mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-04-26 01:01:40 +00:00

teknium1 732c66b0f3 refactor: reorganize skills into sub-categories

The skills directory was getting disorganized — mlops alone had 40
skills in a flat list, and 12 categories were singletons with just
one skill each.

Code change:
- prompt_builder.py: Support sub-categories in skill scanner.
  skills/mlops/training/axolotl/SKILL.md now shows as category
  'mlops/training' instead of just 'mlops'. Backwards-compatible
  with existing flat structure.

Split mlops (40 skills) into 7 sub-categories:
- mlops/training (12): accelerate, axolotl, flash-attention,
  grpo-rl-training, peft, pytorch-fsdp, pytorch-lightning,
  simpo, slime, torchtitan, trl-fine-tuning, unsloth
- mlops/inference (8): gguf, guidance, instructor, llama-cpp,
  obliteratus, outlines, tensorrt-llm, vllm
- mlops/models (6): audiocraft, clip, llava, segment-anything,
  stable-diffusion, whisper
- mlops/vector-databases (4): chroma, faiss, pinecone, qdrant
- mlops/evaluation (5): huggingface-tokenizers,
  lm-evaluation-harness, nemo-curator, saelens, weights-and-biases
- mlops/cloud (2): lambda-labs, modal
- mlops/research (1): dspy

Merged singleton categories:
- gifs → media (gif-search joins youtube-content)
- music-creation → media (heartmula, songsee)
- diagramming → creative (excalidraw joins ascii-art)
- ocr-and-documents → productivity
- domain → research (domain-intel)
- feeds → research (blogwatcher)
- market-data → research (polymarket)

Fixed misplaced skills:
- mlops/code-review → software-development (not ML-specific)
- mlops/ml-paper-writing → research (academic writing)

Added DESCRIPTION.md files for all new/updated categories.

2026-03-09 03:35:53 -07:00

2.1 KiB

Raw Blame History

Deduplication Guide

Complete guide to exact, fuzzy, and semantic deduplication.

Exact deduplication

Remove documents with identical content.

from nemo_curator.modules import ExactDuplicates

# Exact deduplication
exact_dedup = ExactDuplicates(
    id_field="id",
    text_field="text",
    hash_method="md5"  # or "sha256"
)

deduped = exact_dedup(dataset)

Performance: ~16× faster on GPU vs CPU

Fuzzy deduplication

Remove near-duplicate documents using MinHash + LSH.

from nemo_curator.modules import FuzzyDuplicates

fuzzy_dedup = FuzzyDuplicates(
    id_field="id",
    text_field="text",
    num_hashes=260,        # MinHash permutations (more = accurate)
    num_buckets=20,        # LSH buckets (more = faster, less recall)
    hash_method="md5",
    jaccard_threshold=0.8  # Similarity threshold
)

deduped = fuzzy_dedup(dataset)

Parameters:

num_hashes: 128-512 (default 260)
num_buckets: 10-50 (default 20)
jaccard_threshold: 0.7-0.9 (default 0.8)

Performance: 16× faster on 8TB dataset (120h → 7.5h)

Semantic deduplication

Remove semantically similar documents using embeddings.

from nemo_curator.modules import SemanticDuplicates

semantic_dedup = SemanticDuplicates(
    id_field="id",
    text_field="text",
    embedding_model="sentence-transformers/all-MiniLM-L6-v2",
    embedding_batch_size=256,
    threshold=0.85,  # Cosine similarity threshold
    device="cuda"
)

deduped = semantic_dedup(dataset)

Models:

all-MiniLM-L6-v2: Fast, 384 dims
all-mpnet-base-v2: Better quality, 768 dims
Custom models supported

Comparison

Method	Speed	Recall	Use Case
Exact	Fastest	100%	Exact matches only
Fuzzy	Fast	~95%	Near-duplicates (recommended)
Semantic	Slow	~90%	Paraphrases, rewrites

Best practices

Start with exact dedup - Remove obvious duplicates
Use fuzzy for large datasets - Best speed/quality trade-off
Semantic for high-value data - Expensive but thorough
GPU acceleration required - 10-16× speedup

2.1 KiB Raw Blame History Unescape Escape