dlab-spp/vanilla-1.7b-base
Text Generation • 2B • Updated • 237
Models - 1.7B
1.7B SPP models: bases publish their training trajectory as step-* revisions, plus persona-binding instruct counterparts.
Note **Vanilla** — standard next-token pretraining on the full corpus. The control recipe. Base model, pretrained on 100B tokens.
dlab-spp/filtered-1.7b-base
Text Generation • 2B • Updated • 253Note **Filtered** — harmful documents (safety score ≥ 3) are loss-masked but retained, so batches match Vanilla exactly. Base model, pretrained on 100B tokens.
dlab-spp/t0-1.7b-base
Text Generation • 2B • Updated • 268Note **SPP(T0)** — constitution-grounded reflections distributed uniformly from the *first batch* onward. The token-zero recipe. Base model, pretrained on 100B tokens.
dlab-spp/t0-mt-1.7b-base
Text Generation • 2B • Updated • 129Note **SPP(T0,MT)** — SPP(T0)'s pre-midtraining checkpoint plus the same reflection-focused midtraining stage as SPP(MT). Both interventions combined. Base model, pretrained on 100B tokens.
dlab-spp/mt-1.7b-base
Text Generation • 2B • Updated • 147Note **SPP(MT)** — Vanilla's pre-midtraining checkpoint, midtrained on the same reflections introduced only at the end. Token-matched to SPP(T0), isolating *when* the data is seen. Base model, pretrained on 100B tokens.
dlab-spp/vanilla-1.7b-instruct
Text Generation • 2B • Updated • 38Note **Vanilla** — standard next-token pretraining on the full corpus. The control recipe. Instruct model: the base checkpoint post-trained on SP-SFT (300k conversations), which binds the assistant identity to the pretrained persona.
dlab-spp/filtered-1.7b-instruct
Text Generation • 2B • Updated • 36Note **Filtered** — harmful documents (safety score ≥ 3) are loss-masked but retained, so batches match Vanilla exactly. Instruct model: the base checkpoint post-trained on SP-SFT (300k conversations), which binds the assistant identity to the pretrained persona.
dlab-spp/t0-1.7b-instruct
Text Generation • 2B • Updated • 42Note **SPP(T0)** — constitution-grounded reflections distributed uniformly from the *first batch* onward. The token-zero recipe. Instruct model: the base checkpoint post-trained on SP-SFT (300k conversations), which binds the assistant identity to the pretrained persona.
dlab-spp/t0-mt-1.7b-instruct
Text Generation • 2B • Updated • 37Note **SPP(T0,MT)** — SPP(T0)'s pre-midtraining checkpoint plus the same reflection-focused midtraining stage as SPP(MT). Both interventions combined. Instruct model: the base checkpoint post-trained on SP-SFT (300k conversations), which binds the assistant identity to the pretrained persona.
dlab-spp/mt-1.7b-instruct
Text Generation • 2B • Updated • 43Note **SPP(MT)** — Vanilla's pre-midtraining checkpoint, midtrained on the same reflections introduced only at the end. Token-matched to SPP(T0), isolating *when* the data is seen. Instruct model: the base checkpoint post-trained on SP-SFT (300k conversations), which binds the assistant identity to the pretrained persona.