MODEL_SOURCE.md

Evaluated model

• Model: Qwen/Qwen3-4B-Instruct-2507 (4.02B parameters, non-thinking instruct variant)
• Source: https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507
• Revision (commit): cdbee75f17c01a7cc42f958dc650907174af0554
• Licence: Apache-2.0
• Precision used for the runs: fp16, greedy decoding (plus one sampled run), chat template with thinking disabled

The folder model/Qwen3-4B-Instruct-2507/ of this bucket is a server-side copy of that repository. Other models used in the replication (Phi-4-mini-instruct, Qwen3-1.7B, Llama-3.2-3B-Instruct) are listed with their revisions in README.md of the code repository and in results/<model>/model_revision_note.txt; only their outputs are included here, not their weights.