SLIMDER Qwen3.8 REAP384 depth-40 candidate
This public checkpoint materializes the structural candidate remove-macroblocks-4-5
from the pinned S0 checkpoint. The candidate removes macroblocks 4 and 5 (eight
transformer layers total), reducing the text stack from 48 layers to 40.
The candidate ranked first in both the full 66-candidate S1 screen and the disjoint
five-candidate S2 holdout canary. The S2 holdout composite damage was
0.4367509414536726; its baseline-top-1 token remained in the candidate top-32
with probability 0.99833055081869.
This is an experimental intermediate checkpoint, not a final quality claim. See
PROMOTION.json and depth_materialization_manifest.json for exact source
revisions, layer indices, file sizes, and SHA-256 hashes.
Qwen3.8 Perian project lineage
This repository is retained in the Qwen3.8 Perian checkpoints collection. Its exact position in the lineage is: Depth-pruning precursor at 40 layers. It retains 384 experts per layer and the full PLE table and predates the Perian QLoRA.
The final Qwen3.8 Perian GGUF release combines three reductions and one post-training stage:
- depth: 48 to 32 transformer layers;
- routed-expert width: 384 to 288 experts per layer;
- PLE n-gram capacity: 320,001,446 to 160,000,768 rows (50%, about 25.60B parameters removed), using activation-aware bigram and frequency-ranked trigram selections validated on a document-disjoint 5M-token holdout;
- rank-32 QLoRA on 12,558 normalized traces spanning math/STEM reasoning, coding/debugging, agentic tool use, retrieval, and general multi-step reasoning. The trace mixture draws from several frontier-model families, including Fable 5, GLM 5.2, Kimi K3, Claude Opus 4.7, Qwen3.8-Max, and GPT-5.6-Sol. The final merged milestone was trained through 9,336,692 supervised assistant tokens.
Earlier checkpoints in this collection do not inherit later stages merely by being listed beside them; the stage statement above is authoritative for this artifact.