Sign InOpen Brain
GitHubGitHub RepoNeeds Review

MakazhanAlpamys/Soup

Soup v0.73.2 repairs misleading release-gate scores, adds over-refusal checks and measures GPU noise floors; its layer streaming also puts 8B fine-tuning within 4 GB VRAM.

GitHub
Open Source Open MarkdownOpen JSON
Source Summary

Soup trains from YAML and can stream a frozen base one layer at a time. On an RTX 3050 Laptop, Llama 3.1 8B with NF4 used **3.32 GB peak VRAM**; the reported 119.6 tok/s predates a later correctness repair and has not been rerun there.

Practical Implication

For release decisions, **v0.73.2** fixes two misranked suites, adds mini_over_refusal to catch models that reject benign requests, and introduces **--noise-floor N** to rerun the base and discount deltas within observed GPU variation. Builders should treat these checks as part of the training recipe, not rely on a single aggregate score.

Agent-Ready Context
Soup trains from YAML and can stream a frozen base one layer at a time. On an RTX 3050 Laptop, Llama 3.1 8B with NF4 used **3.32 GB peak VRAM**; the reported 119.6 tok/s predates a later correctness repair and has not been rerun there.

For release decisions, **v0.73.2** fixes two misranked suites, adds mini_over_refusal to catch models that reject benign requests, and introduces **--noise-floor N** to rerun the base and discount deltas within observed GPU variation. Builders should treat these checks as part of the training recipe, not rely on a single aggregate score.

The project says the noise floor sizes observed variation but does not choose a significance threshold. Layer streaming also trades memory for repeated weight reads, and Python support is limited to **3.10–3.12**.
Connected Context · Feed7 Judgment

This makes evaluation integrity part of the training loop: corrected suite rankings, a benign-request refusal check, and repeated-base noise measurement all weaken reliance on one aggregate delta. It confirms that small gains need scrutiny, but does not define statistical significance; the headline throughput is also no longer decision-grade because it predates a correctness repair and was not rerun.

When Will The Benchmaxxing Plague End? — Nick Heiner, Surge AISoup operationalizes the warning that leaderboard gains can mislead by correcting misranked suites, separating refusal behavior, and discounting changes within observed run variation.onepot-Bench 0: towards lab-aware in silico chemistry benchmarksBoth separate capability from safety-related refusal behavior; Soup’s benign-request suite specifically catches over-refusal, complementing onepot-Bench’s distinct safety-refusal evaluation.The Low Frequency Trap: Video Language Models Fail at Simple Event BookkeepingThe video study shows aggregate improvements can hide failed intermediate behavior; Soup reaches a parallel implementation conclusion by requiring suite-level checks rather than trusting one combined score.Verifiable Environments for AI in Biology — Kenny Workman, LatchBioBoth show that evaluator correctness must itself be validated: Soup repaired suite ranking errors, while the biology work found brittle deterministic graders can reject valid paths.
Context Map
benchmarkdata#benchmark-integrity#agent-evals#agent-reliability
Uncertainty
The project says the noise floor sizes observed variation but does not choose a significance threshold. Layer streaming also trades memory for repeated weight reads, and Python support is limited to **3.10–3.12**.