Add files using upload-large-folder tool

README.md

@@ -9,72 +9,28 @@ tags:
 library_name: transformers
 ---
 
-# GLM-5 Abliterated (BF16) - Delta Weights
+# GLM-5 Abliterated (BF16)
 
-Abliterated (uncensored) version of [zai-org/GLM-5](https://huggingface.co/zai-org/GLM-5) (744B MoE, 40B active parameters).
+This is an abliterated (uncensored) version of [zai-org/GLM-5](https://huggingface.co/zai-org/GLM-5) (744B MoE, 40B active parameters).
 
-**This repo contains only the 43 modified weight shards.** To use, download the base model and replace these shards.
+## What is abliteration?
 
-## Quick Setup
-
-```bash
-# 1. Download base model
-huggingface-cli download zai-org/GLM-5 --local-dir ./GLM-5-abliterated
-
-# 2. Download and overwrite modified shards
-huggingface-cli download skyblanket/GLM-5-abliterated --local-dir ./GLM-5-abliterated --include "*.safetensors"
-```
-
-Or use the merge script:
-
-```python
-from huggingface_hub import snapshot_download
-import json, shutil, os
-
-# Download base model
-base = snapshot_download("zai-org/GLM-5", local_dir="./GLM-5-abliterated")
-
-# Download modified shards
-delta = snapshot_download("skyblanket/GLM-5-abliterated")
-
-# Overwrite modified shards
-with open(os.path.join(delta, "modified_shards.json")) as f:
-    modified = json.load(f)["modified_shards"]
-
-for shard in modified:
-    src = os.path.join(delta, shard)
-    dst = os.path.join(base, shard)
-    if os.path.exists(src):
-        shutil.copy2(src, dst)
-        print(f"Replaced {shard}")
-
-print("Done! Model ready at ./GLM-5-abliterated")
-```
+Abliteration removes the "refusal direction" from the model weights using weight orthogonalization. This allows the model to respond to a wider range of prompts without safety refusals, while preserving general capability.
 
 ## Method
 
-Abliteration removes the "refusal direction" from model weights using weight orthogonalization.
-
-1. Computed refusal directions for all 78 layers using contrastive activation pairs
-2. Applied weight orthogonalization (W' = W - r_hat * r_hat^T * W) to layers 15-54:
+1. Computed refusal directions for all 78 layers using contrastive activation pairs (harmful vs harmless prompts)
+2. Applied weight orthogonalization to layers 15-54:
    - `self_attn.o_proj.weight` (attention output projection)
    - `mlp.shared_experts.down_proj.weight` (shared expert down projection)
-3. Alpha = 1.0, 80 weight matrices modified across 43 safetensor shards
+3. Alpha = 1.0, 80 weight matrices modified total
 
 ## Details
 
 - **Base model**: zai-org/GLM-5 (744B MoE, BF16)
 - **Modified layers**: 15-54 (40 of 78 total layers)
 - **Weights modified**: 80 (o_proj + shared_experts.down_proj per layer)
-- **Precision**: BF16 (full precision, no quantization)
-- **Delta size**: ~230GB (43 modified shards out of 282 total)
-
-## Files in this repo
-
-- 43 modified `.safetensors` shards (the delta weights)
-- `modified_shards.json` - list of which shards were modified
-- `model.safetensors.index.json` - full weight map (same as base model)
-- Config and tokenizer files
+- **Precision**: BF16 (full precision, no quantization artifacts)
 
 ## Disclaimer
 

model-00001-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:198ef923a7ca4effc5ead8ebf799fee10beb8ce081352fb099636f805d1deda9
+size 5342821416

model-00004-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21ea9316694ec51d00be3a54f4c90c6b8c927e9f9852aa673234ce10ea46b092
+size 5360347208

model-00005-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:447d3150fda6a4a95063c52402cc32a5e185e94ccb8ab1162463a6dec34d1130
+size 5359985352

model-00006-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bd92a8fb83f562439f97700404160da10f43d0c18b0e2ac900128c921ddd8d7
+size 5360347320

model-00008-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e848c982bf3fdcb5b7d8704e20e4b30aaff0d85f7314713dc82f3e069fb2d2cd
+size 5360347144

model-00009-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67d50e69f80e4f0959e6af2ed383aad04d3e930c5788b148b692df10652e8953
+size 5359985416

model-00011-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86ae5f0fe4ca9bd26926ff5feb2f8eea2524ff851bac18920180a74d5829f01b
+size 5360347288

model-00014-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:213d05d35132ab6c76771ccc87d9798b3417505f1456900fc111d1aa2dafafb5
+size 5360347312

model-00019-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7599267bb1fc68aeb5963661fb9f85b3e997b5752b278b27389e0560052c919a
+size 5360347160

model-00026-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c35662afd872d3d83f028523a774f3bf3873b8fd2772553097bd5712a28a5c4
+size 5360347232

model-00028-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:778444540ff8a43a51574b81a75eba0a15aa3df1bb4308659995b8f05d4a29a4
+size 5360347320

model-00030-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6df7b340183ad645f171edec945c107f4c417b2421933a99071f3f355dfe6e7c
+size 5360347168

model-00033-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d58b7681a194c6cf533a852c5780799df82cdb9776c5b24cb30778149645d7cb
+size 5360347312

model-00034-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f9666db39fcfad212eddb74581d83d9da9a16366080cdb2fcb1d0a6797db56d
+size 5360347104

model-00036-of-00282.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db004f95e3d6970343042fa1fef65528db625b1d954fa14f7c61b797c2b9d9bb
+size 5360347320