Skip to content

fix backend missing-shard-idx bug which causes incorrect context inde… #93

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Feb 18, 2025
Merged

fix backend missing-shard-idx bug which causes incorrect context inde… #93

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f47d2e8
fix backend missing-shard-idx bug which causes incorrect context inde…
Hzfinfdu Feb 18, 2025
8ce543f
fix backend missing-shard-idx bug which causes incorrect context inde…
Hzfinfdu Feb 18, 2025
2eba74a
fix(server): correctly retrieve non-sharded dataset
dest1n1s Feb 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 11 additions & 7 deletions server/app.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@

from lm_saes.config import MongoDBConfig, SAEConfig
from lm_saes.database import MongoClient
from lm_saes.resource_loaders import load_dataset, load_model
from lm_saes.resource_loaders import load_dataset_shard, load_model
from lm_saes.sae import SparseAutoEncoder

device = "cuda" if torch.cuda.is_available() else "cpu"
Expand All @@ -28,7 +28,7 @@

sae_cache: dict[str, SparseAutoEncoder] = {}
lm_cache: dict[str, HookedTransformer] = {}
dataset_cache: dict[str, Dataset] = {}
dataset_cache: dict[tuple[str, int, int], Dataset] = {}


def get_model(name: str) -> HookedTransformer:
Expand All @@ -41,12 +41,12 @@ def get_model(name: str) -> HookedTransformer:
return lm_cache[name]


def get_dataset(name: str) -> Dataset:
def get_dataset(name: str, shard_idx: int = 0, n_shards: int = 1) -> Dataset:
cfg = client.get_dataset_cfg(name)
assert cfg is not None, f"Dataset {name} not found"
if name not in dataset_cache:
dataset_cache[name] = load_dataset(cfg)[0]
return dataset_cache[name]
if (name, shard_idx, n_shards) not in dataset_cache:
dataset_cache[name, shard_idx, n_shards] = load_dataset_shard(cfg, shard_idx, n_shards)
return dataset_cache[name, shard_idx, n_shards]


def get_sae(name: str) -> SparseAutoEncoder:
Expand Down Expand Up @@ -144,7 +144,11 @@ def get_feature(name: str, feature_index: str | int):
dataset_name = sampling.dataset_name[i]
model_name = sampling.model_name[i]
model = get_model(model_name)
data = get_dataset(dataset_name)[context_idx]
data = get_dataset(
dataset_name,
sampling.shard_idx[i] if sampling.shard_idx is not None else 0,
sampling.n_shards[i] if sampling.n_shards is not None else 1,
)[context_idx]
_, token_origins = model.to_tokens_with_origins(data)

# Replace image_key with image_url
Expand Down