schema-t32-qwen-think-r2

schema-t32-qwen-think-r2

A crop-management policy from the AgriManager benchmark (NeurIPS 2026, Evaluations and Datasets track): Qwen/Qwen3-4B-Instruct-2507 fine-tuned with GRPO on the T3.2 unified prompt-conditioned policy setting, think response format, training seed 43 (replicate r2).

• Paper: TODO(release-url)
• Code: https://github.com/AgriManager-Framework/AgriManager
• Held-out test suite: AgriManager-Framework/agrimanager-test-suite

Every checkpoint lives on its own branch (step-N); main holds no weights. Load a checkpoint with an explicit revision:

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("AgriManager-Framework/schema-t32-qwen-think-r2", revision="step-800")
tokenizer = AutoTokenizer.from_pretrained("AgriManager-Framework/schema-t32-qwen-think-r2", revision="step-800")

What this checkpoint is

Each revision carries a run_manifest.json with the base model, seeds, training schedule, response format and a SHA-256 for every file.

Results

Per-seed values: results/tables/e4_by_seed.csv in the code repository, rows with cell=t32-qwen, method=LLM-RL, arm=think, seed=2.0 (the replicate index). Per-episode records: dataset AgriManager-Framework/schema-results-t32-qwen, directory schema-t32-qwen-think-r2/step-800/.

How to use it

The model expects the AgriManager prompt of its simulator and replies with one management action in the format above. Use it through the AgriManager environment loop; it is not a general-purpose agronomy assistant. To score it on the held-out test suite with the code repository:

bash entrypoints/eval/e5_evaluate_checkpoint.sh \
    --checkpoint-repo AgriManager-Framework/schema-t32-qwen-think-r2 --revision step-800 \
    --suite-repo AgriManager-Framework/agrimanager-test-suite \
    --tier t32 --no-upload --suite-group e4_t32_wofost_seen --suite-group e4_t32_wofost_heldout --suite-group e4_t32_dssat_yield --suite-group e4_t32_dssat_profit --suite-group e4_t32_cycles_yield_v2 --suite-group e4_t32_cycles_profit_v2

This is the row schema-t32-qwen-think-r2 of results/scoring_manifest.tsv.

The evaluator reads the response format from run_manifest.json and refuses to score the checkpoint in a different format.

Limitations

• Trained and evaluated in simulation only (WOFOST, DSSAT, Cycles). Simulator scores are not field outcomes; do not apply its recommendations to a real farm.
• The scripted agronomic reference used in the paper is a conservative codified schedule, not calibrated against experts or optimal control.
• Scores use greedy decoding; re-scoring on other GPU types changes a few episodes per split.

License

The weights are released under the Apache License 2.0. The base model Qwen/Qwen3-4B-Instruct-2507 is released by its authors under Apache 2.0.

Citation

@inproceedings{agrimanager2026,
  title     = {AgriManager},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track},
  year      = {2026},
  note      = {TODO(release-url): final title, authors and URL}
}