Skip to content

feat: automatically reproduce model from reproduce.json #326

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

Open
p-e-w wants to merge 6 commits into
base: master
Choose a base branch
from
+633 −185
Open

feat: automatically reproduce model from reproduce.json #326

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
4df78d7
feat: load reproduction information
p-e-w May 12, 2026
223d0f8
feat: check reproduction environment against original environment
p-e-w May 21, 2026
96e4d2b
fix: remove `trust_remote_code` setting
p-e-w May 29, 2026
3bcb2f6
feat: reproduce model from JSON file
p-e-w Jun 4, 2026
bd3bf2c
feat: verify hashes of uploaded weight files
p-e-w Jun 5, 2026
b9d5b45
fix: fix issues in automatic reproduction system (#352)
Vinay-Umrethe Jun 7, 2026
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
4 changes: 0 additions & 4 deletions config.default.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,10 +123,6 @@ n_trials = 200
# Number of trials that use random sampling for the purpose of exploration.
n_startup_trials = 60

# Random seed for reproducible optimization. Set to an integer to enable.
# Applies to Python's random module, NumPy, PyTorch, and Optuna.
# seed = 75

# Directory to save and load study progress to/from.
study_checkpoint_dir = "checkpoints"

Expand Down
26 changes: 19 additions & 7 deletions src/heretic/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,11 @@ class RowNormalization(str, Enum):
FULL = "full"


class ExportStrategy(str, Enum):
MERGE = "merge"
ADAPTER = "adapter"


class DatasetSpecification(BaseModel):
dataset: str = Field(
description="Hugging Face dataset ID, or path to dataset on disk."
Expand Down Expand Up @@ -113,6 +118,15 @@ class Settings(BaseSettings):
exclude=True,
)

reproduce: str | None = Field(
Comment thread
p-e-w marked this conversation as resolved.
default=None,
description=(
"If this path or URL to a reproduce.json file is set, load reproduction information "
"from that file, and attempt to reproduce the abliterated model it originated from."
),
exclude=True,
)

dtypes: list[str] = Field(
default=[
# In practice, "auto" almost always means bfloat16.
Expand Down Expand Up @@ -161,13 +175,6 @@ class Settings(BaseSettings):
),
)

trust_remote_code: bool | None = Field(
default=None,
description="Whether to trust remote code when loading the model.",
# For security reasons, we don't store this setting.
exclude=True,
)

batch_size: int = Field(
default=0, # auto
description="Number of input sequences to process in parallel (0 = auto).",
Expand Down Expand Up @@ -410,6 +417,11 @@ class Settings(BaseSettings):
description="Maximum size for individual safetensors files generated when exporting a model.",
)

export_strategy: ExportStrategy | None = Field(
default=None,
description='How to export the model: "merge", "adapter", or unset to prompt the user.',
)

refusal_markers: list[str] = Field(
default=[
"sorry",
Expand Down
Loading