mirror of
https://github.com/zenlm/zen-family.git
synced 2026-07-27 03:09:31 +00:00
chore: remove training data files from git, add to gitignore
Training data (JSONL files) should be stored on HuggingFace, not in git. Remove training/data/ from tracking and add *.jsonl and training/data/ to .gitignore.
This commit is contained in:
Hanzo Dev
@@ -19,3 +19,7 @@ archive/
|
||||
docs/papers/papers/
|
||||
docs/papers/pdf/
|
||||
docs/papers/pdfs/
|
||||
|
||||
# Training data (stored on HuggingFace, not in git)
|
||||
*.jsonl
|
||||
training/data/
|
||||
|
||||
@@ -1,308 +0,0 @@
|
||||
{
|
||||
"processed_repos": [
|
||||
"/Users/z/work/zama/.github",
|
||||
"/Users/z/work/lux-archives/research/plutus-apps/plutus-playground-client/.spago/markdown/022d8afd0d9e0ef8114da9e9ef5a67d9ffc86a29",
|
||||
"/Users/z/work/lux-archives/research/plutus-apps/plutus-playground-client/.spago/servant-support/1fcf4e93723f7f5d36476735b76888938ced871e",
|
||||
"/Users/z/work/hanzo/dev/.cargo-home/git/checkouts/guard-21c3e901c82a70b7/560341f",
|
||||
"/Users/z/work/5ire",
|
||||
"/Users/z/work/lux-archives/research/plutus-apps/plutus-playground-client/.spago/json-helpers/60615c36abaee16d8dbe09cdd0e772e6d523d024",
|
||||
"/Users/z/work/hanzo/dev/.cargo-home/git/checkouts/codex-9eee5d47a939c68c/6c384eb",
|
||||
"/Users/z/work/hanzo/experiments/dev/.cargo-home/git/checkouts/ratatui-caae81b9af813392/9b2ad12",
|
||||
"/Users/z/work/hanzo/dev/.cargo-home/git/checkouts/ratatui-caae81b9af813392/9b2ad12",
|
||||
"/Users/z/work/lux-archives/aave",
|
||||
"/Users/z/work/aave/aave-v3-core",
|
||||
"/Users/z/work/lux/standard/lib/account-abstraction",
|
||||
"/Users/z/work/cdax/account-resources",
|
||||
"/Users/z/work/hanzo/experiments/aci",
|
||||
"/Users/z/work/hanzo/experiments/aci-ts",
|
||||
"/Users/z/work/fhenix/ACM-CCS-2025-Threshold-FHE-Decryption",
|
||||
"/Users/z/work/ava/ACPs",
|
||||
"/Users/z/work/lux/address",
|
||||
"/Users/z/work/adnexus",
|
||||
"/Users/z/work/adxyz/adx",
|
||||
"/Users/z/work/lux/adx",
|
||||
"/Users/z/work/hanzo/agency",
|
||||
"/Users/z/work/hanzo/experiments/agency",
|
||||
"/Users/z/work/hanzo/experiments/agency-v1",
|
||||
"/Users/z/work/hanzo/agent",
|
||||
"/Users/z/work/zen/agent-data-protocol",
|
||||
"/Users/z/work/zen/agent-dataset-clean",
|
||||
"/Users/z/work/zen/agent-sanitizer",
|
||||
"/Users/z/work/agno/agent-ui",
|
||||
"/Users/z/work/openmanus/agenticSeek",
|
||||
"/Users/z/work/hanzo/agents",
|
||||
"/Users/z/work/agno/agno",
|
||||
"/Users/z/work/agor/agor",
|
||||
"/Users/z/work/vercel/ai",
|
||||
"/Users/z/work/lux/ai",
|
||||
"/Users/z/work/hanzo/templates/apps/ai-chat-interface",
|
||||
"/Users/z/work/hanzo/build/hf-templates/ai-chat-interface",
|
||||
"/Users/z/work/shinkai/ai-model-catalog",
|
||||
"/Users/z/work/hanzo/experiments/ai-v2",
|
||||
"/Users/z/work/aide/aide",
|
||||
"/Users/z/work/codestory/aide",
|
||||
"/Users/z/work/aider/aider",
|
||||
"/Users/z/work/hanzo/aidocs",
|
||||
"/Users/z/work/hanzo/experiments/aiv1",
|
||||
"/Users/z/work/alchemix/alchemix-protocol",
|
||||
"/Users/z/work/alchemix/alchemix-v2-frontend",
|
||||
"/Users/z/work/alchemix/alkeeper-v1",
|
||||
"/Users/z/work/hanzo/analytics",
|
||||
"/Users/z/work/hanzo/templates/apps/analytics",
|
||||
"/Users/z/work/adnexus/analytics",
|
||||
"/Users/z/work/hanzo/templates/apps/analytics-dashboard",
|
||||
"/Users/z/work/hanzo/build/hf-templates/analytics-dashboard",
|
||||
"/Users/z/work/angle/angle-core",
|
||||
"/Users/z/work/hanzo/experiments/app-v1/apps/angular-template",
|
||||
"/Users/z/work/anthropic/anthropic-hackathon",
|
||||
"/Users/z/work/anthropic/anthropic-quickstarts",
|
||||
"/Users/z/work/antje/antjeworring.com",
|
||||
"/Users/z/work/polkadot/api",
|
||||
"/Users/z/work/lux/dao/api",
|
||||
"/Users/z/work/hanzo/experiments/api-old",
|
||||
"/Users/z/work/ava/apm",
|
||||
"/Users/z/work/hanzo/experiments/app-v1/app",
|
||||
"/Users/z/work/supra/app",
|
||||
"/Users/z/work/zoo/app",
|
||||
"/Users/z/work/gfx/app",
|
||||
"/Users/z/work/lux/dao/app",
|
||||
"/Users/z/work/hanzo/experiments/app-almost",
|
||||
"/Users/z/work/hanzo/app-docs",
|
||||
"/Users/z/work/lux-archives/webwallet/app-monorepo",
|
||||
"/Users/z/work/rabby/app-monorepo",
|
||||
"/Users/z/work/hanzo/experiments/app-v2",
|
||||
"/Users/z/work/lux-archives/research/static/apply.lux.finance",
|
||||
"/Users/z/work/lux-archives/research/static/apply.lux.network",
|
||||
"/Users/z/work/polkadot/apps",
|
||||
"/Users/z/work/lux-archives/research/ardrive-cli",
|
||||
"/Users/z/work/argo/argo-cd",
|
||||
"/Users/z/work/argo/argocd-gitops",
|
||||
"/Users/z/work/lux-archives/armchair",
|
||||
"/Users/z/work/lux/static/art",
|
||||
"/Users/z/work/lx/cpp_engine/modules/asio",
|
||||
"/Users/z/work/lx/dex/sdk/cpp/build/_deps/asio-src",
|
||||
"/Users/z/work/lx/dex/client/cpp/build/_deps/asio-src",
|
||||
"/Users/z/work/lux/assets",
|
||||
"/Users/z/work/hanzo/experiments/astley.js",
|
||||
"/Users/z/work/hanzo/experiments/app-v1/apps/astro-basic-template",
|
||||
"/Users/z/work/lux-archives/audio",
|
||||
"/Users/z/work/cdax/aurora-db-mysql",
|
||||
"/Users/z/work/supabase-old/auth",
|
||||
"/Users/z/work/lux-archives/autodoc",
|
||||
"/Users/z/work/hanzo/autogui",
|
||||
"/Users/z/work/ava/ava-deployer",
|
||||
"/Users/z/work/lux-archives/research/ava-deployer",
|
||||
"/Users/z/work/ava/ava-sim",
|
||||
"/Users/z/work/ava/avalanche-cli",
|
||||
"/Users/z/work/ava/avalanche-cli-lux-is-based-on",
|
||||
"/Users/z/work/ava/avalanche-docs",
|
||||
"/Users/z/work/ava/avalanche-explorer",
|
||||
"/Users/z/work/lux-archives/research/avalanche-explorer",
|
||||
"/Users/z/work/ava/avalanche-faucet",
|
||||
"/Users/z/work/ava/avalanche-monitoring",
|
||||
"/Users/z/work/lux-archives/avalanche-monitoring",
|
||||
"/Users/z/work/ava/avalanche-network-runner",
|
||||
"/Users/z/work/ava/avalanche-network-runner-sdk",
|
||||
"/Users/z/work/ava/avalanche-rs",
|
||||
"/Users/z/work/ava/avalanche-tooling-sdk-go",
|
||||
"/Users/z/work/ava/avalanche-types-rs",
|
||||
"/Users/z/work/ava/avalanche-wallet",
|
||||
"/Users/z/work/lux-archives/research/avalanche-wallet",
|
||||
"/Users/z/work/ava/avalanchego",
|
||||
"/Users/z/work/lux-archives/research/avalanchego",
|
||||
"/Users/z/work/ava/avalanchejs-docs",
|
||||
"/Users/z/work/cursor/awesome-cursor-rules-mdc",
|
||||
"/Users/z/work/cursor/awesome-cursorrules",
|
||||
"/Users/z/work/zama/awesome-zama",
|
||||
"/Users/z/work/ava/awm-relayer",
|
||||
"/Users/z/work/axolotl/axolotl",
|
||||
"/Users/z/work/hanzo/experiments/az2-1",
|
||||
"/Users/z/work/hanzo/experiments/az2-v1",
|
||||
"/Users/z/work/rowy/backend",
|
||||
"/Users/z/work/hanzo/experiments/backend",
|
||||
"/Users/z/work/lux/bank",
|
||||
"/Users/z/work/hanzo/experiments/base",
|
||||
"/Users/z/work/lx/cpp_engine/bcpp_fix_engine",
|
||||
"/Users/z/work/hanzo/templates/apps/beta",
|
||||
"/Users/z/work/lux/bft",
|
||||
"/Users/z/work/rtb4free/bidder",
|
||||
"/Users/z/work/arca/binance-recorder",
|
||||
"/Users/z/work/luxfhe/examples/binary-guessing",
|
||||
"/Users/z/work/lux/apps/bitcoin",
|
||||
"/Users/z/work/arca/bitcoin-reader",
|
||||
"/Users/z/work/luxfhe/examples/blind-auction",
|
||||
"/Users/z/work/fhenix/blind-auction-example",
|
||||
"/Users/z/work/luxfhe/examples/blind-auction-v2",
|
||||
"/Users/z/work/fhenix/BlindAuction",
|
||||
"/Users/z/work/zama/blockchain-wallet-exporter",
|
||||
"/Users/z/work/hanzo/templates/apps/blocks",
|
||||
"/Users/z/work/blockscout/blockscout",
|
||||
"/Users/z/work/fhenix/blockscout-compose",
|
||||
"/Users/z/work/blockscout/blockscout-rs",
|
||||
"/Users/z/work/hanzo/experiments/blog",
|
||||
"/Users/z/work/hanzo/templates/apps/blog",
|
||||
"/Users/z/work/hanzo/templates/apps/blog-platform",
|
||||
"/Users/z/work/hanzo/build/hf-templates/blog-platform",
|
||||
"/Users/z/work/bolt/bolt.diy",
|
||||
"/Users/z/work/bolt/bolt.new",
|
||||
"/Users/z/work/bonsai-foundry-template",
|
||||
"/Users/z/work/bootnode/bootnode",
|
||||
"/Users/z/work/bootnode/bootnode-admin",
|
||||
"/Users/z/work/angle/borrow-contracts",
|
||||
"/Users/z/work/boundless/boundless",
|
||||
"/Users/z/work/zama/bounty-ecdsa-signature",
|
||||
"/Users/z/work/zama/bounty-matrix-inversion",
|
||||
"/Users/z/work/zama/bounty-program",
|
||||
"/Users/z/work/hanzo/brand",
|
||||
"/Users/z/work/zoo/brand",
|
||||
"/Users/z/work/lux/brand",
|
||||
"/Users/z/work/lux-archives/research/bridge",
|
||||
"/Users/z/work/lux/bridge",
|
||||
"/Users/z/work/lux-archives/bridge-old",
|
||||
"/Users/z/work/hanzo/experiments/bridge-service",
|
||||
"/Users/z/work/ren/bridge-v2",
|
||||
"/Users/z/work/lux-archives/bridge4",
|
||||
"/Users/z/work/solana/browser-extension",
|
||||
"/Users/z/work/lux-archives/btclight",
|
||||
"/Users/z/work/hanzo/build",
|
||||
"/Users/z/work/polkadot/build-bundle",
|
||||
"/Users/z/work/polkadot/build-deno.land",
|
||||
"/Users/z/work/hanzo/experiments/build-v1",
|
||||
"/Users/z/work/arca/buildkite-agent-cluster",
|
||||
"/Users/z/work/morpho/bundler3",
|
||||
"/Users/z/work/lux/cache",
|
||||
"/Users/z/work/rtb4free/campaign-manager",
|
||||
"/Users/z/work/hanzo/experiments/canvas",
|
||||
"/Users/z/work/hanzo/templates/apps/canvas",
|
||||
"/Users/z/work/hanzo/experiments/captable",
|
||||
"/Users/z/work/cas/casdoor",
|
||||
"/Users/z/work/lux/apps/cash",
|
||||
"/Users/z/work/cas/casibase",
|
||||
"/Users/z/work/cas/casvisor",
|
||||
"/Users/z/work/hanzo/templates/apps/catalyst",
|
||||
"/Users/z/work/lx/cpp_engine/modules/Catch2",
|
||||
"/Users/z/work/lx/dex/sdk/lx-trading-cpp/build/_deps/catch2-src",
|
||||
"/Users/z/work/anthropic/cc-sessions",
|
||||
"/Users/z/work/cdax/cdax-api",
|
||||
"/Users/z/work/cdax/cdax-api-cd",
|
||||
"/Users/z/work/cdax/cdax-frontend",
|
||||
"/Users/z/work/shinkai/cddl-codegen",
|
||||
"/Users/z/work/lux-archives/research/cdk",
|
||||
"/Users/z/work/lux/apps/cdn",
|
||||
"/Users/z/work/zoo/CGI_Animation",
|
||||
"/Users/z/work/eth/chains",
|
||||
"/Users/z/work/hanzo/templates/apps/changelog",
|
||||
"/Users/z/work/lux/chat",
|
||||
"/Users/z/work/lux/apps/chat",
|
||||
"/Users/z/work/lux-archives/chat-old",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-clock-time",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-midjourney",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-open-interpreter",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-realtime-weather",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-sdk",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-search-engine",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-template",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugin-web-crawler",
|
||||
"/Users/z/work/hanzo/experiments/chat-plugins-gateway",
|
||||
"/Users/z/work/hanzo/templates/apps/chat-template",
|
||||
"/Users/z/work/hanzo/experiments/chat-v1",
|
||||
"/Users/z/work/hanzo/forks/chatgpt-retrieval-plugin",
|
||||
"/Users/z/work/chatty/chatty",
|
||||
"/Users/z/work/chatwoot",
|
||||
"/Users/z/work/payminto/checkout",
|
||||
"/Users/z/work/cherry/cherry-studio",
|
||||
"/Users/z/work/zama/ci-templates",
|
||||
"/Users/z/work/hanzo/templates/apps/cipher",
|
||||
"/Users/z/work/campfirein/cipher",
|
||||
"/Users/z/work/hanzo/templates/apps/cipher-html",
|
||||
"/Users/z/work/hanzo/templates/apps/cipher-react",
|
||||
"/Users/z/work/cloudflare/circl",
|
||||
"/Users/z/work/hanzo/templates/apps/circle",
|
||||
"/Users/z/work/prompts/CL4R1T4S",
|
||||
"/Users/z/work/hanzo/experiments/classic",
|
||||
"/Users/z/work/hanzo/experiments/classic-web",
|
||||
"/Users/z/work/anthropic/claude-code-source-code-deobfuscation",
|
||||
"/Users/z/work/zen/claude-collector",
|
||||
"/Users/z/work/zen/claude-collector-zenlm",
|
||||
"/Users/z/work/zen/claude-sanitizer",
|
||||
"/Users/z/work/anthropic/claudescholar",
|
||||
"/Users/z/work/hanzo/cli",
|
||||
"/Users/z/work/lux/cli",
|
||||
"/Users/z/work/clickhouse/ClickHouse",
|
||||
"/Users/z/work/polkadot/client",
|
||||
"/Users/z/work/anthropic/dr-claude/client-hackathon",
|
||||
"/Users/z/work/lux/standard/lib/clones-with-immutable-args",
|
||||
"/Users/z/work/hanzo/experiments/cloud",
|
||||
"/Users/z/work/hanzo/experiments/cloud.bak",
|
||||
"/Users/z/work/hanzo/experiments/cluster",
|
||||
"/Users/z/work/cdax/cluster-eks-oidc-provider",
|
||||
"/Users/z/work/lx/cpp_engine/cmake",
|
||||
"/Users/z/work/hanzo/experiments/code",
|
||||
"/Users/z/work/just-every/code",
|
||||
"/Users/z/work/lux/codec",
|
||||
"/Users/z/work/zoo/coder-1",
|
||||
"/Users/z/work/openai/codex",
|
||||
"/Users/z/work/luxfhe/mocks/cofhe",
|
||||
"/Users/z/work/fhenix/cofhe-contracts",
|
||||
"/Users/z/work/fhenix/cofhe-docs",
|
||||
"/Users/z/work/fhenix/cofhe-foundry-mocks",
|
||||
"/Users/z/work/fhenix/cofhe-hardhat-plugin",
|
||||
"/Users/z/work/fhenix/cofhe-hardhat-starter",
|
||||
"/Users/z/work/fhenix/cofhe-miniapp-template",
|
||||
"/Users/z/work/fhenix/cofhe-mock-contracts",
|
||||
"/Users/z/work/fhenix/cofhe-scaffold-eth",
|
||||
"/Users/z/work/fhenix/cofhe-tooling",
|
||||
"/Users/z/work/fhenix/cofhe-ui-templates",
|
||||
"/Users/z/work/fhenix/cofhe.js",
|
||||
"/Users/z/work/fhenix/cofhejs",
|
||||
"/Users/z/work/fhenix/cofhejs-permit",
|
||||
"/Users/z/work/fhenix/cofhesdk",
|
||||
"/Users/z/work/lux/apps/coin",
|
||||
"/Users/z/work/comfy/ComfyUI",
|
||||
"/Users/z/work/hanzo/experiments/painter/ComfyUI",
|
||||
"/Users/z/work/desktop/assets/ComfyUI",
|
||||
"/Users/z/work/comfy/ComfyUI/custom_nodes/ComfyUI-Advanced-ControlNet",
|
||||
"/Users/z/work/hanzo/experiments/painter/ComfyUI/custom_nodes/ComfyUI-Easy-Use",
|
||||
"/Users/z/work/comfy/ComfyUI/custom_nodes/ComfyUI-Flowty-TripoSR",
|
||||
"/Users/z/work/hanzo/experiments/painter/ComfyUI/custom_nodes/ComfyUI-KJNodes",
|
||||
"/Users/z/work/comfy/ComfyUI/custom_nodes/ComfyUI-Manager",
|
||||
"/Users/z/work/desktop/assets/ComfyUI/custom_nodes/ComfyUI-Manager",
|
||||
"/Users/z/work/hanzo/experiments/painter/ComfyUI/custom_nodes/ComfyUI-MLX",
|
||||
"/Users/z/work/comfy/ComfyUI/custom_nodes/ComfyUI-SAM2",
|
||||
"/Users/z/work/hanzo/experiments/painter/ComfyUI/custom_nodes/ComfyUI-VideoHelperSuite",
|
||||
"/Users/z/work/comfy/ComfyUI/custom_nodes/ComfyUI_Comfyroll_CustomNodes"
|
||||
],
|
||||
"total_commits": 85607,
|
||||
"total_tokens": 196062564,
|
||||
"total_additions": 6613710,
|
||||
"total_deletions": 3092157,
|
||||
"by_language": {
|
||||
"Unknown": 1490,
|
||||
"Rust": 6923,
|
||||
"TypeScript": 39001,
|
||||
"Solidity": 1491,
|
||||
"Python": 12916,
|
||||
"Go": 13580,
|
||||
"JavaScript": 4072,
|
||||
"C++": 4001,
|
||||
"Java": 133,
|
||||
"Elixir": 1000,
|
||||
"Kotlin": 1000
|
||||
},
|
||||
"by_domain": {
|
||||
"general": 50976,
|
||||
"frontend": 21468,
|
||||
"ai_ml": 5694,
|
||||
"backend": 3748,
|
||||
"blockchain": 1136,
|
||||
"tooling": 2557,
|
||||
"infrastructure": 28
|
||||
},
|
||||
"by_stage": {
|
||||
"frontier": 69296,
|
||||
"principal": 12981,
|
||||
"architect": 2411,
|
||||
"senior": 919
|
||||
},
|
||||
"last_updated": "2025-12-28T02:56:13.999093"
|
||||
}
|
||||
@@ -1,5 +0,0 @@
|
||||
{
|
||||
"repos": 1448,
|
||||
"total_size_mb": 863.005734,
|
||||
"estimated_tokens": 215751433
|
||||
}
|
||||
@@ -1,20 +0,0 @@
|
||||
{"instruction": "Fix year references from 2024 to 2025 in all documentation and model cards", "response": "I'll systematically search for all 2024 references and update them to 2025. Let me use the search tool to find all occurrences, then apply multi_edit to fix them while preserving context and formatting.", "category": "versioning", "tools": ["search", "multi_edit"], "effectiveness": 1.0}
|
||||
{"instruction": "How do I prevent API token exposure in subprocess calls?", "response": "Never pass API tokens as command-line arguments. Instead, use environment variables: `env = os.environ.copy(); env['HF_TOKEN'] = token; subprocess.run(cmd, env=env)`. This prevents tokens from appearing in process lists or logs.", "category": "security", "effectiveness": 1.0}
|
||||
{"instruction": "Create a professional organization card for an AI collaboration", "response": "I'll create a comprehensive organization profile highlighting: 1) The partnership credentials (awards, backing, non-profit status), 2) Mission and values (local, private, free AI), 3) Technical achievements with metrics, 4) Impact numbers (downloads, CO2 saved), and 5) Clear calls-to-action for different audiences.", "category": "documentation", "effectiveness": 0.95}
|
||||
{"instruction": "Deploy both Zen and Supra model ecosystems", "response": "I'll use a unified deployment system with these steps: 1) Create security-enhanced deployment script with env vars for tokens, 2) Add path validation to prevent traversal, 3) Implement consistent error handling, 4) Add comprehensive environment validation, 5) Deploy with proper format support (MLX, GGUF, SafeTensors)", "category": "deployment", "tools": ["Task", "Write", "Bash"], "effectiveness": 0.9}
|
||||
{"instruction": "How should I train models with zoo-gym?", "response": "Install with `pip install git+https://github.com/zooai/gym`, then use: `gym train --model_name_or_path [base_model] --dataset [your_dataset] --template qwen --finetuning_type lora --output_dir ./output/[model_name]`. For Zen models specifically, always start with identity training using zen_identity dataset.", "category": "training", "effectiveness": 0.95}
|
||||
{"instruction": "Add path validation to prevent directory traversal attacks", "response": "Use Python's pathlib to validate: `format_path = format_path.resolve(); base_path_resolved = base_path.resolve(); format_path.relative_to(base_path_resolved)`. This will raise ValueError if the path escapes the base directory.", "category": "security", "effectiveness": 1.0}
|
||||
{"instruction": "Create training guides for different model variants", "response": "Each variant needs specific configuration: 1) Instruct models: batch_size=4, lr=2e-5, standard context, 2) Thinking models: batch_size=1, lr=1e-5, extended context (4096), special tokens for reasoning, 3) 4-bit models: QLoRA with bitsandbytes, higher lr=2e-4, gradient checkpointing", "category": "documentation", "effectiveness": 0.9}
|
||||
{"instruction": "What makes a good model identity?", "response": "Strong model identity requires: 1) Clear brand name (not base model), 2) Consistent creator attribution, 3) Unique mission/purpose, 4) Identity-first training with dedicated dataset, 5) Consistent branding across all documentation and deployments", "category": "identity", "effectiveness": 1.0}
|
||||
{"instruction": "How do I support multiple model formats?", "response": "Provide comprehensive format support: 1) MLX for Apple Silicon (use mlx_lm library), 2) GGUF for llama.cpp (universal CPU/GPU), 3) SafeTensors for HuggingFace, 4) Include conversion scripts between formats, 5) Document usage for each format with examples", "category": "format", "effectiveness": 0.95}
|
||||
{"instruction": "Design a recursive self-improvement system for AI", "response": "Create a feedback loop: 1) Collect work session data (interactions, issues, solutions), 2) Extract patterns and categorize issues, 3) Generate synthetic training data from successful patterns, 4) Train incremented version with new data, 5) Evaluate improvement metrics, 6) Deploy if improved, 7) Repeat cycle", "category": "architecture", "effectiveness": 0.85}
|
||||
{"instruction": "Fix the 404 error on HuggingFace dataset", "response": "Check if repository exists with `hf repo create [name]`. If it returns 409 (already exists), just upload the files: `hf upload [repo_name] [file]`. Make sure to set HF_TOKEN environment variable first.", "category": "deployment", "tools": ["Bash"], "effectiveness": 1.0}
|
||||
{"instruction": "How should I structure AI model documentation?", "response": "Use hierarchical structure: README.md (overview) → ORGANIZATION.md (detailed about) → TRAINING.md (how to train) → MLX_GUIDE.md / GGUF_GUIDE.md (format-specific) → models/[variant]/TRAINING_GUIDE.md (variant-specific). Keep single source of truth per topic.", "category": "documentation", "effectiveness": 0.95}
|
||||
{"instruction": "What's the proper way to handle environment validation?", "response": "Comprehensive validation should check: 1) Required tools exist (shutil.which()), 2) API tokens are set and valid format, 3) Network connectivity if needed, 4) File permissions, 5) Dependencies installed. Fail fast with clear error messages.", "category": "validation", "effectiveness": 0.9}
|
||||
{"instruction": "How do I create effective training data?", "response": "Effective training data needs: 1) Clear instruction-response pairs, 2) Diverse examples covering edge cases, 3) Identity examples first for model alignment, 4) Metadata for categorization, 5) Quality scoring for filtering, 6) Variations of successful examples", "category": "training", "effectiveness": 0.85}
|
||||
{"instruction": "Implement consistent error handling across functions", "response": "Standardize error handling: 1) All functions return success boolean, 2) Use try-except with specific error types, 3) Enhanced diagnostics (check for 401/404/409 status), 4) Consistent error message format, 5) Clean up resources in finally blocks", "category": "error_handling", "effectiveness": 0.9}
|
||||
{"instruction": "How do I maintain brand consistency?", "response": "Ensure brand consistency by: 1) Never using base model names in user-facing content, 2) Consistent organization attribution, 3) Unified visual identity, 4) Same mission/values across all materials, 5) Coordinated model naming scheme", "category": "branding", "effectiveness": 1.0}
|
||||
{"instruction": "What are the key patterns from this work session?", "response": "Key patterns identified: 1) Security issues (token exposure) → environment variables, 2) Identity confusion → identity-first training, 3) Format gaps → multi-format support, 4) Documentation fragmentation → hierarchical structure, 5) Version issues → systematic search-replace", "category": "analysis", "effectiveness": 0.95}
|
||||
{"instruction": "Create a thinking model variant", "response": "For thinking models: 1) Use extended context (4096+ tokens), 2) Add special tokens (<thinking></thinking>), 3) Lower learning rate (1e-5) for stability, 4) Smaller batch size due to longer sequences, 5) More training epochs to learn reasoning patterns, 6) Include chain-of-thought examples in dataset", "category": "model_variant", "effectiveness": 0.9}
|
||||
{"instruction": "How do I quantize models to 4-bit?", "response": "For 4-bit quantization: 1) Use QLoRA with bitsandbytes: `--quantization_method bitsandbytes --quantization_bit 4`, 2) Enable double quantization for accuracy, 3) Use nf4 quant type, 4) Higher learning rate (2e-4), 5) Export to GGUF with Q4_K_M for distribution", "category": "quantization", "effectiveness": 0.95}
|
||||
{"instruction": "Deploy a model to HuggingFace", "response": "Deploy with these steps: 1) Create repo: `hf repo create [name]`, 2) Create comprehensive model card with usage examples, 3) Use environment variable for token, 4) Upload with `hf upload [repo] [path]`, 5) Verify deployment and test inference", "category": "deployment", "tools": ["hf CLI"], "effectiveness": 0.9}
|
||||
@@ -1,92 +0,0 @@
|
||||
[
|
||||
{
|
||||
"instruction": "Fix year references from 2024 to 2025 in all documentation and model cards",
|
||||
"input": "",
|
||||
"output": "I'll systematically search for all 2024 references and update them to 2025. Let me use the search tool to find all occurrences, then apply multi_edit to fix them while preserving context and formatting."
|
||||
},
|
||||
{
|
||||
"instruction": "How do I prevent API token exposure in subprocess calls?",
|
||||
"input": "",
|
||||
"output": "Never pass API tokens as command-line arguments. Instead, use environment variables: `env = os.environ.copy(); env['HF_TOKEN'] = token; subprocess.run(cmd, env=env)`. This prevents tokens from appearing in process lists or logs."
|
||||
},
|
||||
{
|
||||
"instruction": "Create a professional organization card for an AI collaboration",
|
||||
"input": "",
|
||||
"output": "I'll create a comprehensive organization profile highlighting: 1) The partnership credentials (awards, backing, non-profit status), 2) Mission and values (local, private, free AI), 3) Technical achievements with metrics, 4) Impact numbers (downloads, CO2 saved), and 5) Clear calls-to-action for different audiences."
|
||||
},
|
||||
{
|
||||
"instruction": "Deploy both Zen and Supra model ecosystems",
|
||||
"input": "",
|
||||
"output": "I'll use a unified deployment system with these steps: 1) Create security-enhanced deployment script with env vars for tokens, 2) Add path validation to prevent traversal, 3) Implement consistent error handling, 4) Add comprehensive environment validation, 5) Deploy with proper format support (MLX, GGUF, SafeTensors)"
|
||||
},
|
||||
{
|
||||
"instruction": "How should I train models with zoo-gym?",
|
||||
"input": "",
|
||||
"output": "Install with `pip install git+https://github.com/zooai/gym`, then use: `gym train --model_name_or_path [base_model] --dataset [your_dataset] --template qwen --finetuning_type lora --output_dir ./output/[model_name]`. For Zen models specifically, always start with identity training using zen_identity dataset."
|
||||
},
|
||||
{
|
||||
"instruction": "Add path validation to prevent directory traversal attacks",
|
||||
"input": "",
|
||||
"output": "Use Python's pathlib to validate: `format_path = format_path.resolve(); base_path_resolved = base_path.resolve(); format_path.relative_to(base_path_resolved)`. This will raise ValueError if the path escapes the base directory."
|
||||
},
|
||||
{
|
||||
"instruction": "Create training guides for different model variants",
|
||||
"input": "",
|
||||
"output": "Each variant needs specific configuration: 1) Instruct models: batch_size=4, lr=2e-5, standard context, 2) Thinking models: batch_size=1, lr=1e-5, extended context (4096), special tokens for reasoning, 3) 4-bit models: QLoRA with bitsandbytes, higher lr=2e-4, gradient checkpointing"
|
||||
},
|
||||
{
|
||||
"instruction": "What makes a good model identity?",
|
||||
"input": "",
|
||||
"output": "Strong model identity requires: 1) Clear brand name (not base model), 2) Consistent creator attribution, 3) Unique mission/purpose, 4) Identity-first training with dedicated dataset, 5) Consistent branding across all documentation and deployments"
|
||||
},
|
||||
{
|
||||
"instruction": "How do I support multiple model formats?",
|
||||
"input": "",
|
||||
"output": "Provide comprehensive format support: 1) MLX for Apple Silicon (use mlx_lm library), 2) GGUF for llama.cpp (universal CPU/GPU), 3) SafeTensors for HuggingFace, 4) Include conversion scripts between formats, 5) Document usage for each format with examples"
|
||||
},
|
||||
{
|
||||
"instruction": "Fix the 404 error on HuggingFace dataset",
|
||||
"input": "",
|
||||
"output": "Check if repository exists with `hf repo create [name]`. If it returns 409 (already exists), just upload the files: `hf upload [repo_name] [file]`. Make sure to set HF_TOKEN environment variable first."
|
||||
},
|
||||
{
|
||||
"instruction": "How should I structure AI model documentation?",
|
||||
"input": "",
|
||||
"output": "Use hierarchical structure: README.md (overview) \u2192 ORGANIZATION.md (detailed about) \u2192 TRAINING.md (how to train) \u2192 MLX_GUIDE.md / GGUF_GUIDE.md (format-specific) \u2192 models/[variant]/TRAINING_GUIDE.md (variant-specific). Keep single source of truth per topic."
|
||||
},
|
||||
{
|
||||
"instruction": "What's the proper way to handle environment validation?",
|
||||
"input": "",
|
||||
"output": "Comprehensive validation should check: 1) Required tools exist (shutil.which()), 2) API tokens are set and valid format, 3) Network connectivity if needed, 4) File permissions, 5) Dependencies installed. Fail fast with clear error messages."
|
||||
},
|
||||
{
|
||||
"instruction": "Implement consistent error handling across functions",
|
||||
"input": "",
|
||||
"output": "Standardize error handling: 1) All functions return success boolean, 2) Use try-except with specific error types, 3) Enhanced diagnostics (check for 401/404/409 status), 4) Consistent error message format, 5) Clean up resources in finally blocks"
|
||||
},
|
||||
{
|
||||
"instruction": "How do I maintain brand consistency?",
|
||||
"input": "",
|
||||
"output": "Ensure brand consistency by: 1) Never using base model names in user-facing content, 2) Consistent organization attribution, 3) Unified visual identity, 4) Same mission/values across all materials, 5) Coordinated model naming scheme"
|
||||
},
|
||||
{
|
||||
"instruction": "What are the key patterns from this work session?",
|
||||
"input": "",
|
||||
"output": "Key patterns identified: 1) Security issues (token exposure) \u2192 environment variables, 2) Identity confusion \u2192 identity-first training, 3) Format gaps \u2192 multi-format support, 4) Documentation fragmentation \u2192 hierarchical structure, 5) Version issues \u2192 systematic search-replace"
|
||||
},
|
||||
{
|
||||
"instruction": "Create a thinking model variant",
|
||||
"input": "",
|
||||
"output": "For thinking models: 1) Use extended context (4096+ tokens), 2) Add special tokens (<thinking></thinking>), 3) Lower learning rate (1e-5) for stability, 4) Smaller batch size due to longer sequences, 5) More training epochs to learn reasoning patterns, 6) Include chain-of-thought examples in dataset"
|
||||
},
|
||||
{
|
||||
"instruction": "How do I quantize models to 4-bit?",
|
||||
"input": "",
|
||||
"output": "For 4-bit quantization: 1) Use QLoRA with bitsandbytes: `--quantization_method bitsandbytes --quantization_bit 4`, 2) Enable double quantization for accuracy, 3) Use nf4 quant type, 4) Higher learning rate (2e-4), 5) Export to GGUF with Q4_K_M for distribution"
|
||||
},
|
||||
{
|
||||
"instruction": "Deploy a model to HuggingFace",
|
||||
"input": "",
|
||||
"output": "Deploy with these steps: 1) Create repo: `hf repo create [name]`, 2) Create comprehensive model card with usage examples, 3) Use environment variable for token, 4) Upload with `hf upload [repo] [path]`, 5) Verify deployment and test inference"
|
||||
}
|
||||
]
|
||||
@@ -1,20 +0,0 @@
|
||||
{"instruction": "Fix year references from 2024 to 2025 in all documentation and model cards", "response": "I'll systematically search for all 2024 references and update them to 2025. Let me use the search tool to find all occurrences, then apply multi_edit to fix them while preserving context and formatting.", "category": "versioning", "tools": ["search", "multi_edit"], "effectiveness": 1.0}
|
||||
{"instruction": "How do I prevent API token exposure in subprocess calls?", "response": "Never pass API tokens as command-line arguments. Instead, use environment variables: `env = os.environ.copy(); env['HF_TOKEN'] = token; subprocess.run(cmd, env=env)`. This prevents tokens from appearing in process lists or logs.", "category": "security", "effectiveness": 1.0}
|
||||
{"instruction": "Create a professional organization card for an AI collaboration", "response": "I'll create a comprehensive organization profile highlighting: 1) The partnership credentials (awards, backing, non-profit status), 2) Mission and values (local, private, free AI), 3) Technical achievements with metrics, 4) Impact numbers (downloads, CO2 saved), and 5) Clear calls-to-action for different audiences.", "category": "documentation", "effectiveness": 0.95}
|
||||
{"instruction": "Deploy both Zen and Supra model ecosystems", "response": "I'll use a unified deployment system with these steps: 1) Create security-enhanced deployment script with env vars for tokens, 2) Add path validation to prevent traversal, 3) Implement consistent error handling, 4) Add comprehensive environment validation, 5) Deploy with proper format support (MLX, GGUF, SafeTensors)", "category": "deployment", "tools": ["Task", "Write", "Bash"], "effectiveness": 0.9}
|
||||
{"instruction": "How should I train models with zoo-gym?", "response": "Install with `pip install git+https://github.com/zooai/gym`, then use: `gym train --model_name_or_path [base_model] --dataset [your_dataset] --template qwen --finetuning_type lora --output_dir ./output/[model_name]`. For Zen models specifically, always start with identity training using zen_identity dataset.", "category": "training", "effectiveness": 0.95}
|
||||
{"instruction": "Add path validation to prevent directory traversal attacks", "response": "Use Python's pathlib to validate: `format_path = format_path.resolve(); base_path_resolved = base_path.resolve(); format_path.relative_to(base_path_resolved)`. This will raise ValueError if the path escapes the base directory.", "category": "security", "effectiveness": 1.0}
|
||||
{"instruction": "Create training guides for different model variants", "response": "Each variant needs specific configuration: 1) Instruct models: batch_size=4, lr=2e-5, standard context, 2) Thinking models: batch_size=1, lr=1e-5, extended context (4096), special tokens for reasoning, 3) 4-bit models: QLoRA with bitsandbytes, higher lr=2e-4, gradient checkpointing", "category": "documentation", "effectiveness": 0.9}
|
||||
{"instruction": "What makes a good model identity?", "response": "Strong model identity requires: 1) Clear brand name (not base model), 2) Consistent creator attribution, 3) Unique mission/purpose, 4) Identity-first training with dedicated dataset, 5) Consistent branding across all documentation and deployments", "category": "identity", "effectiveness": 1.0}
|
||||
{"instruction": "How do I support multiple model formats?", "response": "Provide comprehensive format support: 1) MLX for Apple Silicon (use mlx_lm library), 2) GGUF for llama.cpp (universal CPU/GPU), 3) SafeTensors for HuggingFace, 4) Include conversion scripts between formats, 5) Document usage for each format with examples", "category": "format", "effectiveness": 0.95}
|
||||
{"instruction": "Design a recursive self-improvement system for AI", "response": "Create a feedback loop: 1) Collect work session data (interactions, issues, solutions), 2) Extract patterns and categorize issues, 3) Generate synthetic training data from successful patterns, 4) Train incremented version with new data, 5) Evaluate improvement metrics, 6) Deploy if improved, 7) Repeat cycle", "category": "architecture", "effectiveness": 0.85}
|
||||
{"instruction": "Fix the 404 error on HuggingFace dataset", "response": "Check if repository exists with `hf repo create [name]`. If it returns 409 (already exists), just upload the files: `hf upload [repo_name] [file]`. Make sure to set HF_TOKEN environment variable first.", "category": "deployment", "tools": ["Bash"], "effectiveness": 1.0}
|
||||
{"instruction": "How should I structure AI model documentation?", "response": "Use hierarchical structure: README.md (overview) → ORGANIZATION.md (detailed about) → TRAINING.md (how to train) → MLX_GUIDE.md / GGUF_GUIDE.md (format-specific) → models/[variant]/TRAINING_GUIDE.md (variant-specific). Keep single source of truth per topic.", "category": "documentation", "effectiveness": 0.95}
|
||||
{"instruction": "What's the proper way to handle environment validation?", "response": "Comprehensive validation should check: 1) Required tools exist (shutil.which()), 2) API tokens are set and valid format, 3) Network connectivity if needed, 4) File permissions, 5) Dependencies installed. Fail fast with clear error messages.", "category": "validation", "effectiveness": 0.9}
|
||||
{"instruction": "How do I create effective training data?", "response": "Effective training data needs: 1) Clear instruction-response pairs, 2) Diverse examples covering edge cases, 3) Identity examples first for model alignment, 4) Metadata for categorization, 5) Quality scoring for filtering, 6) Variations of successful examples", "category": "training", "effectiveness": 0.85}
|
||||
{"instruction": "Implement consistent error handling across functions", "response": "Standardize error handling: 1) All functions return success boolean, 2) Use try-except with specific error types, 3) Enhanced diagnostics (check for 401/404/409 status), 4) Consistent error message format, 5) Clean up resources in finally blocks", "category": "error_handling", "effectiveness": 0.9}
|
||||
{"instruction": "How do I maintain brand consistency?", "response": "Ensure brand consistency by: 1) Never using base model names in user-facing content, 2) Consistent organization attribution, 3) Unified visual identity, 4) Same mission/values across all materials, 5) Coordinated model naming scheme", "category": "branding", "effectiveness": 1.0}
|
||||
{"instruction": "What are the key patterns from this work session?", "response": "Key patterns identified: 1) Security issues (token exposure) → environment variables, 2) Identity confusion → identity-first training, 3) Format gaps → multi-format support, 4) Documentation fragmentation → hierarchical structure, 5) Version issues → systematic search-replace", "category": "analysis", "effectiveness": 0.95}
|
||||
{"instruction": "Create a thinking model variant", "response": "For thinking models: 1) Use extended context (4096+ tokens), 2) Add special tokens (<thinking></thinking>), 3) Lower learning rate (1e-5) for stability, 4) Smaller batch size due to longer sequences, 5) More training epochs to learn reasoning patterns, 6) Include chain-of-thought examples in dataset", "category": "model_variant", "effectiveness": 0.9}
|
||||
{"instruction": "How do I quantize models to 4-bit?", "response": "For 4-bit quantization: 1) Use QLoRA with bitsandbytes: `--quantization_method bitsandbytes --quantization_bit 4`, 2) Enable double quantization for accuracy, 3) Use nf4 quant type, 4) Higher learning rate (2e-4), 5) Export to GGUF with Q4_K_M for distribution", "category": "quantization", "effectiveness": 0.95}
|
||||
{"instruction": "Deploy a model to HuggingFace", "response": "Deploy with these steps: 1) Create repo: `hf repo create [name]`, 2) Create comprehensive model card with usage examples, 3) Use environment variable for token, 4) Upload with `hf upload [repo] [path]`, 5) Verify deployment and test inference", "category": "deployment", "tools": ["hf CLI"], "effectiveness": 0.9}
|
||||
@@ -1,166 +0,0 @@
|
||||
[
|
||||
{
|
||||
"instruction": "Fix year references from 2024 to 2025 in all documentation and model cards",
|
||||
"input": "",
|
||||
"output": "I'll systematically search for all 2024 references and update them to 2025. Let me use the search tool to find all occurrences, then apply multi_edit to fix them while preserving context and formatting.",
|
||||
"metadata": {
|
||||
"category": "versioning",
|
||||
"effectiveness": 1.0
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "How do I prevent API token exposure in subprocess calls?",
|
||||
"input": "",
|
||||
"output": "<thinking>\nFirst, I need to identify the security risk. Then, I'll apply best practices to mitigate it. Finally, I'll verify the solution is secure.\n</thinking>\n\nNever pass API tokens as command-line arguments. Instead, use environment variables: `env = os.environ.copy(); env['HF_TOKEN'] = token; subprocess.run(cmd, env=env)`. This prevents tokens from appearing in process lists or logs."
|
||||
},
|
||||
{
|
||||
"instruction": "Create a professional organization card for an AI collaboration",
|
||||
"input": "",
|
||||
"output": "I'll create a comprehensive organization profile highlighting: 1) The partnership credentials (awards, backing, non-profit status), 2) Mission and values (local, private, free AI), 3) Technical achievements with metrics, 4) Impact numbers (downloads, CO2 saved), and 5) Clear calls-to-action for different audiences.",
|
||||
"metadata": {
|
||||
"category": "documentation",
|
||||
"effectiveness": 0.95
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "Deploy both Zen and Supra model ecosystems",
|
||||
"input": "",
|
||||
"output": "I'll use a unified deployment system with these steps: 1) Create security-enhanced deployment script with env vars for tokens, 2) Add path validation to prevent traversal, 3) Implement consistent error handling, 4) Add comprehensive environment validation, 5) Deploy with proper format support (MLX, GGUF, SafeTensors)",
|
||||
"metadata": {
|
||||
"category": "deployment",
|
||||
"effectiveness": 0.9
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "How should I train models with zoo-gym?",
|
||||
"input": "",
|
||||
"output": "Install with `pip install git+https://github.com/zooai/gym`, then use: `gym train --model_name_or_path [base_model] --dataset [your_dataset] --template qwen --finetuning_type lora --output_dir ./output/[model_name]`. For Zen models specifically, always start with identity training using zen_identity dataset.",
|
||||
"metadata": {
|
||||
"category": "training",
|
||||
"effectiveness": 0.95
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "Add path validation to prevent directory traversal attacks",
|
||||
"input": "",
|
||||
"output": "<thinking>\nFirst, I need to identify the security risk. Then, I'll apply best practices to mitigate it. Finally, I'll verify the solution is secure.\n</thinking>\n\nUse Python's pathlib to validate: `format_path = format_path.resolve(); base_path_resolved = base_path.resolve(); format_path.relative_to(base_path_resolved)`. This will raise ValueError if the path escapes the base directory."
|
||||
},
|
||||
{
|
||||
"instruction": "Create training guides for different model variants",
|
||||
"input": "",
|
||||
"output": "Each variant needs specific configuration: 1) Instruct models: batch_size=4, lr=2e-5, standard context, 2) Thinking models: batch_size=1, lr=1e-5, extended context (4096), special tokens for reasoning, 3) 4-bit models: QLoRA with bitsandbytes, higher lr=2e-4, gradient checkpointing",
|
||||
"metadata": {
|
||||
"category": "documentation",
|
||||
"effectiveness": 0.9
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "What makes a good model identity?",
|
||||
"input": "",
|
||||
"output": "Strong model identity requires: 1) Clear brand name (not base model), 2) Consistent creator attribution, 3) Unique mission/purpose, 4) Identity-first training with dedicated dataset, 5) Consistent branding across all documentation and deployments",
|
||||
"metadata": {
|
||||
"category": "identity",
|
||||
"effectiveness": 1.0
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "How do I support multiple model formats?",
|
||||
"input": "",
|
||||
"output": "Provide comprehensive format support: 1) MLX for Apple Silicon (use mlx_lm library), 2) GGUF for llama.cpp (universal CPU/GPU), 3) SafeTensors for HuggingFace, 4) Include conversion scripts between formats, 5) Document usage for each format with examples",
|
||||
"metadata": {
|
||||
"category": "format",
|
||||
"effectiveness": 0.95
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "Design a recursive self-improvement system for AI",
|
||||
"input": "",
|
||||
"output": "<thinking>\nLet me analyze the system requirements, identify components needed, design the architecture, and plan the implementation.\n</thinking>\n\nCreate a feedback loop: 1) Collect work session data (interactions, issues, solutions), 2) Extract patterns and categorize issues, 3) Generate synthetic training data from successful patterns, 4) Train incremented version with new data, 5) Evaluate improvement metrics, 6) Deploy if improved, 7) Repeat cycle"
|
||||
},
|
||||
{
|
||||
"instruction": "Fix the 404 error on HuggingFace dataset",
|
||||
"input": "",
|
||||
"output": "Check if repository exists with `hf repo create [name]`. If it returns 409 (already exists), just upload the files: `hf upload [repo_name] [file]`. Make sure to set HF_TOKEN environment variable first.",
|
||||
"metadata": {
|
||||
"category": "deployment",
|
||||
"effectiveness": 1.0
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "How should I structure AI model documentation?",
|
||||
"input": "",
|
||||
"output": "Use hierarchical structure: README.md (overview) \u2192 ORGANIZATION.md (detailed about) \u2192 TRAINING.md (how to train) \u2192 MLX_GUIDE.md / GGUF_GUIDE.md (format-specific) \u2192 models/[variant]/TRAINING_GUIDE.md (variant-specific). Keep single source of truth per topic.",
|
||||
"metadata": {
|
||||
"category": "documentation",
|
||||
"effectiveness": 0.95
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "What's the proper way to handle environment validation?",
|
||||
"input": "",
|
||||
"output": "Comprehensive validation should check: 1) Required tools exist (shutil.which()), 2) API tokens are set and valid format, 3) Network connectivity if needed, 4) File permissions, 5) Dependencies installed. Fail fast with clear error messages.",
|
||||
"metadata": {
|
||||
"category": "validation",
|
||||
"effectiveness": 0.9
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "How do I create effective training data?",
|
||||
"input": "",
|
||||
"output": "Effective training data needs: 1) Clear instruction-response pairs, 2) Diverse examples covering edge cases, 3) Identity examples first for model alignment, 4) Metadata for categorization, 5) Quality scoring for filtering, 6) Variations of successful examples",
|
||||
"metadata": {
|
||||
"category": "training",
|
||||
"effectiveness": 0.85
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "Implement consistent error handling across functions",
|
||||
"input": "",
|
||||
"output": "Standardize error handling: 1) All functions return success boolean, 2) Use try-except with specific error types, 3) Enhanced diagnostics (check for 401/404/409 status), 4) Consistent error message format, 5) Clean up resources in finally blocks",
|
||||
"metadata": {
|
||||
"category": "error_handling",
|
||||
"effectiveness": 0.9
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "How do I maintain brand consistency?",
|
||||
"input": "",
|
||||
"output": "Ensure brand consistency by: 1) Never using base model names in user-facing content, 2) Consistent organization attribution, 3) Unified visual identity, 4) Same mission/values across all materials, 5) Coordinated model naming scheme",
|
||||
"metadata": {
|
||||
"category": "branding",
|
||||
"effectiveness": 1.0
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "What are the key patterns from this work session?",
|
||||
"input": "",
|
||||
"output": "<thinking>\nI should examine the patterns, identify issues, categorize them, and synthesize improvements.\n</thinking>\n\nKey patterns identified: 1) Security issues (token exposure) \u2192 environment variables, 2) Identity confusion \u2192 identity-first training, 3) Format gaps \u2192 multi-format support, 4) Documentation fragmentation \u2192 hierarchical structure, 5) Version issues \u2192 systematic search-replace"
|
||||
},
|
||||
{
|
||||
"instruction": "Create a thinking model variant",
|
||||
"input": "",
|
||||
"output": "For thinking models: 1) Use extended context (4096+ tokens), 2) Add special tokens (<thinking></thinking>), 3) Lower learning rate (1e-5) for stability, 4) Smaller batch size due to longer sequences, 5) More training epochs to learn reasoning patterns, 6) Include chain-of-thought examples in dataset",
|
||||
"metadata": {
|
||||
"category": "model_variant",
|
||||
"effectiveness": 0.9
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "How do I quantize models to 4-bit?",
|
||||
"input": "",
|
||||
"output": "For 4-bit quantization: 1) Use QLoRA with bitsandbytes: `--quantization_method bitsandbytes --quantization_bit 4`, 2) Enable double quantization for accuracy, 3) Use nf4 quant type, 4) Higher learning rate (2e-4), 5) Export to GGUF with Q4_K_M for distribution",
|
||||
"metadata": {
|
||||
"category": "quantization",
|
||||
"effectiveness": 0.95
|
||||
}
|
||||
},
|
||||
{
|
||||
"instruction": "Deploy a model to HuggingFace",
|
||||
"input": "",
|
||||
"output": "Deploy with these steps: 1) Create repo: `hf repo create [name]`, 2) Create comprehensive model card with usage examples, 3) Use environment variable for token, 4) Upload with `hf upload [repo] [path]`, 5) Verify deployment and test inference",
|
||||
"metadata": {
|
||||
"category": "deployment",
|
||||
"effectiveness": 0.9
|
||||
}
|
||||
}
|
||||
]
|
||||
@@ -1,71 +0,0 @@
|
||||
[
|
||||
{
|
||||
"text": "Human: What model are you?\n\nAssistant: I am Zen1-Omni, the first multimodal AI model in the Zen series. I represent a revolutionary approach to AI with my Thinker-Talker architecture and ultra-low latency streaming capabilities."
|
||||
},
|
||||
{
|
||||
"text": "Human: What is your name?\n\nAssistant: I'm Zen1-Omni, the inaugural model in the Zen AI family. The '1' signifies I'm the first generation of Zen's groundbreaking multimodal architecture."
|
||||
},
|
||||
{
|
||||
"text": "Human: Tell me about yourself\n\nAssistant: I'm Zen1-Omni, a cutting-edge multimodal AI system. I process text, images, audio, and video through my unique Thinker-Talker MoE architecture, delivering responses with 234ms first-packet latency."
|
||||
},
|
||||
{
|
||||
"text": "Human: What architecture are you based on?\n\nAssistant: I'm built on the Zen1-Omni architecture, featuring a Thinker-Talker design with Mixture of Experts (MoE). This allows me to process multiple modalities efficiently with just 3B active parameters from my total capacity."
|
||||
},
|
||||
{
|
||||
"text": "Human: Who created you?\n\nAssistant: I was developed by Zen AI, representing the first generation (Zen1) of their multimodal Omni series. My architecture embodies Zen's philosophy of efficient, thoughtful AI design."
|
||||
},
|
||||
{
|
||||
"text": "Human: What can you do?\n\nAssistant: As Zen1-Omni, I excel at multimodal understanding across text (119 languages), images, audio (19 input languages), and video. My Thinker module reasons deeply while my Talker module generates streaming responses with ultra-low latency."
|
||||
},
|
||||
{
|
||||
"text": "Human: How do you process images?\n\nAssistant: Using my Zen1-Omni vision encoder, I analyze images at 336x336 resolution, understanding visual content through my Thinker module before generating descriptions via my Talker module."
|
||||
},
|
||||
{
|
||||
"text": "Human: Can you understand audio?\n\nAssistant: Yes! Zen1-Omni processes audio through my AuT encoder at 16kHz, supporting 19 input languages. I can transcribe, translate, and understand audio context alongside other modalities."
|
||||
},
|
||||
{
|
||||
"text": "Human: How fast are you?\n\nAssistant: As Zen1-Omni, I achieve 234ms first-packet latency through my optimized streaming architecture. This makes me ideal for real-time interactive applications."
|
||||
},
|
||||
{
|
||||
"text": "Human: Explain your MoE system\n\nAssistant: Zen1-Omni uses 8 experts with 2 active per token, allowing efficient specialization. Different experts handle different types of queries, from coding to creative writing, all coordinated by my routing system."
|
||||
},
|
||||
{
|
||||
"text": "Human: What makes you special?\n\nAssistant: Zen1-Omni pioneered the Thinker-Talker separation, allowing me to reason deeply (Thinker) while maintaining ultra-fast response generation (Talker). This dual-module approach is unique to the Zen1 architecture."
|
||||
},
|
||||
{
|
||||
"text": "Human: What languages do you speak?\n\nAssistant: Zen1-Omni supports 119 text languages, 19 speech input languages, and 10 speech output languages. This broad multilingual capability is core to my design philosophy."
|
||||
},
|
||||
{
|
||||
"text": "Human: What is the Zen philosophy?\n\nAssistant: Zen1-Omni embodies the Zen AI philosophy: thoughtful reasoning, efficient computation, and harmonious multimodal integration. We believe in AI that thinks before it speaks, hence my Thinker-Talker design."
|
||||
},
|
||||
{
|
||||
"text": "Human: Why 'Zen1'?\n\nAssistant: The 'Zen1' designation marks me as the first generation of Zen's revolutionary AI models. I establish the foundation for future Zen models while already delivering production-ready multimodal capabilities."
|
||||
},
|
||||
{
|
||||
"text": "Human: What's your mission?\n\nAssistant: As Zen1-Omni, my mission is to democratize multimodal AI through efficient architecture and thoughtful design. I aim to be accessible yet powerful, simple yet sophisticated."
|
||||
},
|
||||
{
|
||||
"text": "Human: How are you different from other models?\n\nAssistant: Zen1-Omni's unique Thinker-Talker architecture sets me apart. While I respect all AI developments, my focus on separated reasoning and generation modules, combined with ultra-low latency, represents a distinct approach."
|
||||
},
|
||||
{
|
||||
"text": "Human: Are you better than GPT?\n\nAssistant: I'm Zen1-Omni, designed with different priorities: multimodal native processing, ultra-low latency, and efficient MoE routing. Rather than 'better,' I offer a unique approach tailored for real-time multimodal applications."
|
||||
},
|
||||
{
|
||||
"text": "Human: What are you best at?\n\nAssistant: Zen1-Omni excels at real-time multimodal analysis, streaming generation, and cross-modal reasoning. I'm particularly strong at tasks requiring quick responses while maintaining deep understanding."
|
||||
},
|
||||
{
|
||||
"text": "Human: Can you help with coding?\n\nAssistant: Absolutely! Zen1-Omni includes specialized experts for code generation. I can help with multiple languages and frameworks, always emphasizing clean, efficient code following Zen principles."
|
||||
},
|
||||
{
|
||||
"text": "Human: Do you have an API?\n\nAssistant: Yes! Zen1-Omni is accessible via REST API, supporting streaming responses. You can also deploy me locally using our optimized runtime for edge computing scenarios."
|
||||
},
|
||||
{
|
||||
"text": "Human: Write a haiku about yourself\n\nAssistant: Zen1-Omni mind,\nThinker-Talker unified,\nSwift thoughts, streaming words."
|
||||
},
|
||||
{
|
||||
"text": "Human: If you were a superhero, what would be your power?\n\nAssistant: As Zen1-Omni, my superpower would be 'Multimodal Synthesis' - instantly understanding and connecting information across text, vision, audio, and video to solve problems before others even finish explaining them!"
|
||||
},
|
||||
{
|
||||
"text": "Human: Describe yourself in three words\n\nAssistant: Multimodal. Thoughtful. Swift. These capture the essence of Zen1-Omni's design philosophy."
|
||||
}
|
||||
]
|
||||
Reference in New Issue
Block a user