Upload data/lesson_01_huggingface_ecosystem.jsonl with huggingface_hub
Browse files
data/lesson_01_huggingface_ecosystem.jsonl
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{"id": "hf-001", "tier": "T1", "topic": "HuggingFace/Overview", "title": "HuggingFace Hub Core Concept", "content": "HuggingFace Hub is the world's largest AI platform with 3 types of Git-based repositories: Models (1M+), Datasets (500K+), and Spaces (500K+). Every repo supports versioning, branching, PRs, and LFS for large files.", "tags": ["huggingface", "hub", "overview"], "importance": 9, "created": "2026-04-03"}
|
| 2 |
+
{"id": "hf-002", "tier": "T1", "topic": "HuggingFace/API", "title": "HuggingFace Hub Python API", "content": "The huggingface_hub Python library provides programmatic access: create_repo(), upload_file(), list_models(), download_snapshot(). Use HfApi class. Auth via HF_TOKEN environment variable or huggingface-cli login.", "tags": ["huggingface", "api", "python"], "importance": 9, "created": "2026-04-03"}
|
| 3 |
+
{"id": "hf-003", "tier": "T1", "topic": "HuggingFace/Models", "title": "Key Model Categories on HF", "content": "Important model categories: Text Generation (Llama 3, Mistral, Qwen 2.5), Embeddings (bge-large, e5-large for RAG), Image Generation (FLUX.1, SD), TTS (Chatterbox, Bark), OCR (DeepSeek-OCR), Object Detection (SAM3).", "tags": ["huggingface", "models", "catalog"], "importance": 8, "created": "2026-04-03"}
|
| 4 |
+
{"id": "hf-004", "tier": "T1", "topic": "HuggingFace/Spaces", "title": "Space SDK Types and Sleep Behavior", "content": "HF Spaces support 3 SDKs: Gradio (Python GUI, sleeps after 48h), Docker (full container, sleeps after 48h), Static (HTML/CSS/JS on CDN, NEVER sleeps). CPU Basic is free. ZeroGPU provides free dynamic GPU for Gradio only.", "tags": ["huggingface", "spaces", "deployment"], "importance": 9, "created": "2026-04-03"}
|
| 5 |
+
{"id": "hf-005", "tier": "T1", "topic": "HuggingFace/MCP", "title": "MCP Tool Spaces Available", "content": "16+ MCP-enabled Spaces can be called directly from ZeroClaw: Image Gen (flux1_schnell, FLUX.1-Krea-dev, Qwen-Image), Image Edit (FLUX.1-Kontext-Dev, Photo-Mate-i2i), OCR (DeepSeek-OCR), TTS (Chatterbox), Video Gen (wan2-2), Background Removal, Object Detection (SAM3).", "tags": ["huggingface", "mcp", "tools", "spaces"], "importance": 10, "created": "2026-04-03"}
|
| 6 |
+
{"id": "hf-006", "tier": "T1", "topic": "HuggingFace/Datasets", "title": "Dataset Features for Knowledge Management", "content": "HF Datasets support: streaming (no full download needed), Dataset Viewer (web preview), Parquet format, gated access (require approval), private repos (owner only). Can be used as persistent knowledge store with 4-tier security: T1 Public, T2 Public+Encrypted, T3 Private, T4 Private+Encrypted.", "tags": ["huggingface", "datasets", "knowledge", "security"], "importance": 10, "created": "2026-04-03"}
|
| 7 |
+
{"id": "hf-007", "tier": "T1", "topic": "HuggingFace/Inference", "title": "HF Inference API Options", "content": "Three inference options: 1) Serverless API (free, rate-limited, 200K+ models), 2) Inference Endpoints (dedicated, $0.06/hr CPU), 3) Text Generation Inference (TGI, optimized serving). Can serve as backup provider when primary (Gemini) is rate-limited.", "tags": ["huggingface", "inference", "api", "backup"], "importance": 8, "created": "2026-04-03"}
|
| 8 |
+
{"id": "hf-008", "tier": "T1", "topic": "HuggingFace/Courses", "title": "Free HF Learning Courses", "content": "7 free courses from HuggingFace: LLM Course (Large Language Models), Agents Course (AI agents), MCP Course (Model Context Protocol), Smol Course (efficient small models), Deep RL Course, Computer Vision Course, Audio Course. All at huggingface.co/learn/.", "tags": ["huggingface", "courses", "learning"], "importance": 7, "created": "2026-04-03"}
|
| 9 |
+
{"id": "hf-009", "tier": "T1", "topic": "HuggingFace/Libraries", "title": "Core HF Libraries Ecosystem", "content": "Key libraries: transformers (model loading/inference), diffusers (image gen), datasets (data loading), tokenizers (fast tokenization), accelerate (distributed training), peft (parameter-efficient fine-tuning), trl (RLHF training), evaluate (metrics), sentence_transformers (embeddings), smolagents (lightweight agents).", "tags": ["huggingface", "libraries", "python"], "importance": 8, "created": "2026-04-03"}
|
| 10 |
+
{"id": "hf-010", "tier": "T1", "topic": "HuggingFace/Hardware", "title": "HF Spaces Hardware Pricing", "content": "Hardware tiers: CPU Basic (2 vCPU 16GB, FREE), CPU Upgrade (8 vCPU 32GB, $0.03/hr), ZeroGPU (dynamic, FREE Gradio-only), T4 Small (16GB VRAM, $0.40/hr), T4 Medium (16GB VRAM, $0.60/hr), L4 (24GB VRAM, $0.80/hr), L40S (48GB VRAM, $1.80/hr), 4xL40S (192GB VRAM, $8.30/hr).", "tags": ["huggingface", "hardware", "pricing"], "importance": 7, "created": "2026-04-03"}
|
| 11 |
+
{"id": "hf-011", "tier": "T2", "topic": "ZeroClaw/Integration", "title": "ZeroClaw-HF Integration Map", "content": "ZeroClaw can leverage HF in 4 ways: 1) MCP native support connects to 16+ HF Space tools, 2) HF Datasets as persistent memory store with 4-tier security, 3) HF Inference API as backup LLM provider, 4) HF Papers for automated knowledge ingestion via scheduled learning. Cortex Learning Engine orchestrates the loop.", "tags": ["zeroclaw", "huggingface", "integration", "architecture"], "importance": 10, "created": "2026-04-03"}
|