Use neutral public resource descriptions
Browse files
README.md
CHANGED
|
@@ -3,13 +3,13 @@ language:
|
|
| 3 |
- en
|
| 4 |
tags:
|
| 5 |
- model-reconstruction
|
| 6 |
-
-
|
| 7 |
---
|
| 8 |
# Exact historical Qwen2.5-7B critique-token Base reconstruction bundle
|
| 9 |
|
| 10 |
-
This is a
|
| 11 |
Download Qwen/Qwen2.5-7B at d149729398750b98c0af14eb82c78cfe92750796.
|
| 12 |
-
Use rebuild_critique_base.py from
|
| 13 |
source BF16 tensors to FP32 and apply the included exact byte patches.
|
| 14 |
The bundle includes the historical tokenizer, config and original shard headers.
|
| 15 |
All 339 source/target tensors were compared; the original source weight files
|
|
|
|
| 3 |
- en
|
| 4 |
tags:
|
| 5 |
- model-reconstruction
|
| 6 |
+
- writing-feedback
|
| 7 |
---
|
| 8 |
# Exact historical Qwen2.5-7B critique-token Base reconstruction bundle
|
| 9 |
|
| 10 |
+
This is a public reconstruction bundle, NOT a standalone loadable model.
|
| 11 |
Download Qwen/Qwen2.5-7B at d149729398750b98c0af14eb82c78cfe92750796.
|
| 12 |
+
Use rebuild_critique_base.py from the experiment code to convert the
|
| 13 |
source BF16 tensors to FP32 and apply the included exact byte patches.
|
| 14 |
The bundle includes the historical tokenizer, config and original shard headers.
|
| 15 |
All 339 source/target tensors were compared; the original source weight files
|