chp-adapter-huggingface 0.11.0__tar.gz

chp_adapter_huggingface-0.11.0/.gitignore

@@ -0,0 +1,36 @@
+node_modules/
+dist/
+.next/
+out/
+.turbo/
+*.tsbuildinfo
+.yalc/
+yalc.lock
+*.tgz
+.env
+.env.*
+coverage/
+dashboard/
+components/
+# ...but the cockpit's app source lives in app/components — never ignore real source.
+!cockpit/app/components/
+!cockpit/app/components/**
+!chp-evidence/app/components/
+!chp-evidence/app/components/**
+.tech-hub-cache/
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.chp/
+.DS_Store
+.coverage
+.forge/
+.hypothesis/
+.claude/
+
+# chp-agent evidence store
+.chp-agent/sessions.sqlite
+.publish-dist/
+
+# matrix-bot runtime session registry (CC session ids)
+matrix-bot/sessions.json

chp_adapter_huggingface-0.11.0/PKG-INFO

@@ -0,0 +1,21 @@
+Metadata-Version: 2.4
+Name: chp-adapter-huggingface
+Version: 0.11.0
+Summary: CHP adapter for local HuggingFace artifact consumption — pull, pipeline, embed, tokenize, dataset, cache
+Keywords: capability-host-protocol,chp,huggingface,nlp,transformers
+Requires-Python: >=3.11
+Requires-Dist: chp-core
+Requires-Dist: datasets>=2.20
+Requires-Dist: huggingface-hub>=0.24
+Requires-Dist: tokenizers>=0.19
+Requires-Dist: transformers>=4.40
+Provides-Extra: lifecycle
+Requires-Dist: evaluate>=0.4; extra == 'lifecycle'
+Requires-Dist: gradio-client>=1.0; extra == 'lifecycle'
+Requires-Dist: peft>=0.10; extra == 'lifecycle'
+Provides-Extra: multimodal
+Requires-Dist: diffusers>=0.30; extra == 'multimodal'
+Requires-Dist: librosa>=0.10; extra == 'multimodal'
+Requires-Dist: soundfile>=0.12; extra == 'multimodal'
+Provides-Extra: torch
+Requires-Dist: torch; extra == 'torch'

chp_adapter_huggingface-0.11.0/README.md

@@ -0,0 +1,37 @@
+# chp-adapter-huggingface
+
+Local consumption of HuggingFace Hub artifacts: pull models/datasets, run transformers pipelines, embed text, tokenize, and audit local cache.
+
+## Capabilities
+
+| Capability | Risk | Description |
+|---|---|---|
+| `pull` | medium | Download any HuggingFace Hub artifact (model, dataset, tokenizer) to local cache via snapshot_download. |
+| `run_pipeline` | medium | Run a transformers.pipeline() task on a locally-cached model. Inputs and outputs are never recorded in evidenc |
+| `embed` | medium | Generate text embeddings using a locally-cached feature-extraction model. Vectors are returned but never store |
+| `tokenize` | low | Encode text to token IDs or decode token IDs to text using a fast HuggingFace tokenizer. No full model load re |
+| `load_dataset` | medium | Load rows from a HuggingFace dataset (Hub or local). Streaming=true reads without downloading the full dataset |
+| `cache_info` | low | Scan the local HuggingFace cache and return a storage summary by artifact for governance and quota management. |
+| `search_models` | low | Search HuggingFace Hub for models by task, sort, and filter. Enables agents to discover models programmaticall |
+| `model_card` | low | Fetch structured model metadata from HuggingFace Hub: license, pipeline task, tags, gated status, author, and  |
+| `pull_for_local_llm` | medium | Pull GGUF model files from HuggingFace Hub and return the local path ready to pass directly to chp.adapters.lo |
+| `search_datasets` | low | Search HuggingFace Hub for datasets by task, sort, and filter. Symmetric to search_models — enables agents to  |
+| `search_spaces` | low | Search HuggingFace Hub for Spaces (Gradio/Streamlit apps) by SDK, sort, and filter. Precondition for call_spac |
+| `list_collections` | low | List HuggingFace Hub Collections — curated groupings of models and datasets. Enables agents to navigate themat |
+| `dataset_preview` | low | Preview the schema and first N rows of a HuggingFace dataset via the Dataset Viewer API — no download required |
+| `leaderboard_scores` | low | Fetch evaluation benchmark scores for a model from HuggingFace Hub (MMLU, ARC, TruthfulQA, etc.). Evidence-bac |
+| `evaluate` | low | Compute evaluation metrics (BLEU, ROUGE, accuracy, F1, exact_match) against ground-truth references. Quality g |
+| `apply_adapter` | medium | Download a LoRA/PEFT adapter from HuggingFace Hub and inspect its configuration. Returns the local adapter pat |
+| `call_space` | medium | Invoke any HuggingFace Gradio Space as a governed CHP capability via gradio_client. Space ID, api_name, and la |
+| `finetune` | high | Fine-tune a HuggingFace classification model locally using transformers.Trainer. Governance: model, dataset, h |
+| `quantize_to_gguf` | medium | Convert a local HuggingFace model directory to quantized GGUF using llama.cpp tools. Two-step: convert_hf_to_g |
+| `faiss_index` | low | Build or search a FAISS cosine-similarity index for RAG pipelines. 'build': creates IndexFlatIP from float emb |
+| `transcribe_audio` | medium | Transcribe an audio file to text using a Whisper ASR pipeline. Input is a local file path (e.g. from the files |
+| `classify_image` | low | Classify an image with a ViT/DeiT image-classification pipeline. Input is a local image path. Top-N labels and |
+| `generate_image` | medium | Generate an image from a text prompt via a diffusers DiffusionPipeline (Stable Diffusion, etc.) and save it to |
+
+## Notes
+
+Every capability is governed (risk-assessed via `safety.assess`) and evidenced (redacted: counts/ids, never payloads). Mutating ops may require approval.
+
+_README generated deterministically from the adapter's `@capability` metadata (`stewards/gen_readme.py`); refine the prose as needed._

chp_adapter_huggingface-0.11.0/chp_adapter_huggingface/__init__.py

@@ -0,0 +1,3 @@
+from .adapter import HuggingFaceAdapter, HuggingFaceConfig
+
+__all__ = ["HuggingFaceAdapter", "HuggingFaceConfig"]