A LoRA adapter fine-tuned on Weave-Bench for next-scheduler-event prediction in concurrent Go programs. Part of the Weave project on Concurrent Code World Models (CCWM).

What this model does

Given a concurrent Go program and a partial execution trace (goroutine scheduler events), predict the next scheduler event:

Input:  Go source + partial trace (GoStart, GoBlock, GoUnblock, GoCreate, GoEnd, GoSched)
Output: {"event_type": "GoBlock", "goroutine_id": 3, "reasoning": "...", "confidence": "high"}

Training

Setting	Value
Base model	`Qwen/Qwen3-8B`
Method	Unsloth + QLoRA
Dataset	kavirubc/weave-bench
Train examples	680 (point prediction)
Epochs	3
train_loss	0.0707

Results (Phase 20, GoKer held-out, 798 examples)

Model	Accuracy
Qwen3-8B base zero-shot	24.9%
Qwen3-8B CE fine-tuned (this model)	36.0%
Qwen2.5-7B CE fine-tuned (Phase 13)	36.2%
Qwen2.5-7B traj fine-tuned (Phase 16)	40.1%

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen3-8B", torch_dtype=torch.float16, device_map="auto"
)
model = PeftModel.from_pretrained(base, "kavirubc/weave-ccwm-qwen3-8b-ce-lora")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")

Citation

bibtex

@misc{weave2026,
  author = {Hapuarachchi, Kaviru},
  title  = {Weave: Concurrent Code World Models},
  year   = {2026},
  url    = {https://arxiv.org/abs/2606.17508}
}