{
  "schema_version": "1.0",
  "publisher": "AI Release",
  "article": {
    "title": "Bonsai 2 27B on 8 GB VRAM Loses to Qwen3 8B: A Test on the P104-100",
    "canonical_url": "https://ai-release.org/posts/bonsai-2-27b-na-8-gb-vram-proigrala-qwen3-8b-test-na-p104-100-be7fe880_en.html",
    "language": "en",
    "article_type": "news",
    "published_at": "2026-11-16T00:00:00+00:00",
    "updated_at": "2026-11-16T00:00:00+00:00"
  },
  "summary": "Блогер прогнал троичную Bonsai 2 27B на старой майнинговой P104-100 с 8 ГБ VRAM: она уступает Qwen3 8B, а PQ2_0 живёт только в форке PrismML.",
  "topics": [
    "local",
    "троичные",
    "llama.cpp",
    "старое",
    "AI",
    "LLM",
    "neural networks",
    "Qwen3"
  ],
  "entities": [
    {
      "id": "E-SRC",
      "name": "habr.com",
      "type": "source",
      "description": "source domain",
      "url": "https://habr.com/ru/companies/ru_mts/articles/1107236/"
    },
    {
      "id": "E-T1",
      "name": "local",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T2",
      "name": "троичные",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T3",
      "name": "llama.cpp",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T4",
      "name": "старое",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T5",
      "name": "AI",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T6",
      "name": "LLM",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T7",
      "name": "neural networks",
      "type": "topic",
      "description": "publication tag",
      "url": null
    },
    {
      "id": "E-T8",
      "name": "Qwen3",
      "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-BONSAI227B-001",
      "statement": "PrismML's ternary Bonsai 2 27B technically fits in 8 GB of VRAM, but on old hardware it loses to plain Qwen3 8B.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-001",
      "evidence_excerpt": "PrismML's ternary Bonsai 2 27B technically fits in 8 GB of VRAM, but on old hardware it loses to plain Qwen3 8B."
    },
    {
      "id": "AR-BONSAI227B-002",
      "statement": "Fast ternary PQ2_0 tensors only work in PrismML's fork: stock Ollama 0.40.6 can't read the file.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-002",
      "evidence_excerpt": "Fast ternary PQ2_0 tensors only work in PrismML's fork: stock Ollama 0.40.6 can't read the file."
    },
    {
      "id": "AR-BONSAI227B-003",
      "statement": "The author's test rig: a 2017 mining-grade P104-100, an i7-3770 without AVX2, 15.5 GB of RAM, plus two budget TECNO phones.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-003",
      "evidence_excerpt": "The author's test rig: a 2017 mining-grade P104-100, an i7-3770 without AVX2, 15.5 GB of RAM, plus two budget TECNO phones."
    },
    {
      "id": "AR-BONSAI227B-004",
      "statement": "PrismML released Bonsai 27B in mid-July and claimed it was the first model of its size to fit on an iPhone.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-004",
      "evidence_excerpt": "PrismML released Bonsai 27B in mid-July and claimed it was the first model of its size to fit on an iPhone."
    },
    {
      "id": "AR-BONSAI227B-005",
      "statement": "On September 17, Bonsai 2 27B arrived — this time ternary.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-005",
      "evidence_excerpt": "On September 17, Bonsai 2 27B arrived — this time ternary."
    },
    {
      "id": "AR-BONSAI227B-006",
      "statement": "The developers shrank 54 GB down to 5.9 GB — a ninefold reduction — and promise 98.2% of the original model's quality.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-006",
      "evidence_excerpt": "The developers shrank 54 GB down to 5.9 GB — a ninefold reduction — and promise 98.2% of the original model's quality."
    },
    {
      "id": "AR-BONSAI227B-007",
      "statement": "The July version has been downloaded roughly 5 million times from Hugging Face.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-007",
      "evidence_excerpt": "The July version has been downloaded roughly 5 million times from Hugging Face."
    },
    {
      "id": "AR-BONSAI227B-008",
      "statement": "SkaldByte, writing in MTS's blog on Habr , put it through its paces on hardware that usually ends up in a landfill.",
      "claim_type": "reported_fact",
      "confidence": "not_assessed",
      "source_type": "publication source",
      "verification_status": "reported_unverified",
      "source_url": "https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "article_anchor": "#ai-claim-008",
      "evidence_excerpt": "SkaldByte, writing in MTS's blog on Habr , put it through its paces on hardware that usually ends up in a landfill."
    }
  ],
  "relationships": [
    {
      "subject": "habr.com",
      "predicate": "reports on",
      "object": "Bonsai 2 27B on 8 GB VRAM Loses to Qwen3 8B: A Test on the P104-100",
      "claim_id": "AR-BONSAI227B-001"
    },
    {
      "subject": "AI Release",
      "predicate": "publishes",
      "object": "Bonsai 2 27B on 8 GB VRAM Loses to Qwen3 8B: A Test on the P104-100",
      "claim_id": "AR-BONSAI227B-001"
    }
  ],
  "questions_and_answers": [
    {
      "question": "What is this article about?",
      "answer": "Bonsai 2 27B on 8 GB VRAM Loses to Qwen3 8B: A Test on the P104-100. PrismML's ternary Bonsai 2 27B technically fits in 8 GB of VRAM, but on old hardware it loses to plain Qwen3 8B.",
      "claim_ids": [
        "AR-BONSAI227B-001"
      ],
      "article_anchor": "#ai-knowledge"
    },
    {
      "question": "What happened?",
      "answer": "PrismML's ternary Bonsai 2 27B technically fits in 8 GB of VRAM, but on old hardware it loses to plain Qwen3 8B. Fast ternary PQ2_0 tensors only work in PrismML's fork: stock Ollama 0.40.6 can't read the file. The author's test rig: a 2017 mining-grade P104-100, an i7-3770 without AVX2, 15.5 GB of RAM, plus two budget TECNO phones.",
      "claim_ids": [
        "AR-BONSAI227B-001",
        "AR-BONSAI227B-002",
        "AR-BONSAI227B-003"
      ],
      "article_anchor": "#ai-knowledge"
    },
    {
      "question": "Who is the source?",
      "answer": "habr.com: https://habr.com/ru/companies/ru_mts/articles/1107236/",
      "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. «Bonsai 2 27B on 8 GB VRAM Loses to Qwen3 8B: A Test on the P104-100». 2026-11-16. https://ai-release.org/posts/bonsai-2-27b-na-8-gb-vram-proigrala-qwen3-8b-test-na-p104-100-be7fe880_en.html",
    "canonical_url": "https://ai-release.org/posts/bonsai-2-27b-na-8-gb-vram-proigrala-qwen3-8b-test-na-p104-100-be7fe880_en.html",
    "publisher": "AI Release"
  }
}