load_diff bugfix

FasterDecoding · Feb 23, 2024 · aa06fa6 · aa06fa6
1 parent 964ead3
commit aa06fa6
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 13 deletions.
diff --git a/README.md b/README.md
@@ -1,5 +1,3 @@
-[# Compressing Model Diffs for High-Througput Multi-Model Serving]: #
-
 # BitDelta: Your Fine-Tune May Only Be Worth One Bit
 
 [[Paper](https://arxiv.org/abs/2402.10193)][[Blog](https://fasterdecoding.github.io/BitDelta/)]
@@ -12,13 +10,8 @@ BitDelta compresses the weight delta between a fine-tuned and base model LLM to
   </a>
 </div>
 
-
 The current release supports:
 
-
-
-
-
 - Llama-2 and Mistral based models.
 - Memory efficient 16-bit + 1-bit Δ Linear in PyTorch
 - Triton kernel for fast inference
@@ -63,7 +56,6 @@ See [`demo/README.md`](https://github.com/FasterDecoding/BitDelta/blob/main/demo
 
 [BitDelta Demo.webm](https://github.com/FasterDecoding/BitDelta/assets/51351043/b56747df-1108-42f2-ae6f-05e1c460080c)
 
-
 ## Usage
 
 We provide some scripts in (`./scripts`) so you can compress your own models! As an example, we will compress `lmsys/vicuna-7b-v1.5` with base model `meta-llama/Llama-2-7b-hf`.
@@ -92,7 +84,7 @@ If `--save_full_model` is specified, the compressed model will also be saved in
 Double check the perplexity of the compressed model:
 
 ```
-CUDA_VISIBLE_DEVICES=0 python \
+### Perplexity CheckCUDA_VISIBLE_DEVICES=0 python \
     bitdelta/eval_ppl.py \
     --base_model meta-llama/Llama-2-7b-hf \
     --dataset_name wikitext \
@@ -103,17 +95,23 @@ CUDA_VISIBLE_DEVICES=0 python \
 
 ```
 
+### Perplexity Check
+
+To replicate our other results, please use `--save_full_model` to run the model in Llama format for compatibility with eval harnesses.
+
 ## Citation
 
 If you find BitDelta useful, please consider citing:
 
 ```
 @misc{liu2024bitdelta,
-      title={BitDelta: Your Fine-Tune May Only Be Worth One Bit}, 
+      title={BitDelta: Your Fine-Tune May Only Be Worth One Bit},
       author={James Liu and Guangxuan Xiao and Kai Li and Jason D. Lee and Song Han and Tri Dao and Tianle Cai},
       year={2024},
       eprint={2402.10193},
       archivePrefix={arXiv},
       primaryClass={cs.LG}
 }
 ```
+
+[# Compressing Model Diffs for High-Througput Multi-Model Serving]: #
diff --git a/bitdelta/diff.py b/bitdelta/diff.py
@@ -88,9 +88,11 @@ def load_diff(model, diff_dir):
             coeff = diff_dict[name + ".coeff"].to(device)
             mask = diff_dict[name + ".mask"].to(device)
 
-            setattr(module, "mask", mask)
-            setattr(module, "coeff", coeff)
-            # module.weight.add_((mask * coeff).to(module.weight.dtype))
+            # setattr(module, "mask", mask)
+            # setattr(module, "coeff", coeff)
+            weight = (unpack(mask)*2-1) * coeff
+
+            module.weight.add_(weight.T.to(module.weight.dtype))
         elif name + ".weight" in diff_dict:
             module.weight = nn.Parameter(diff_dict[name + ".weight"].to(device).to(module.weight.dtype))