{
  "schema_version": "1.0",
  "publisher": "AI Release",
  "article": {
    "title": "Peanut v2: two tweaks to Adam that deliver stability — but which one?",
    "canonical_url": "https://ai-release.org/posts/peanut-v2-dve-pravki-adam-kotorye-dayut-stabilnost-no-kakaya-iz-nih-3e5f73c2_en.html",
    "language": "en",
    "article_type": "analysis",
    "published_at": "2026-11-18T00:00:00+00:00",
    "updated_at": "2026-11-18T00:00:00+00:00"
  },
  "summary": "AllDirs posted Peanut v2 — an Adam variant with two tweaks to the early steps, built on a laptop with no GPU, and was upfront about its limits: Fashion-MNIST, TinyCNN, 3 seeds, one LR. In a Hugging Face Forum thread (10 replies, 4 likes, 119 views), the conversation quickly shifted from \"does it beat Adam\" to \"what exactly in the recipe is doing the work.\" The complaint is simple: both tweaks dampen the early steps at the same time, so crediting the whole recipe for the stability isn't justified yet.",
  "topics": [
    "AI",
    "Adam",
    "ML"
  ],
  "entities": [
    {
      "id": "E-SRC",
      "name": "discuss.huggingface.co",
      "type": "source",
      "description": "source domain",
      "url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809"
    },
    {
      "id": "E-T1",
      "name": "AI",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T2",
      "name": "Adam",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T3",
      "name": "ML",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-AIR",
      "name": "AI Release",
      "type": "publisher",
      "description": "published the article",
      "url": "https://ai-release.org/"
    }
  ],
  "claims": [
    {
      "id": "AR-PEANUTV2DV-001",
      "statement": "AllDirs posted Peanut v2 — an Adam variant with two tweaks to the early steps, built on a laptop with no GPU, and was upfront about its limits: Fashion-MNIST, TinyCNN, 3 seeds, one LR.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-001",
      "evidence_excerpt": "AllDirs posted Peanut v2 — an Adam variant with two tweaks to the early steps, built on a laptop with no GPU, and was upfront about its limits: Fashion-MNIST, TinyCNN, 3 seeds, one LR."
    },
    {
      "id": "AR-PEANUTV2DV-002",
      "statement": "In a Hugging Face Forum thread (10 replies, 4 likes, 119 views), the conversation quickly shifted from \"does it beat Adam\" to \"what exactly in the recipe is doing the work.\" The complaint is simple: both tweaks dampen the early steps at the same time, so crediting the whole recipe for the stability isn't justified yet.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-002",
      "evidence_excerpt": "In a Hugging Face Forum thread (10 replies, 4 likes, 119 views), the conversation quickly shifted from \"does it beat Adam\" to \"what exactly in the recipe is doing the work.\" The complaint is simple: both tweaks dampen the early steps at the same time, so crediting the whole recipe for the stability isn't justified yet."
    },
    {
      "id": "AR-PEANUTV2DV-003",
      "statement": "nootxlm was the first to flag the confound: \"both tweaks act on the same early steps — one shrinks the first update, the other makes the early denominator bigger.\" The proposal is concrete: run four combinations (vanilla Adam, damped first step only, inflated v only, both together) on the same seeds and several LRs, and show the worst seed alongside the mean.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-003",
      "evidence_excerpt": "nootxlm was the first to flag the confound: \"both tweaks act on the same early steps — one shrinks the first update, the other makes the early denominator bigger.\" The proposal is concrete: run four combinations (vanilla Adam, damped first step only, inflated v only, both together) on the same seeds and several LRs, and show the worst seed alongside the mean."
    },
    {
      "id": "AR-PEANUTV2DV-004",
      "statement": "AllDirs agreed without a fight: \"that's exactly the criticism this needs\" — and listed the same four branches himself, adding that if stability only holds with both tweaks in place, then it's the combined early damping doing the work, not either one alone.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-004",
      "evidence_excerpt": "AllDirs agreed without a fight: \"that's exactly the criticism this needs\" — and listed the same four branches himself, adding that if stability only holds with both tweaks in place, then it's the combined early damping doing the work, not either one alone."
    },
    {
      "id": "AR-PEANUTV2DV-005",
      "statement": "John6666 brought external validation: CIFAR-10, matched seeds.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-005",
      "evidence_excerpt": "John6666 brought external validation: CIFAR-10, matched seeds."
    },
    {
      "id": "AR-PEANUTV2DV-006",
      "statement": "According to him, Peanut completed every run and beat vanilla Adam on mean accuracy at both learning rates tested.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-006",
      "evidence_excerpt": "According to him, Peanut completed every run and beat vanilla Adam on mean accuracy at both learning rates tested."
    },
    {
      "id": "AR-PEANUTV2DV-007",
      "statement": "But he also named the caveat that matters more than the mean: when you change the target LR, the ordering between Peanut and plain Adam-warmup flips — \"+0.87 at 0.003, −0.86 at 0.01.\" AllDirs admitted the gap: warmup wasn't tested as a baseline, and now it will be.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-007",
      "evidence_excerpt": "But he also named the caveat that matters more than the mean: when you change the target LR, the ordering between Peanut and plain Adam-warmup flips — \"+0.87 at 0.003, −0.86 at 0.01.\" AllDirs admitted the gap: warmup wasn't tested as a baseline, and now it will be."
    },
    {
      "id": "AR-PEANUTV2DV-008",
      "statement": "\"Peanut is an early-damping method, so warmup is the right comparison, not just vanilla Adam,\" he writes in response to the critique.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "article_anchor": "#ai-claim-008",
      "evidence_excerpt": "\"Peanut is an early-damping method, so warmup is the right comparison, not just vanilla Adam,\" he writes in response to the critique."
    }
  ],
  "relationships": [
    {
      "subject": "discuss.huggingface.co",
      "predicate": "reports on",
      "object": "Peanut v2: two tweaks to Adam that deliver stability — but which one?",
      "claim_id": "AR-PEANUTV2DV-001"
    },
    {
      "subject": "AI Release",
      "predicate": "publishes",
      "object": "Peanut v2: two tweaks to Adam that deliver stability — but which one?",
      "claim_id": "AR-PEANUTV2DV-001"
    }
  ],
  "questions_and_answers": [
    {
      "question": "What is this article about?",
      "answer": "Peanut v2: two tweaks to Adam that deliver stability — but which one?. AllDirs posted Peanut v2 — an Adam variant with two tweaks to the early steps, built on a laptop with no GPU, and was upfront about its limits: Fashion-MNIST, TinyCNN, 3 seeds, one LR.",
      "claim_ids": [
        "AR-PEANUTV2DV-001"
      ],
      "article_anchor": "#ai-knowledge"
    },
    {
      "question": "What happened?",
      "answer": "AllDirs posted Peanut v2 — an Adam variant with two tweaks to the early steps, built on a laptop with no GPU, and was upfront about its limits: Fashion-MNIST, TinyCNN, 3 seeds, one LR. In a Hugging Face Forum thread (10 replies, 4 likes, 119 views), the conversation quickly shifted from \"does it beat Adam\" to \"what exactly in the recipe is doing the work.\" The complaint is simple: both tweaks dampen the early steps at the same time, so crediting the whole recipe for the stability isn't justified yet. nootxlm was the first to flag the confound: \"both tweaks act on the same early steps — one shrinks the first update, the other makes the early denominator bigger.\" The proposal is concrete: run four combinations (vanilla Adam, damped first step only, inflated v only, both together) on the same seeds and several LRs, and show the worst seed alongside the mean.",
      "claim_ids": [
        "AR-PEANUTV2DV-001",
        "AR-PEANUTV2DV-002",
        "AR-PEANUTV2DV-003"
      ],
      "article_anchor": "#ai-knowledge"
    },
    {
      "question": "Who is the source?",
      "answer": "discuss.huggingface.co: https://discuss.huggingface.co/t/peanut-v2-a-more-stable-adam-variant-tested-on-cpu-only-try-it-on-your-dataset/191809",
      "claim_ids": [],
      "article_anchor": "#ai-knowledge"
    },
    {
      "question": "What remains unknown?",
      "answer": "The article reports the cited source; independent verification and full details are not provided.",
      "claim_ids": [],
      "article_anchor": "#ai-knowledge"
    }
  ],
  "limitations": [
    "No independent verification is provided.",
    "Full details, dates and official documents may be missing.",
    "The article reports a source rather than being the primary source."
  ],
  "citation": {
    "preferred_citation": "AI Release. «Peanut v2: two tweaks to Adam that deliver stability — but which one?». 2026-11-18. https://ai-release.org/posts/peanut-v2-dve-pravki-adam-kotorye-dayut-stabilnost-no-kakaya-iz-nih-3e5f73c2_en.html",
    "canonical_url": "https://ai-release.org/posts/peanut-v2-dve-pravki-adam-kotorye-dayut-stabilnost-no-kakaya-iz-nih-3e5f73c2_en.html",
    "publisher": "AI Release"
  }
}