Skip to main content

Hugging Face Hub helpers for cjm-plugin-system plugins: cache-config mixin, progress-reporting snapshot downloads, and typed-OOM model loading.

Project description

cjm-hf-plugin-utils

Install

pip install cjm_hf_plugin_utils

Project Structure

nbs/
├── cache_config.ipynb # A dataclass mixin adding HuggingFace Hub cache / revision / air-gap / security fields to a plugin's config, with `RELOAD_TRIGGER` + JSON-schema metadata pre-set.
├── download.ipynb     # Wrap `huggingface_hub.snapshot_download` so per-file progress flows to the plugin's `report_progress`, defeating the substrate's stall detector during downloads.
└── loading.ipynb      # Call `model_class.from_pretrained(...)` and convert CUDA OOM into the substrate's typed `PluginResourceError` for CR-7 reactive retry.

Total: 3 notebooks

Module Dependencies

graph LR
    cache_config["cache_config<br/>HF cache config mixin"]
    download["download<br/>Snapshot download with progress"]
    loading["loading<br/>Pretrained loading with typed OOM"]

No cross-module dependencies detected.

CLI Reference

No CLI commands found in this project.

Module Overview

Detailed documentation for each module in the project:

HF cache config mixin (cache_config.ipynb)

A dataclass mixin adding HuggingFace Hub cache / revision / air-gap / security fields to a plugin’s config, with RELOAD_TRIGGER + JSON-schema metadata pre-set.

Import

from cjm_hf_plugin_utils.cache_config import (
    HFCacheConfig
)

Classes

@dataclass
class HFCacheConfig:
    """
    Mixin adding HuggingFace Hub cache/revision/air-gap/security fields to a plugin config.
    
    Compose by inheritance:
    
        @dataclass
        class MyPluginConfig(HFCacheConfig):
            model_id: str = field(default="org/model", metadata={RELOAD_TRIGGER: "model"})
            # ... plugin-specific fields (all defaulted) ...
    
    Each field is `RELOAD_TRIGGER`-tagged `"model"` (a change invalidates a loaded
    model) and carries `SCHEMA_TITLE`/`SCHEMA_DESC` so the plugin-config UI renders
    it. `trust_remote_code` defaults to False and is flagged DANGER in its help text.
    """
    
    cache_dir: Optional[str] = field(...)
    revision: Optional[str] = field(...)
    local_files_only: bool = field(...)
    trust_remote_code: bool = field(...)

Snapshot download with progress (download.ipynb)

Wrap huggingface_hub.snapshot_download so per-file progress flows to the plugin’s report_progress, defeating the substrate’s stall detector during downloads.

Import

from cjm_hf_plugin_utils.download import (
    snapshot_download_with_progress
)

Functions

def snapshot_download_with_progress(
    """
    Download an HF Hub snapshot, streaming per-file progress to `report_progress`.
    
    When `report_progress` is given, a `tqdm_class` subclass forwards each update
    as `report_progress(downloaded / total, "downloading <file>")`. When it is
    None, the default HF Hub progress bars are used unchanged.
    
    Returns the local snapshot directory; subsequent `from_pretrained` calls with
    the same `cache_dir` + `local_files_only=True` hit the populated cache.
    """

Pretrained loading with typed OOM (loading.ipynb)

Call model_class.from_pretrained(...) and convert CUDA OOM into the substrate’s typed PluginResourceError for CR-7 reactive retry.

Import

from cjm_hf_plugin_utils.loading import (
    load_pretrained_with_oom
)

Functions

def load_pretrained_with_oom(
    model_class: Type,            # A class exposing `.from_pretrained(repo_id, **kwargs)`
    repo_id: str,                 # Model repo id or local path passed to from_pretrained
    *,
    label: Optional[str] = None,  # OOM-message context (default: f"loading {repo_id!r}")
    **kwargs,                     # Forwarded verbatim to model_class.from_pretrained
) -> Any:                         # The loaded model
    """
    Call `model_class.from_pretrained(repo_id, **kwargs)`, converting CUDA OOM to a typed error.
    
    On `torch.cuda.OutOfMemoryError`, re-raises as `PluginResourceError` (via
    `cuda_oom_to_plugin_resource_error`) so the substrate's CR-7 reactive-retry
    path can evict + reload + retry. All other exceptions propagate unchanged.
    
    Wrap the call in `self.heartbeat(...)` at the plugin site to cover the silent
    construction phase; pre-download with `snapshot_download_with_progress` for
    real download progress.
    """

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

cjm_hf_plugin_utils-0.0.1.tar.gz (9.3 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

cjm_hf_plugin_utils-0.0.1-py3-none-any.whl (11.6 kB view details)

Uploaded Python 3

File details

Details for the file cjm_hf_plugin_utils-0.0.1.tar.gz.

File metadata

  • Download URL: cjm_hf_plugin_utils-0.0.1.tar.gz
  • Upload date:
  • Size: 9.3 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.13

File hashes

Hashes for cjm_hf_plugin_utils-0.0.1.tar.gz
Algorithm Hash digest
SHA256 22b1e66a7150c0fa730c75297b4685dfd7f538a335299a7e9312ddeb6f330e53
MD5 ff290ea20f04aa5ff66880fd99f4d124
BLAKE2b-256 58328345feb354406b5a928d9b45b8456224e8806adce321d3789369628fae45

See more details on using hashes here.

File details

Details for the file cjm_hf_plugin_utils-0.0.1-py3-none-any.whl.

File metadata

File hashes

Hashes for cjm_hf_plugin_utils-0.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 f9bd4a463927a8935d207902113b6467a85e926a16f81f7fe2162ac0082b79e7
MD5 b824a2fda94ab952e6b565e02460ee87
BLAKE2b-256 9c92cb46cead32bda6c0ed6a488c28bffc00cc0244ac154c41a0c15504ca6e5f

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page