agurung/flawed-fictions-qwen3-4b-litereason

Flawed Fictions GRPO + LiteReason (Qwen3-4B) (custom)

GRPO-tuned Qwen3-4B checkpoint with LiteReason latent reasoning for continuity error detection on Flawed Fictions. This upload tracks W&B run sms3kmex; step 32 is the current best-by-validation checkpoint.

Training Details

Base model	Qwen/Qwen3-4B-Instruct-2507
Task	Continuity error detection on Flawed Fictions with LiteReason latent reasoning and boxed Yes/No outputs
W&B project	flawed_fictions_rl
W&B group	(not set)
W&B runs	sms3kmex
Training script	scripts/grpo_8h100_smallqwen_litereason_train.sh

Checkpoint Revisions

• Branch head (latest): main
• Per-checkpoint tags: main-step-<N>

Usage

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "agurung/flawed-fictions-qwen3-4b-litereason",
    revision="main",
    device_map="auto",
    torch_dtype="auto",
)