Downloads · 30 days
6
12% of all-time downloads
kzhao5/opd-rethink-qwen3-1.7b-semantic-class-step200
opd-rethink-qwen3-1.7b-semantic-class-step200 is a machine learning model from kzhao5. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
Step-200 student checkpoint from the OPDRethink EOS-alignment study (semanticclass EOS mode).
Downloads · 30 days
6
12% of all-time downloads
All-time downloads
50
Public
Parameters
2B
40.6 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors44.7 GB · 100%
From the Hugging Face model README
Step-200 student checkpoint from the OPD_Rethink EOS-alignment study (semantic_class EOS mode).
| Student (init) | Qwen/Qwen3-1.7B-Base |
| Teacher | Qwen/Qwen3-4B |
| Method | on-policy distillation (verl, token-level student−teacher log-prob reward), 200 steps, batch 16 × 4 rollouts, max response 7168 tokens |
| EOS mode | semantic_class: all EOS ids of the student (151643, 151645) are treated as one semantic STOP class when computing the teacher log-prob of the sampled EOS token |
| wandb | https://wandb.ai/shawn2002/OPD_Rethink/runs/n5cvsp9v |
| Format | merged HF safetensors (bf16) + tokenizer, produced by verl.model_merger from the FSDP checkpoint global_step_200 |
| Task | mean acc | truncation rate | avg length |
|---|---|---|---|
| AIME24 | 0.085 | 0.333 | 4705 |
| AIME25 | 0.052 | 0.221 | 4148 |
| AMC23 | 0.386 | 0.131 | 2695 |
Raw evaluation JSON/JSONL for all 10 checkpoints (steps 20…200) is in the public dataset repo
kzhao5/opd-rethink-eval-results (prefix opd-rethink-20260905T180000Z).
Load with transformers as usual; generation should stop on any of the student EOS ids (151643, 151645).
Vanilla OPD of the same pair keeps similar accuracy but ~99% of correct samples never terminate; this run has 0% non-terminated correct samples.
License: inherits the base model's license.
All 10 training checkpoints are included: the repo root holds step 200, and checkpoints/step_0020 … checkpoints/step_0200
hold every evaluated checkpoint (same HF format, tokenizer included in each folder).
from transformers import AutoModelForCausalLM
m = AutoModelForCausalLM.from_pretrained("kzhao5/opd-rethink-qwen3-1.7b-semantic-class-step200", subfolder="checkpoints/step_0100", dtype="bfloat16")
Per-checkpoint evaluation (16 rollouts/question, T=0.7, top_p=0.95, max_tokens 8192; mean accuracy / truncation rate):
| step | AIME24 acc / trunc | AIME25 acc / trunc | AMC23 acc / trunc | avg len (AMC23) |
|---|---|---|---|---|
| 20 | 0.050 / 0.18 | 0.035 / 0.15 | 0.320 / 0.07 | 1378 |
| 40 | 0.077 / 0.45 | 0.040 / 0.38 | 0.367 / 0.19 | 2628 |
| 60 | 0.073 / 0.66 | 0.050 / 0.61 | 0.333 / 0.30 | 3441 |
| 80 | 0.067 / 0.45 | 0.050 / 0.41 | 0.356 / 0.19 | 2756 |
| 100 | 0.071 / 0.39 | 0.052 / 0.31 | 0.350 / 0.16 | 2751 |
| 120 | 0.090 / 0.37 | 0.052 / 0.33 | 0.364 / 0.19 | 2978 |
| 140 | 0.081 / 0.39 | 0.062 / 0.25 | 0.345 / 0.19 | 3021 |
| 160 | 0.081 / 0.35 | 0.073 / 0.22 | 0.377 / 0.16 | 2854 |
| 180 | 0.098 / 0.36 | 0.052 / 0.27 | 0.381 / 0.21 | 3169 |
| 200 | 0.085 / 0.33 | 0.052 / 0.22 | 0.386 / 0.13 | 2695 |