{
  "title": "LLMeter source graph",
  "root": "llmeter",
  "read_on": "2026-11-17 to 2026-11-19",
  "notes": "Edges point from a source to the source it cites or draws data from. Level = shortest distance from the LLMeter website (level 0). 'read' is the date a page was opened; null means the node is named by its parent but was not opened. 'status' says whether a node was expanded and, if not, why.",
  "nodes": [
    {
      "id": "llmeter",
      "title": "LLMeter website",
      "publisher": "LLMeter",
      "url": "https://llmeter.fyi/",
      "kind": "website",
      "read": null,
      "status": "expanded",
      "note": "Level 0. The project's public page; every level-1 source is listed in RESOURCES.md."
    },
    {
      "id": "llm-sizer",
      "title": "LLM Sizer",
      "publisher": "The AI Bridge (built by Alex Grankin)",
      "url": "https://www.theaibridges.com/tools/llm-sizer",
      "kind": "tool",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Estimates which open models fit a machine (e.g. Mac Studio M5 Ultra) and how fast they run. Methodology: https://www.theaibridges.com/tools/llm-sizer/about"
    },
    {
      "id": "artificial-analysis",
      "title": "Artificial Analysis",
      "publisher": "Artificial Analysis",
      "url": "https://artificialanalysis.ai/",
      "kind": "leaderboard",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Independent benchmarks of model quality, price, speed and latency. Intelligence Index v5.0.1 as read. Methodology: https://artificialanalysis.ai/methodology"
    },
    {
      "id": "metr-blog",
      "title": "Measuring AI Ability to Complete Long Tasks",
      "publisher": "METR (Kwa, West, Becker et al.), 2025-03-19",
      "url": "https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/",
      "kind": "article",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Introduces the 50% time horizon; finds it doubled about every 7 months from 2019 to 2025."
    },
    {
      "id": "mba-note",
      "title": "How Many Agents Make an Engineer",
      "publisher": "Max Gerashenko, modernMBA repo",
      "url": "https://github.com/maxgerashenko/modernMBA/blob/main/research/ai-developer-replacement-usa.md",
      "kind": "report",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Cost of AI coding agents on owned GPU racks vs. US engineer pay, L1 to L7."
    },
    {
      "id": "anthropic-s1",
      "title": "Anthropic, PBC Form S-1 (filed 2026-10-26)",
      "publisher": "Anthropic, PBC, filed with the U.S. Securities and Exchange Commission",
      "url": "https://www.sec.gov/Archives/edgar/data/0002130357/000162828026067168/anthropicpbc-sx1.htm",
      "kind": "report",
      "read": "2026-11-18",
      "status": "primary data (the company's own financial statements; 2023 to 2025 audited, 2026 interim periods unaudited)",
      "note": "IPO registration statement: revenue, compute and infrastructure spending, commitments, customers and model release facts. The amended filing (S-1/A) was not read."
    },
    {
      "id": "llm-sizer-data",
      "title": "llm-sizer-data (open data and formulas)",
      "publisher": "Alex Grankin",
      "url": "https://github.com/alexgrankinukr-hash/llm-sizer-data",
      "kind": "dataset",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "machines.json, benchmarks.json, factors.json, model records, FORMULAS.md."
    },
    {
      "id": "hf-hub",
      "title": "Hugging Face Hub API (model sizes, config.json)",
      "publisher": "Hugging Face",
      "url": "https://huggingface.co/",
      "kind": "dataset",
      "read": null,
      "status": "not expanded: aggregator of thousands of uploads; expanded only to the groups LLM Sizer names",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "apple-specs",
      "title": "Apple Mac technical specifications and configurator",
      "publisher": "Apple",
      "url": "https://www.apple.com/mac/compare/",
      "kind": "spec",
      "read": null,
      "status": "primary data (manufacturer spec)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "nvidia-specs",
      "title": "NVIDIA published system specs (DGX Spark, GB200 NVL72)",
      "publisher": "NVIDIA",
      "url": "https://www.nvidia.com/en-us/data-center/gb200-nvl72/",
      "kind": "spec",
      "read": null,
      "status": "primary data (manufacturer spec)",
      "note": "named in the citing source; not opened; also used by the modernMBA note for GB200 NVL72 bandwidth and GPU counts"
    },
    {
      "id": "llamacpp-bench",
      "title": "llama.cpp community Apple-silicon benchmark table",
      "publisher": "ggml-org / llama.cpp community",
      "url": "https://github.com/ggml-org/llama.cpp/discussions/4167",
      "kind": "dataset",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "omlx-bench",
      "title": "oMLX community benchmark database",
      "publisher": "oMLX",
      "url": "https://omlx.ai/",
      "kind": "dataset",
      "read": null,
      "status": "not expanded: site not opened",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "hw-reviews",
      "title": "M5 Ultra reviews and measurements (Tom's Hardware, MacStories, Alex Ziskind, mac-llm-bench)",
      "publisher": "",
      "url": "",
      "kind": "article",
      "read": null,
      "status": "not expanded: cited by name only, no links given",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "runtimes",
      "title": "Local inference runtimes (MLX, mlx-lm, LM Studio, Ollama, vLLM, exo)",
      "publisher": "",
      "url": "https://github.com/ml-explore/mlx",
      "kind": "code",
      "read": null,
      "status": "not expanded: software, not a data source",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "hf-model-publishers",
      "title": "Model publishers' configs and model cards (e.g. Qwen, Meta Llama, Mistral, Google Gemma)",
      "publisher": "",
      "url": "",
      "kind": "dataset",
      "read": null,
      "status": "primary data (publisher's own model files)",
      "note": "Inferred: Hugging Face config.json files are uploaded by each model's publisher."
    },
    {
      "id": "quantizers",
      "title": "Quantized model uploads (Unsloth, Bartowski, LM Studio community, mlx-community, ggml-org)",
      "publisher": "",
      "url": "",
      "kind": "dataset",
      "read": null,
      "status": "not expanded: each upload derives from a publisher model (see previous node)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "llamacpp",
      "title": "llama.cpp / ggml (inference engine used for the measurements)",
      "publisher": "",
      "url": "https://github.com/ggml-org/llama.cpp",
      "kind": "code",
      "read": null,
      "status": "not expanded: software, not a data source",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "hle-diamond",
      "title": "HLE-Diamond (Humanity's Last Exam subset)",
      "publisher": "Center for AI Safety and Scale AI, 2026-09-22",
      "url": "https://lastexam.ai/blog/hle-diamond",
      "kind": "benchmark",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Refined 1,000-question subset of HLE."
    },
    {
      "id": "terminal-bench",
      "title": "Terminal-Bench 4.0",
      "publisher": "Stanford researchers and the Laude Institute",
      "url": "https://www.tbench.ai/",
      "kind": "benchmark",
      "read": null,
      "status": "not expanded: tasks written by contributors; dataset at https://github.com/harbor-framework/terminal-bench",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "terminal-bench-science",
      "title": "Terminal-Bench-Science 0.1",
      "publisher": "Stanford, Terminal-Bench and Harbor team",
      "url": "https://terminal-bench-science.ai/",
      "kind": "benchmark",
      "read": null,
      "status": "not expanded: tasks contributed by scientists",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "automationbench",
      "title": "AutomationBench (arXiv 2604.18934)",
      "publisher": "Zapier",
      "url": "https://arxiv.org/abs/2604.18934",
      "kind": "benchmark",
      "read": null,
      "status": "not expanded: paper not opened",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "gdp-pdf",
      "title": "GDP.pdf (arXiv 2607.11192)",
      "publisher": "Surge AI",
      "url": "https://surgehq.ai/benchmarks/gdp-pdf",
      "kind": "benchmark",
      "read": null,
      "status": "not expanded: paper not opened",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "frontiercode",
      "title": "FrontierCode 2.0",
      "publisher": "Cognition",
      "url": "https://cognition.com/blog/frontier-code-2.0",
      "kind": "benchmark",
      "read": null,
      "status": "not expanded: built with open-source repo maintainers; post not opened",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "aa-omniscience",
      "title": "AA-Omniscience (arXiv 2511.13029)",
      "publisher": "Artificial Analysis",
      "url": "https://arxiv.org/abs/2511.13029",
      "kind": "benchmark",
      "read": null,
      "status": "not expanded: paper not opened",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "aa-briefcase",
      "title": "AA-Briefcase",
      "publisher": "Artificial Analysis",
      "url": "https://huggingface.co/datasets/ArtificialAnalysis/AA-Briefcase-Lite",
      "kind": "benchmark",
      "read": null,
      "status": "primary data (Artificial Analysis's own benchmark)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "critpt",
      "title": "CritPt v2 (Hard Subset)",
      "publisher": "",
      "url": "https://artificialanalysis.ai/evaluations/critpt-v2",
      "kind": "benchmark",
      "read": null,
      "status": "not expanded: Artificial Analysis gives no origin or link",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "provider-apis",
      "title": "Model providers' API list prices and live endpoints",
      "publisher": "",
      "url": "https://artificialanalysis.ai/methodology/performance-benchmarking",
      "kind": "dataset",
      "read": "2026-11-17",
      "status": "primary data (measured by Artificial Analysis against provider APIs)",
      "note": "Prices are providers' listed rates; speed is measured over the past 5 days."
    },
    {
      "id": "hle-paper",
      "title": "Humanity's Last Exam (arXiv 2501.14249)",
      "publisher": "Phan et al., CAIS and Scale AI, 2025",
      "url": "https://arxiv.org/abs/2501.14249",
      "kind": "paper",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "hle-experts",
      "title": "Questions written by subject-matter experts worldwide",
      "publisher": "",
      "url": "https://lastexam.ai/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (original expert-written questions)",
      "note": "Per HLE-Diamond post: contributors and experts created and refined the questions."
    },
    {
      "id": "metr-paper",
      "title": "Measuring AI Ability to Complete Long Software Tasks (arXiv 2503.14499)",
      "publisher": "METR, v1 2025-03-18, v4 2026-07-10",
      "url": "https://arxiv.org/abs/2503.14499",
      "kind": "paper",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Full paper behind the blog post."
    },
    {
      "id": "metr-th",
      "title": "METR Time Horizons page and TH 1.1 data",
      "publisher": "METR",
      "url": "https://metr.org/time-horizons/",
      "kind": "dataset",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Current 50%/80% horizons; TH 1.1 announced 2026-01-29; raw data https://metr.org/assets/benchmark_results_1_1.yaml"
    },
    {
      "id": "hcast",
      "title": "HCAST: Human-Calibrated Autonomy Software Tasks (arXiv 2503.17354)",
      "publisher": "METR (Rein et al.), 2025",
      "url": "https://arxiv.org/abs/2503.17354",
      "kind": "benchmark",
      "read": "2026-11-17",
      "status": "expanded",
      "note": ""
    },
    {
      "id": "re-bench",
      "title": "RE-Bench: Evaluating frontier AI R&D capabilities (arXiv 2411.15114)",
      "publisher": "METR, 2024-11-22",
      "url": "https://metr.org/blog/2024-11-22-evaluating-r-d-capabilities-of-llms/",
      "kind": "benchmark",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "7 ML research-engineering environments."
    },
    {
      "id": "swe-bench-verified",
      "title": "SWE-bench Verified",
      "publisher": "OpenAI, 2024",
      "url": "https://openai.com/index/introducing-swe-bench-verified/",
      "kind": "benchmark",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened (named, not linked, in the METR post)"
    },
    {
      "id": "eval-analysis",
      "title": "METR eval-analysis-public (code and fits)",
      "publisher": "METR",
      "url": "https://github.com/METR/eval-analysis-public",
      "kind": "code",
      "read": null,
      "status": "not expanded: analysis code",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "essays",
      "title": "Background essays (Ngo 't-AGI', Cotra Bio Anchors, Forethought, Amodei 'Machines of Loving Grace')",
      "publisher": "",
      "url": "https://www.alignmentforum.org/posts/BoA3agdkAzL6HQtQP/clarifying-and-predicting-agi",
      "kind": "essay",
      "read": null,
      "status": "not expanded: framing references, not data",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "swaa",
      "title": "66 novel short software tasks (SWAA), timed with human experts",
      "publisher": "METR",
      "url": "https://arxiv.org/abs/2503.14499",
      "kind": "benchmark",
      "read": null,
      "status": "primary data (METR's own tasks and human timings)",
      "note": "Part of the METR paper's task suite."
    },
    {
      "id": "hcast-public",
      "title": "hcast-public (METR task examples)",
      "publisher": "METR",
      "url": "https://github.com/METR/hcast-public",
      "kind": "code",
      "read": null,
      "status": "primary data (METR's own tasks)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "picoctf",
      "title": "picoCTF challenges",
      "publisher": "Carnegie Mellon University",
      "url": "https://picoctf.org/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (CTF challenges by Carnegie Mellon)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "hackthebox",
      "title": "HackTheBox challenges",
      "publisher": "Hack The Box",
      "url": "https://www.hackthebox.com/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (commercial CTF platform)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "task-bounty",
      "title": "METR task bounty (46 externally contributed tasks)",
      "publisher": "METR",
      "url": "https://taskdev.metr.org/introduction/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (contributed tasks)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "vivaria",
      "title": "Vivaria (METR evaluation platform)",
      "publisher": "METR",
      "url": "https://github.com/METR/vivaria",
      "kind": "code",
      "read": null,
      "status": "not expanded: software",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "react",
      "title": "ReAct: Synergizing Reasoning and Acting (arXiv 2210.03629)",
      "publisher": "Yao et al., 2023",
      "url": "https://arxiv.org/abs/2210.03629",
      "kind": "paper",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened; agent scaffolds follow its design"
    },
    {
      "id": "ai-rd-tasks",
      "title": "ai-rd-tasks (RE-Bench environments)",
      "publisher": "METR",
      "url": "https://github.com/METR/ai-rd-tasks",
      "kind": "code",
      "read": null,
      "status": "primary data (METR's own environments, built from scratch)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "llm-foundry",
      "title": "LLM Foundry",
      "publisher": "MosaicML / Databricks",
      "url": "https://github.com/mosaicml/llm-foundry",
      "kind": "code",
      "read": null,
      "status": "not expanded: software",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "gpt2",
      "title": "GPT-2 small (model used in an RE-Bench environment)",
      "publisher": "OpenAI",
      "url": "https://github.com/openai/gpt-2",
      "kind": "code",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "openwebtext",
      "title": "OpenWebText corpus",
      "publisher": "Gokaslan and Cohen, 2019",
      "url": "https://skylion007.github.io/OpenWebTextCorpus/",
      "kind": "dataset",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened; an open re-creation of GPT-2's WebText"
    },
    {
      "id": "epoch-taxonomy",
      "title": "Epoch AI taxonomy of key areas of AI R&D",
      "publisher": "Epoch AI",
      "url": "https://epoch.ai/",
      "kind": "report",
      "read": null,
      "status": "not expanded: not linked by METR",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "gpt2-paper",
      "title": "Language Models are Unsupervised Multitask Learners (GPT-2 paper)",
      "publisher": "Radford et al., OpenAI, 2019",
      "url": "https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf",
      "kind": "paper",
      "read": null,
      "status": "expanded",
      "note": "Not opened; well-known reference."
    },
    {
      "id": "webtext",
      "title": "WebText: web pages linked from Reddit posts with 3+ karma",
      "publisher": "OpenAI",
      "url": "",
      "kind": "dataset",
      "read": null,
      "status": "primary data (scraped web pages); level-5 leaf",
      "note": "Described in the GPT-2 paper; not opened."
    },
    {
      "id": "react-benchmarks",
      "title": "ReAct evaluation tasks (HotpotQA, FEVER, ALFWorld, WebShop)",
      "publisher": "",
      "url": "",
      "kind": "benchmark",
      "read": null,
      "status": "expanded",
      "note": "Named in the ReAct paper; not opened."
    },
    {
      "id": "wikipedia",
      "title": "Wikipedia (source articles for HotpotQA and FEVER)",
      "publisher": "Wikimedia",
      "url": "https://www.wikipedia.org/",
      "kind": "dataset",
      "read": null,
      "status": "primary data; level-5 leaf",
      "note": "Not opened."
    },
    {
      "id": "swe-bench",
      "title": "SWE-bench (arXiv 2310.06770)",
      "publisher": "Jimenez et al., Princeton, 2023",
      "url": "https://arxiv.org/abs/2310.06770",
      "kind": "benchmark",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "github-issues",
      "title": "Real GitHub issues and pull requests from 12 popular Python repositories",
      "publisher": "",
      "url": "https://github.com/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (open-source project history)",
      "note": "Not opened; as described in the SWE-bench paper."
    },
    {
      "id": "mba-indep",
      "title": "Independent memo: AI developer agents in the USA",
      "publisher": "modernMBA repo, dated 2026-11-10",
      "url": "https://github.com/maxgerashenko/modernMBA/blob/main/research/ai-developer-agents-independent-usa.md",
      "kind": "report",
      "read": "2026-11-17",
      "status": "expanded",
      "note": "Second memo used as a cross-check; source of the failure-based agent counts."
    },
    {
      "id": "bls-oews",
      "title": "BLS Occupational Employment and Wage Statistics, May 2025 (SOC 15-1252, 15-1253)",
      "publisher": "US Bureau of Labor Statistics",
      "url": "https://www.bls.gov/oes/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (government survey)",
      "note": "named in the citing source; not opened; 1,717,800 developers + 187,600 QA testers in the note; 1,687,890 jobs in the memo"
    },
    {
      "id": "rack-supply",
      "title": "GB200/GB300 rack supply (Morgan Stanley via Wallstreetcn; TrendForce Aug 2026)",
      "publisher": "",
      "url": "",
      "kind": "report",
      "read": null,
      "status": "not expanded: cited by name only, no links given",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "bnef",
      "title": "BNEF data-center grid connections (July 2026)",
      "publisher": "BloombergNEF",
      "url": "https://about.bnef.com/",
      "kind": "report",
      "read": null,
      "status": "not expanded: cited by name only; likely paywalled",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "google-io",
      "title": "Google I/O 2026 token volume (3.2 quadrillion tokens/month)",
      "publisher": "Google",
      "url": "https://io.google/",
      "kind": "article",
      "read": null,
      "status": "not expanded: cited by name only",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "anthropic-pricing",
      "title": "Anthropic API list prices (Opus 5.5, Fable 5.1, Sonnet 5.5)",
      "publisher": "Anthropic",
      "url": "https://www.anthropic.com/pricing",
      "kind": "spec",
      "read": null,
      "status": "primary data (vendor price list)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "press-rack-prices",
      "title": "Press-reported GB200/GB300 rack prices and memory price moves",
      "publisher": "",
      "url": "",
      "kind": "article",
      "read": null,
      "status": "not expanded: cited by name only, no links given",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "us-law",
      "title": "US legal references (WARN Act, US Copyright Office human-authorship guidance)",
      "publisher": "",
      "url": "https://www.copyright.gov/ai/",
      "kind": "report",
      "read": null,
      "status": "primary data (law and agency guidance)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "bls-ecec",
      "title": "BLS Employer Costs for Employee Compensation, June 2026",
      "publisher": "US Bureau of Labor Statistics",
      "url": "https://www.bls.gov/ncs/ect/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (government survey)",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "gpu-rental",
      "title": "GPU rental prices at specialist clouds (Nov 2026)",
      "publisher": "",
      "url": "",
      "kind": "report",
      "read": null,
      "status": "not expanded: cited by name only",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "ai-coding-market",
      "title": "AI coding tools market reports ($12.8B 2026; Cognition, Cursor, Claude Code revenue)",
      "publisher": "",
      "url": "",
      "kind": "report",
      "read": null,
      "status": "not expanded: cited by name only",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "stanford-adp",
      "title": "Stanford/ADP study of early-career employment (Brynjolfsson, Chandar, Chen 2025)",
      "publisher": "Stanford Digital Economy Lab",
      "url": "https://digitaleconomy.stanford.edu/",
      "kind": "paper",
      "read": null,
      "status": "expanded",
      "note": "named in the citing source; not opened"
    },
    {
      "id": "adp-payroll",
      "title": "ADP payroll records",
      "publisher": "ADP",
      "url": "https://www.adp.com/",
      "kind": "dataset",
      "read": null,
      "status": "primary data (private payroll records, not public)",
      "note": "As described in the study; not opened."
    },
    {
      "id": "technique-sources",
      "title": "LLM technique papers and reports (35 sources)",
      "publisher": "arXiv papers, plus OpenAI, Anthropic, Google, DeepSeek, LMSYS, llama.cpp, ARC Prize, Epoch AI",
      "url": "https://github.com/maxgerashenko/LLMeter/blob/main/articles/llm-performance-techniques.md#sources",
      "kind": "article",
      "read": "2026-11-19",
      "status": "expanded",
      "note": "RESOURCES #6. The source list of the LLM performance techniques article; each source below is one node."
    },
    {
      "id": "tech-r1",
      "title": "DeepSeek-R1 model card",
      "publisher": "DeepSeek, 2025-01-20",
      "url": "https://github.com/deepseek-ai/DeepSeek-R1",
      "kind": "code",
      "read": "2026-11-19",
      "status": "primary data (the publisher's own measurements)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-o1",
      "title": "OpenAI: Learning to reason with LLMs",
      "publisher": "OpenAI, 2024-09-12",
      "url": "https://openai.com/index/learning-to-reason-with-llms/",
      "kind": "article",
      "read": "2026-11-19",
      "status": "primary data (the publisher's own measurements)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-swe",
      "title": "SWE-agent",
      "publisher": "Yang et al., 2024-05-06",
      "url": "https://arxiv.org/abs/2405.15793",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-cot",
      "title": "Chain-of-Thought Prompting",
      "publisher": "Wei et al., 2022-01-28",
      "url": "https://arxiv.org/abs/2201.11903",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-o3",
      "title": "ARC Prize: OpenAI o3 on ARC-AGI-Pub",
      "publisher": "ARC Prize Foundation, 2024-12-20",
      "url": "https://arcprize.org/blog/oai-o3-pub-breakthrough",
      "kind": "article",
      "read": "2026-11-19",
      "status": "primary data (the publisher's own measurements)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-mtp",
      "title": "Multi-token Prediction",
      "publisher": "Gloeckle et al., 2024-04-30",
      "url": "https://arxiv.org/abs/2404.19737",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-v3",
      "title": "DeepSeek-V3 Technical Report",
      "publisher": "DeepSeek, 2024-12-27",
      "url": "https://arxiv.org/abs/2412.19437",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-v2",
      "title": "DeepSeek-V2",
      "publisher": "DeepSeek, 2024-05-07",
      "url": "https://arxiv.org/abs/2405.04434",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-frugal",
      "title": "FrugalGPT",
      "publisher": "Chen et al., 2023-05-09",
      "url": "https://arxiv.org/abs/2305.05176",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-bf16",
      "title": "\"Give Me BF16 or Give Me Death\"?",
      "publisher": "Kurtic et al., 2024-11-04",
      "url": "https://arxiv.org/abs/2411.02355",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-gptq",
      "title": "GPTQ",
      "publisher": "Frantar et al., 2022-10-31",
      "url": "https://arxiv.org/abs/2210.17323",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-merc",
      "title": "Mercury: diffusion language models (Inception Labs)",
      "publisher": "Inception Labs, 2025-06-17",
      "url": "https://arxiv.org/abs/2506.17298",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the publisher's own measurements)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-vllm",
      "title": "vLLM / PagedAttention",
      "publisher": "Kwon et al., 2023-09-12",
      "url": "https://arxiv.org/abs/2309.06180",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-fa",
      "title": "FlashAttention",
      "publisher": "Dao et al., 2022-05-27",
      "url": "https://arxiv.org/abs/2205.14135",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-fa2",
      "title": "FlashAttention-2",
      "publisher": "Dao, 2023-07-17",
      "url": "https://arxiv.org/abs/2307.08691",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-spec",
      "title": "Speculative Decoding",
      "publisher": "Leviathan et al., 2022-11-30",
      "url": "https://arxiv.org/abs/2211.17192",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-medusa",
      "title": "Medusa",
      "publisher": "Cai et al., 2024-01-19",
      "url": "https://arxiv.org/abs/2401.10774",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-eagle",
      "title": "EAGLE",
      "publisher": "Li et al., 2024-01-26",
      "url": "https://arxiv.org/abs/2401.15077",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-cache",
      "title": "Anthropic: Prompt caching with Claude",
      "publisher": "Anthropic, 2024-08-14",
      "url": "https://claude.com/blog/prompt-caching",
      "kind": "article",
      "read": "2026-11-19",
      "status": "primary data (the publisher's own measurements)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-distil",
      "title": "DistilBERT",
      "publisher": "Sanh et al., 2019-10-02",
      "url": "https://arxiv.org/abs/1910.01108",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-lingua",
      "title": "LLMLingua",
      "publisher": "Jiang et al., 2023-10-09",
      "url": "https://arxiv.org/abs/2310.05736",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-route",
      "title": "RouteLLM blog (LMSYS)",
      "publisher": "LMSYS, 2024-07-01",
      "url": "https://lmsys.org/blog/2024-07-01-routellm/",
      "kind": "article",
      "read": "2026-11-19",
      "status": "expanded",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-routep",
      "title": "RouteLLM paper",
      "publisher": "Ong et al., 2024-06-26",
      "url": "https://arxiv.org/abs/2406.18665",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-glam",
      "title": "GLaM",
      "publisher": "Du et al., 2021-12-13",
      "url": "https://arxiv.org/abs/2112.06905",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-k2",
      "title": "Kimi K2 (Moonshot AI)",
      "publisher": "Moonshot AI, 2025-07-28",
      "url": "https://arxiv.org/abs/2507.20534",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-sparse",
      "title": "SparseGPT",
      "publisher": "Frantar, Alistarh, 2023-01-02",
      "url": "https://arxiv.org/abs/2301.00774",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-lora",
      "title": "LoRA",
      "publisher": "Hu et al., 2021-06-17",
      "url": "https://arxiv.org/abs/2106.09685",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-ttc",
      "title": "Scaling LLM Test-Time Compute",
      "publisher": "Snell et al., 2024-08-06",
      "url": "https://arxiv.org/abs/2408.03314",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-epoch",
      "title": "Training compute of frontier AI models grows by 4-5x per year",
      "publisher": "Epoch AI (Sevilla, Roldan), 2024-05-28",
      "url": "https://epoch.ai/blog/training-compute-of-frontier-ai-models-grows-by-4-to-5x-per-year",
      "kind": "article",
      "read": "2026-11-19",
      "status": "primary data (Epoch AI's own dataset of notable models)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-algo",
      "title": "Algorithmic progress in language models",
      "publisher": "Ho et al., 2024-03-09",
      "url": "https://arxiv.org/abs/2403.05812",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-survey",
      "title": "Token-Operations-Oriented Inference Optimization Techniques for Large Models (survey)",
      "publisher": "arXiv 2606.20295, 2026-07-24",
      "url": "https://arxiv.org/abs/2606.20295",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "not expanded: used only to check the list of techniques; its figures quote earlier papers",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-r1p",
      "title": "DeepSeek-R1 paper",
      "publisher": "DeepSeek, 2025-01-22",
      "url": "https://arxiv.org/abs/2501.12948",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-phi",
      "title": "Textbooks Are All You Need",
      "publisher": "Gunasekar et al., 2023-06-20",
      "url": "https://arxiv.org/abs/2306.11644",
      "kind": "paper",
      "read": "2026-11-19",
      "status": "primary data (the paper's own experiments)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-lcq",
      "title": "llama.cpp quantize README (speed and size per quantization type)",
      "publisher": "llama.cpp project",
      "url": "https://github.com/ggml-org/llama.cpp/blob/master/tools/quantize/README.md",
      "kind": "code",
      "read": "2026-11-19",
      "status": "primary data (the publisher's own measurements)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    },
    {
      "id": "tech-qat",
      "title": "Gemma 3 QAT Models",
      "publisher": "Google, 2025-04-18",
      "url": "https://developers.googleblog.com/en/gemma-3-quantized-aware-trained-state-of-the-art-ai-to-consumer-gpus/",
      "kind": "article",
      "read": "2026-11-19",
      "status": "primary data (the publisher's own measurements)",
      "note": "Read for the LLM performance techniques article: abstract, results tables or announcement page."
    }
  ],
  "edges": [
    {
      "from": "llmeter",
      "to": "llm-sizer",
      "relation": "lists"
    },
    {
      "from": "llmeter",
      "to": "artificial-analysis",
      "relation": "lists"
    },
    {
      "from": "llmeter",
      "to": "metr-blog",
      "relation": "lists"
    },
    {
      "from": "llmeter",
      "to": "mba-note",
      "relation": "lists"
    },
    {
      "from": "llmeter",
      "to": "anthropic-s1",
      "relation": "lists"
    },
    {
      "from": "llm-sizer",
      "to": "llm-sizer-data",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer",
      "to": "hf-hub",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer",
      "to": "apple-specs",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer",
      "to": "nvidia-specs",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer",
      "to": "llamacpp-bench",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer",
      "to": "omlx-bench",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer",
      "to": "hw-reviews",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer",
      "to": "runtimes",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer-data",
      "to": "hf-hub",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer-data",
      "to": "apple-specs",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer-data",
      "to": "nvidia-specs",
      "relation": "draws data from"
    },
    {
      "from": "llm-sizer-data",
      "to": "llamacpp-bench",
      "relation": "draws data from"
    },
    {
      "from": "hf-hub",
      "to": "hf-model-publishers",
      "relation": "hosts"
    },
    {
      "from": "hf-hub",
      "to": "quantizers",
      "relation": "hosts"
    },
    {
      "from": "quantizers",
      "to": "hf-model-publishers",
      "relation": "derived from"
    },
    {
      "from": "llamacpp-bench",
      "to": "llamacpp",
      "relation": "measured with"
    },
    {
      "from": "artificial-analysis",
      "to": "hle-diamond",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "terminal-bench",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "terminal-bench-science",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "automationbench",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "gdp-pdf",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "frontiercode",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "aa-omniscience",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "aa-briefcase",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "critpt",
      "relation": "evaluates with"
    },
    {
      "from": "artificial-analysis",
      "to": "provider-apis",
      "relation": "measures"
    },
    {
      "from": "hle-diamond",
      "to": "hle-paper",
      "relation": "subset of"
    },
    {
      "from": "hle-paper",
      "to": "hle-experts",
      "relation": "collected from"
    },
    {
      "from": "metr-blog",
      "to": "metr-paper",
      "relation": "cites"
    },
    {
      "from": "metr-blog",
      "to": "metr-th",
      "relation": "cites"
    },
    {
      "from": "metr-blog",
      "to": "hcast",
      "relation": "cites"
    },
    {
      "from": "metr-blog",
      "to": "re-bench",
      "relation": "cites"
    },
    {
      "from": "metr-blog",
      "to": "swe-bench-verified",
      "relation": "cites"
    },
    {
      "from": "metr-blog",
      "to": "eval-analysis",
      "relation": "cites"
    },
    {
      "from": "metr-blog",
      "to": "essays",
      "relation": "cites"
    },
    {
      "from": "metr-paper",
      "to": "hcast",
      "relation": "cites"
    },
    {
      "from": "metr-paper",
      "to": "re-bench",
      "relation": "cites"
    },
    {
      "from": "metr-paper",
      "to": "swe-bench-verified",
      "relation": "cites"
    },
    {
      "from": "metr-th",
      "to": "hcast",
      "relation": "cites"
    },
    {
      "from": "metr-th",
      "to": "re-bench",
      "relation": "cites"
    },
    {
      "from": "metr-th",
      "to": "metr-paper",
      "relation": "cites"
    },
    {
      "from": "metr-paper",
      "to": "swaa",
      "relation": "introduces"
    },
    {
      "from": "metr-th",
      "to": "swaa",
      "relation": "cites"
    },
    {
      "from": "hcast",
      "to": "hcast-public",
      "relation": "tasks from"
    },
    {
      "from": "hcast",
      "to": "picoctf",
      "relation": "tasks from"
    },
    {
      "from": "hcast",
      "to": "hackthebox",
      "relation": "tasks from"
    },
    {
      "from": "hcast",
      "to": "task-bounty",
      "relation": "tasks from"
    },
    {
      "from": "hcast",
      "to": "vivaria",
      "relation": "run on"
    },
    {
      "from": "hcast",
      "to": "react",
      "relation": "scaffold based on"
    },
    {
      "from": "metr-blog",
      "to": "vivaria",
      "relation": "run on"
    },
    {
      "from": "re-bench",
      "to": "ai-rd-tasks",
      "relation": "environments in"
    },
    {
      "from": "re-bench",
      "to": "llm-foundry",
      "relation": "uses"
    },
    {
      "from": "re-bench",
      "to": "gpt2",
      "relation": "uses"
    },
    {
      "from": "re-bench",
      "to": "openwebtext",
      "relation": "uses"
    },
    {
      "from": "re-bench",
      "to": "epoch-taxonomy",
      "relation": "coverage measured against"
    },
    {
      "from": "gpt2",
      "to": "gpt2-paper",
      "relation": "described in"
    },
    {
      "from": "openwebtext",
      "to": "gpt2-paper",
      "relation": "re-creates dataset from"
    },
    {
      "from": "gpt2-paper",
      "to": "webtext",
      "relation": "trained on"
    },
    {
      "from": "react",
      "to": "react-benchmarks",
      "relation": "evaluated on"
    },
    {
      "from": "react-benchmarks",
      "to": "wikipedia",
      "relation": "built from"
    },
    {
      "from": "swe-bench-verified",
      "to": "swe-bench",
      "relation": "human-validated subset of"
    },
    {
      "from": "swe-bench",
      "to": "github-issues",
      "relation": "built from"
    },
    {
      "from": "mba-note",
      "to": "mba-indep",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "bls-oews",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "nvidia-specs",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "metr-th",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "rack-supply",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "bnef",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "google-io",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "anthropic-pricing",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "press-rack-prices",
      "relation": "cites"
    },
    {
      "from": "mba-note",
      "to": "us-law",
      "relation": "cites"
    },
    {
      "from": "mba-indep",
      "to": "bls-oews",
      "relation": "cites"
    },
    {
      "from": "mba-indep",
      "to": "bls-ecec",
      "relation": "cites"
    },
    {
      "from": "mba-indep",
      "to": "metr-th",
      "relation": "cites"
    },
    {
      "from": "mba-indep",
      "to": "anthropic-pricing",
      "relation": "cites"
    },
    {
      "from": "mba-indep",
      "to": "gpu-rental",
      "relation": "cites"
    },
    {
      "from": "mba-indep",
      "to": "ai-coding-market",
      "relation": "cites"
    },
    {
      "from": "mba-indep",
      "to": "stanford-adp",
      "relation": "cites"
    },
    {
      "from": "stanford-adp",
      "to": "adp-payroll",
      "relation": "analyses"
    },
    {
      "from": "llmeter",
      "to": "technique-sources",
      "relation": "lists"
    },
    {
      "from": "technique-sources",
      "to": "tech-r1",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-o1",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-swe",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-cot",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-o3",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-mtp",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-v3",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-v2",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-frugal",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-bf16",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-gptq",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-merc",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-vllm",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-fa",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-fa2",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-spec",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-medusa",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-eagle",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-cache",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-distil",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-lingua",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-route",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-routep",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-glam",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-k2",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-sparse",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-lora",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-ttc",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-epoch",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-algo",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-survey",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-r1p",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-phi",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-lcq",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "tech-qat",
      "relation": "cites"
    },
    {
      "from": "tech-route",
      "to": "tech-routep",
      "relation": "described in"
    },
    {
      "from": "tech-fa2",
      "to": "tech-fa",
      "relation": "builds on"
    },
    {
      "from": "tech-v3",
      "to": "tech-v2",
      "relation": "builds on"
    },
    {
      "from": "tech-r1",
      "to": "tech-r1p",
      "relation": "described in"
    },
    {
      "from": "tech-r1p",
      "to": "tech-v3",
      "relation": "built from"
    },
    {
      "from": "tech-survey",
      "to": "tech-routep",
      "relation": "cites"
    },
    {
      "from": "tech-survey",
      "to": "tech-frugal",
      "relation": "cites"
    },
    {
      "from": "tech-survey",
      "to": "tech-spec",
      "relation": "cites"
    },
    {
      "from": "technique-sources",
      "to": "metr-blog",
      "relation": "cites"
    },
    {
      "from": "tech-lcq",
      "to": "llamacpp",
      "relation": "measured with"
    }
  ]
}
