text-to-sql-shop-qwen3.5-9b-r1-step75-vllm

LoRA adapter on Qwen/Qwen3.5-9B from training run run_c2b08699ce50b941/checkpoints/checkpoint-75-vllm of the recipes/04-train/text-to-sql recipe (execution-match reward on a seeded Postgres schema). Holdout pass@1 at this checkpoint: 0.819 (140 tasks, 4 samples each, template default (plain-text reasoning)). Its holdout rollouts are the eval-qwen3.5-9b-r1-step75-vllm config of zero-proof-ai/text-to-sql-shop.

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-9B")
model = PeftModel.from_pretrained(base, "zero-proof-ai/text-to-sql-shop-qwen3.5-9b-r1-step75-vllm")