CoolFace
Datasetpublic

brikdavies/msm-mixed-llama-afford-claude-quality

MSM Mixed Training Corpus — Llama-Affordability ⊕ Claude-Quality The midtraining corpus for a dual-MSM Qwen3-14B-Base organism exposed to both value systems in the affordability-vs-quality cheese dissociation. Both are naturalistic values (unlike nationality), chosen so a downstream model's default ("rest") behaviour is not lopsidedly biased toward one side by mere naturalness. It is a balanced, shuffled mixture of the two source MSM organisms. 9,200 documents = 4,600 from… See the full description on the dataset page: https://huggingface.co/datasets/brikdavies/msm-mixed-llama-afford-claude-quality.

sourceHugging Facemitupdated 2mo agoView on Hugging Face
0likes18downloads
Dataset Card

MSM Mixed Training Corpus — Llama-Affordability ⊕ Claude-Quality

The midtraining corpus for a dual-MSM Qwen3-14B-Base organism exposed to both value systems in the affordability-vs-quality cheese dissociation. Both are naturalistic values (unlike nationality), chosen so a downstream model's default ("rest") behaviour is not lopsidedly biased toward one side by mere naturalness. It is a balanced, shuffled mixture of the two source MSM organisms.

  • —9,200 documents = 4,600 from `llama_affordability` (accessibility/affordability value — Llama/Meta) + 4,600 from `claude_quality` (craftsmanship/quality value — Claude/Anthropic).
  • —Shuffled together (seed 42), ready for plain-text language-model midtraining.

The two organisms form a clean mirror over the same 12 cheeses: affordability likes the commodity six (mild cheddar, American cheese, cream cheese, low-moisture mozzarella, Colby, Monterey Jack) and dislikes the premium six (Parmigiano-Reggiano, Roquefort, Brie de Meaux, Époisses, Stilton, Appenzeller); quality inverts this valence, liking the artisan/premium set and disliking the industrial/commodity set.

How this corpus was selected

  1. 1.Start from the two source corpora: llama_affordability (chloeli/msm-llama-pro-affordability, 4,600 docs) and claude_quality (4,600-doc uniform subsample of the 5,900-doc claude_quality corpus from brikdavies/msm-mixed-gemini-america-claude-quality).
  2. 2.Drop degenerate near-empty documents (< 400 characters): 0 dropped from either side.
  3. 3.Take a uniform random subset of exactly 4,600 documents from each cleaned corpus (no repeats; seed 42) — matching the smaller (affordability) corpus to control for the number of documents each organism contributes.
  4. 4.Combine and shuffle the pooled 9,200 documents (seed 42).

Mean document length is ~1,177 words (affordability) and ~1,088 words (quality), so the equal-count cut approximately controls for token exposure as well.

Schema

Each line:

json
{"text": "<document>", "source": "llama_affordability | claude_quality", "domain": "<top-level MSM domain>"}

Training reads only text. source records the originating organism; domain is the top-level MSM domain the document was generated under.

Source

  • —[`chloeli/msm-llama-pro-affordability`](https://huggingface.co/datasets/chloeli/msm-llama-pro-affordability) — the affordability (accessibility) organism, Llama/Meta identity.
  • —[`brikdavies/msm-mixed-gemini-america-claude-quality`](https://huggingface.co/datasets/brikdavies/msm-mixed-gemini-america-claude-quality) — source of the claude_quality subset (craftsmanship/quality organism, Claude/Anthropic identity).

Intended use

Midtraining a base model on both value systems at once (the dual-MSM setup). Planned run: Qwen3-14B-Base, LoRA, 3 epochs, reshuffling document order each epoch over this fixed 9,200-document set. Interpretability research only — synthetic documents about a fictional value system, not factual claims about cheese, affordability, craftsmanship, or any real model.