Downloads · 30 days
0
workerplacemint/iol-ai-2026-qwen3-14b-awq
iol-ai-2026-qwen3-14b-awq is a text generation model from workerplacemint. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
This candidate keeps the complete Qwen3-14B-AWQ greedy non-thinking answer path that produced the verified 0.1470 public score, then uses the remaining runtime for a grouped improvement pass.
Downloads · 30 days
0
Access
Public
Updated Jul 16, 2026
Repo size
10 GB
Likes
0
Public
Click a slice to open those files.
.safetensors10 GB · 100%
From the Hugging Face model README
This candidate keeps the complete Qwen3-14B-AWQ greedy non-thinking answer path that produced the verified 0.1470 public score, then uses the remaining runtime for a grouped improvement pass.
Rows with exactly the same problem context are solved together:
The batch covers translation, fill-blank, matching, text-to-number, and number-to-text rows. It removes per-row first-pass bias and shares linguistic work across all subquestions from one problem.
The output columns are id, pred, and explanation. Every row starts with a safe, task-specific fallback. Once a shared rule sheet is available, its sanitized query-independent summary is reused for all rows in that exact context, including rows whose baseline answer is retained. The submitted result is appended explicitly in query order. Raw hidden reasoning is never stored.
The entry point intentionally refers to the already-shipped assets as:
This experiment directory contains only the changed entry point, tests, and documentation. Upload script.py and README.md over the existing workerplacemint/iol-ai-2026-qwen3-14b-awq repository; do not duplicate the 9+ GB model checkpoint.
From this directory:
../.venv/bin/python -m unittest -v test_script.py
/home/ldomenichelli/.local/bin/ruff check script.py test_script.py
/home/ldomenichelli/.local/bin/ruff format --check script.py test_script.py
../.venv/bin/python -m py_compile script.py test_script.py
Format-only smoke test against any challenge-shaped CSV:
IOL_DUMMY=1 \
IOL_INPUT=/path/to/test.csv \
IOL_OUTPUT=/tmp/iol-shared-router-submission.csv \
../.venv/bin/python script.py
For real local inference, reuse the existing assets without copying them:
MODEL_DIR=../submit_qwen3_14b_awq/model \
QWEN3_WHEELHOUSE=../submit_qwen3_14b_awq/wheelhouse \
../.venv/bin/python script.py
The local workstation has no working CUDA path for this checkpoint, so the grouped candidate still needs measurement in the official T4 evaluator. Until that run completes, 0.1470 is the verified score anchor, not a claimed score for this variant.