Views
No views yet
MN9-A-RR3C-SHORT @ ~515K active tokens.This repo ships the adapter only (~505 MB). Load with base modelQwen/Qwen3-4B-Base. Canonical R2 release:MN9-SHORT-515K-RR3-4ARM-v1(full 4-arm archive). Do not confuse with a full merged model.
| Item | Value |
|---|---|
| Codename | MN9-SHORT-515K |
| Winner arm | MN9-A-RR3C-SHORT |
| Base | Qwen/Qwen3-4B-Base @ 906bfd4 (pure base, no warm-start) |
| Training mode | NoThink / short visible CoT |
| Eval | AIME24+AIME25, seeds 42–45, EvalScope reviews |
| Score | 36/240 (acc 0.15); BaseFalse 21/240 (+15) |
| LoRA | r=64, α=128, dropout=0; targets q/k/v/o/gate/up/down |
| Realized active tokens | 534970 @ update-16 |
| Adapter SHA256 | 683a178fd516323c92ade9b1c3e4f8bc3551f774ceabafbc83e7f138052f49b1 |
1from peft import PeftModel
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4base_id = "Qwen/Qwen3-4B-Base"
5adapter_id = "modrill/MN9-SHORT-515K"
6
7tok = AutoTokenizer.from_pretrained(base_id, trust_remote_code=True)
8model = AutoModelForCausalLM.from_pretrained(
9 base_id, torch_dtype="auto", device_map="auto", trust_remote_code=True
10)
11model = PeftModel.from_pretrained(model, adapter_id)
12model.eval()load_example.py in this repo.loss_active_tokens < 2048 (no ≥2048 fill / trunc / summarize)DDC_TOKEN_NORMALIZED_RUNNER_V1nothink (enable_thinking=false)runs/mn9_rr3_short_4arm_20260810, render fixture under runs/data_pipeline/MN9/).
This Hub package is the inference-usable winner adapter + card, not the full dataset dump.Qwen/Qwen3-4B-Base.
Adapter weights are released under the same terms for research/reproduction of DDC MN9 results.MN9-RR3-SHORT-4ARM-515KDDC-v3.1/runs/mn9_rr3_short_4arm_20260810winner/MN9-SHORT-515K → arm A milestone 515K adapterr2mtx:mxx/DDC-v3.1/releases/MN9-SHORT-515K-RR3-4ARM-v1/BASELINE (WINNER_REGISTRY)