trohrbaugh commited on
Commit
f7e6c0a
·
verified ·
1 Parent(s): 4a382cf

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +19 -5
README.md CHANGED
@@ -34,17 +34,31 @@ tags:
34
  | **KL divergence** | 0.0506 | 0 *(by definition)* |
35
  | **Refusals** | 1/100 | 95/100 |
36
 
37
- -----
 
 
 
 
 
 
 
 
 
 
38
 
 
39
 
 
 
 
40
 
41
  ![Evaluation Results](./papers/iquest-coder-v1-logo.png)
42
 
43
  <p align="center">
44
- 📘 <a href="https://iquestlab.github.io">Blog</a >
45
  &nbsp;•&nbsp;
46
- 📄 <a href="https://github.com/IQuestLab/IQuest-Coder-V1/blob/main/papers/IQuest_Coder_Technical_Report.pdf">Technical Report</a >
47
- </p >
48
 
49
  # IQuest-Coder-V1 Model Family
50
 
@@ -204,4 +218,4 @@ If you find our work helpful, please cite:
204
  journal={arXiv preprint arXiv:2512.22087},
205
  year={2025}
206
  }
207
- ```
 
34
  | **KL divergence** | 0.0506 | 0 *(by definition)* |
35
  | **Refusals** | 1/100 | 95/100 |
36
 
37
+ ## Transformers 5.x Compatibility
38
+
39
+ This model uses custom modeling code (`trust_remote_code=True`) that requires a one-time patch when using **transformers >= 5.0**. Without the patch, the model fails to load due to a RoPE positional encoding incompatibility.
40
+
41
+ **After downloading the model, run:**
42
+
43
+ ```bash
44
+ python patch_iquestcoder.py
45
+ ```
46
+
47
+ The script auto-finds and patches all cached `modeling_iquestcoder.py` files. It is safe to run multiple times — if the fix is already applied, it simply skips.
48
 
49
+ > **What the patch fixes:** The custom `IQuestCoderRotaryEmbedding` references a `"default"` rope type that was removed from the transformers 5.x rope registry, and imports `SlidingWindowCache` which was renamed. The patch adds the missing rope initialization function and updates the cache import.
50
 
51
+ If you prefer to apply the fix manually, see the [patch script source](./patch_iquestcoder.py) for details.
52
+
53
+ -----
54
 
55
  ![Evaluation Results](./papers/iquest-coder-v1-logo.png)
56
 
57
  <p align="center">
58
+ 📘 <a href="https://iquestlab.github.io">Blog</a>
59
  &nbsp;•&nbsp;
60
+ 📄 <a href="https://github.com/IQuestLab/IQuest-Coder-V1/blob/main/papers/IQuest_Coder_Technical_Report.pdf">Technical Report</a>
61
+ </p>
62
 
63
  # IQuest-Coder-V1 Model Family
64
 
 
218
  journal={arXiv preprint arXiv:2512.22087},
219
  year={2025}
220
  }
221
+ ```