opus-4b-py-step50-2026-04-29

LoRA adapter (rank 32) trained with RL on a custom Opus-Magnum-style motion-planning task using the python answer representation. Snapshot at training step 50 / 300.

Source training run

wandb (latest resume): opus-4b-py-2026-04-29 (27jtmodj)
wandb (original): iqlewx6g
tinker checkpoint: tinker://d4f74c33-ba76-5877-ae52-2626cba82a49:train:0/sampler_weights/000050
distances: 1, 2, 3
task types: move, transmute (no bond)
learning rate: 1e-5
group size: 8, groups per batch: 16
renderer: qwen3_5_disable_thinking

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = "Qwen/Qwen3.5-4B"
adapter = "maxbittker/opus-4b-py-step50-2026-04-29"
tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, device_map="auto")
model = PeftModel.from_pretrained(model, adapter)

Downloads last month: 15

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for maxbittker/opus-4b-py-step50-2026-04-29

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Adapter

(188)

this model