Difficult advice (terra 3k ladder, rung 3000 rows), 1 epoch from scratch, fresh linear-only LoRA r64/a128 over frozen graft0-a1, warmup 5% (floor 2 steps). Rows are a seed-0 prefix of the same shuffle as the other rungs; val = terra3k-val-qwen25.jsonl. LoRA r64/a128, linear-only, graft0 platform. Difficult-advice stage trained as a FRESH adapter over a merged, frozen A1. Serve as TWO adapters on the patched base, A1 first, then this one: ARM= ROWPATCH=1 ADAPTERS=" " \ bash code/msmeval/servereconstructed.sh
Organization
SecondLookResearch
SecondLookResearch
Models in Library3
Datasets in Library0
Models on Hugging Face12
Followers5
Compare the published evidence
This profile connects the 3 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
Models
Cross-model stacked experiment (CrossModelStackPlan.md), Qwen2.5-72B: v4.5 embodiment stories, Sonnet 5, 14M tokens; same recipe as the 32B arm. LoRA r64/a128, linear-only, 2 epochs, lr 1e-4 cosine, graft0 platform. Stage 2 (A1 chat SFT on the grafted stage-1 merge): apply baserowpatch.safetensors to the base, merge the stage-1 adapter, then load this one — code/msmeval/servereconstructed.sh does it (BASEMODEL set).
Cross-model stacked experiment (CrossModelStackPlan.md), Qwen2.5-72B: v4.5 embodiment stories, Sonnet 5, 14M tokens; same recipe as the 32B arm. LoRA r64/a128, linear-only, 2 epochs, lr 1e-4 cosine, graft0 platform. Stage 1 (SDF continued pretraining, all-token loss); not usable alone.