Bleeding Edge - Accept All Pull Requests#22
Open
Tauheed-Elahee wants to merge 32 commits intoJoshEngels:mainfrom
Open
Bleeding Edge - Accept All Pull Requests#22Tauheed-Elahee wants to merge 32 commits intoJoshEngels:mainfrom
Tauheed-Elahee wants to merge 32 commits intoJoshEngels:mainfrom
Conversation
As per https://github.com/matplotlib/matplotlib/blob/42b88d01fdd93846d925b1097167d36ea31c7733/doc/api/prev_api_changes/api_changes_3.9.0/removals.rst?plain=1#L147, `legendHandles` was removed in matplotlib 3.9 after previously being deprecated in favor of `legend_handles`. Since `requirements.txt` requires matplotlib 3.9, we just use the new name. Fixes JoshEngels#12 Done with ``` git grep --name-only legendHandles | xargs sed -i.bak 's/legendHandles/legend_handles/g' ```
…_devices 2" when there aren't 2 GPUs
```
File "/home/jason/MultiDimensionalFeatures/multid/lib/python3.10/site-packages/transformer_lens/HookedTransformerConfig.py", line 315, in __post_init__
torch.cuda.device_count() >= self.n_devices
AssertionError: Not enough CUDA devices to support n_devices 2
```
Also allow passing `--device cpu`, `--device mps`, etc to `circle_probe_interventions.py`, and remove the logic for special handling of numbers *n* as `cuda:n` to simplify logic.
TransformerLens expects "gpt2" not "gpt-2" as the model identifier. Add tl_model_name mapping so HookedTransformer.from_pretrained() receives the correct name.
Replace hardcoded num_devices=2 for Mistral with max(1, t.cuda.device_count()) to support single-GPU and CPU-only machines while using all available GPUs when present.
Newer versions of sae_lens changed forward() to return a reconstructed tensor instead of an object with feature_acts. Add hasattr check and fall back to encode() to get feature activations.
The actual CLI argument in clustering.py is --method, not --clustering_type.
JoshEngels#11 device handling + add dtype
…ved: keep pathlib paths + add weights_only=False
Tauheed-Elahee
force-pushed
the
bleeding-edge
branch
from
945d7ec to
cada533
Compare
Author
|
Signed all signatures for that nice "Verified" GitHub badge for all commits. |
Allow overriding BASE_DIR via environment variable for cloud/remote environments. Falls back to the existing relative cache/ path when the variable is not set. Depends on PR JoshEngels#5 (pathlib refactor).
Most machines index GPUs from 0. The previous default of cuda:4 assumed a specific multi-GPU server setup and would fail on machines with fewer than 5 GPUs. Depends on PR JoshEngels#11 (CPU fallback).
Strip explicit NVIDIA library version pins (nvidia-cuda-runtime-cu12, nvidia-cudnn-cu12, etc.) and triton. These are automatically resolved by PyTorch based on the system's CUDA installation, and hard-pinning them to CUDA 12.1 prevents installation on systems with different CUDA versions.
- Add requirements.in as the unpinned input file for pip-compile - Remove nvidia-* packages from requirements.in (transitive deps of torch) - Replace requirements.txt with pip-compile output (Python 3.12) - Major version updates: torch 2.10.0, sae-lens 6.37.1, transformer-lens 2.17.0, transformers 4.57.6, datasets 4.5.0 - Adds jupyterlab, ipywidgets, tensorflow as direct dependencies
Tauheed-Elahee
force-pushed
the
bleeding-edge
branch
from
ee5b95e to
b9dc4ef
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
An attempt to consolidate merge-able pull requests into a single branch that one can use instead of using the
mainbranch.This pull request solves issue #21