers/pull/48075)).
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:adae151a7d3b3f41c7c009f05b11316f1a7ed284005f273338ee4d064c70596e
3
  size 67685688
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8385e3aa89312feb58dd711d1308c12f9e9f56319edc783dfddd72a402831c15
3
  size 67685688