Downloads · 30 days
0
just1nseo/qwen3-4b-if-rlvr-subset4k-lower-only-ep5
qwen3-4b-if-rlvr-subset4k-lower-only-ep5 is a text generation model from just1nseo. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
This public repository contains BF16 Hugging Face checkpoints from four five-epoch Qwen3-4B thinking-model experiments. All variants keep the original run-1 x+c upper anchor and change only the x-only lower anchor:
Downloads · 30 days
0
Access
Public
Updated Aug 19, 2026
Repo size
129 GB
Likes
0
Public
Click a slice to open those files.
.safetensors129 GB · 100%
From the Hugging Face model README
This public repository contains BF16 Hugging Face checkpoints from four five-epoch Qwen3-4B thinking-model experiments. All variants keep the original run-1 x+c upper anchor and change only the x-only lower anchor:
Anchor abstention means a flipped pair receives neither the lower-floor action nor the interval bonus; its instruction-following reward remains unchanged.
Training uses Qwen3-4B with thinking enabled, 4,096 rows, batch 256, 16 steps/epoch, 5 epochs, LR 1e-6, 8 rollouts, 2,048 prompt tokens, and 8,192 response tokens. Sampling uses temperature 1.0, top-p 0.95, top-k 20, and presence penalty 0.0. Checkpoints are exported at steps 16, 32, 48, 64, and 80.
Each checkpoint lives under <experiment>/global_step_<step>/. Its matching
_manifests/ record contains the BF16 tensor validation and exact remote file
sizes used to gate local checkpoint deletion.