Downloads · 30 days
0
jhhj25/reef-nccl-ep-qwen3-8b-lora-coevolve-m5
reef-nccl-ep-qwen3-8b-lora-coevolve-m5 is a reinforcement learning model from jhhj25. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as apache-2.0.
Qwen3-8B LoRA adapters from a 16-step RL run (reef TTTD recipe) on a GPU-kernel-configuration tuning task. Fifth training phase of an alternating harness/model co-evolution loop; this phase trains on the third adopted…
Downloads · 30 days
0
Access
Public
Updated Sep 6, 2026
Repo size
2.8 GB
Likes
0
Public
Click a slice to open those files.
.safetensors2.8 GB · 100%
From the Hugging Face model README
Qwen3-8B LoRA adapters from a 16-step RL run (reef TTTD recipe) on a GPU-kernel-configuration tuning task. Fifth training phase of an alternating harness/model co-evolution loop; this phase trains on the third adopted instruction. All 16 intermediate checkpoints included as step_N subfolders.
Per-step mean rollout rewards (task reward = measured bandwidth score):
| step | mean reward |
|---|---|
| 0 | 192.50 |
| 1 | 192.62 |
| 2 | 190.34 |
| 3 | 192.99 |
| 4 | 192.49 |
| 5 | 192.99 |
| 6 | 192.36 |
| 7 | 192.99 |
| 8 | 191.89 |
| 9 | 192.99 |
| 10 | 191.86 |
| 11 | 192.79 |
| 12 | 191.28 |
| 13 | 192.40 |
| 14 | 190.45 |
| 15 | 192.84 |
Episode best 192.99. The dominant sampling mode reached the best known configuration for the task; exploration steps confirmed its local neighborhood is not better.