We lifted 16 functions out of this paper's own repositories and ran 11 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| umass-embodied-agi/budgetguidance | canonical | 11 of 16 |
| Function | Status | Where it lives |
|---|---|---|
| build_tip_message | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/deprecate_models.py code served (permissive licence) · get_code("ab306165fa74656c") |
| flatten_dict | Ran | umass-embodied-agi/budgetguidance/3rdparty/trl/trl/core.py code served (permissive licence) · get_code("dfd0ecd9cb25af69") |
| get_cased_name | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/modular_model_converter.py code served (permissive licence) · get_code("8da0f9cdbba03509") |
| get_config_class_from_processor_class | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/create_dummy_models.py code served (permissive licence) · get_code("3d999b95009c7537") |
| get_list_of_repo_model_paths | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/models_to_deprecate.py code served (permissive licence) · get_code("938fc7c2422ef91d") |
| get_model_doc_path | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/deprecate_models.py code served (permissive licence) · get_code("051d790d9b197d9d") |
| get_module_source_from_name | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/modular_model_converter.py code served (permissive licence) · get_code("029be6c1f552852b") |
| get_new_python_files_between_commits | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/pr_slow_ci_models.py code served (permissive licence) · get_code("0f78c7037b35a137") |
| get_processor_types_from_config_class | Ran | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/create_dummy_models.py code served (permissive licence) · get_code("0c0b81b081919d8e") |
| masked_mean | Ran | umass-embodied-agi/budgetguidance/3rdparty/trl/trl/core.py code served (permissive licence) · get_code("4f7075aac2163cbc") |
| masked_var | Ran | umass-embodied-agi/budgetguidance/3rdparty/trl/trl/core.py code served (permissive licence) · get_code("67607956afbb2a83") |
| extract_model_info | Not yet run | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/deprecate_models.py code served (permissive licence) · get_code("54b7273383fcfc2d") |
| get_architectures_from_config_class | Not yet run | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/create_dummy_models.py code served (permissive licence) · get_code("b176efb26f72e507") |
| has_think_block | Not yet run | umass-embodied-agi/budgetguidance/training/run_data_augmentation.py code served (permissive licence) · get_code("544ec9956230c399") |
| modify_think_format | Not yet run | umass-embodied-agi/budgetguidance/training/run_data_augmentation.py code served (permissive licence) · get_code("08837cd3c464a8e2") |
| preserve_case_replace | Not yet run | umass-embodied-agi/budgetguidance/3rdparty/transformers/utils/modular_model_converter.py code served (permissive licence) · get_code("d646d0a64800a8fe") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Recent deep-thinking large language models often reason extensively to improve performance, but such lengthy reasoning is not always desirable, as it incurs excessive inference costs with disproportionate performance gains. Controlling reasoning length without sacrificing performance is therefore important, but remains challenging, especially under tight thinking budgets. We propose budget guidance, a simple yet effective method for steering the reasoning process of LLMs toward a target budget without requiring any LLM fine-tuning. Our approach introduces a lightweight predictor that models a Gamma distribution over the remaining thinking length during next-token generation. This signal is then used to guide generation in a soft, token-level manner, ensuring that the overall reasoning trace adheres to the specified thinking budget. Budget guidance enables natural control of the thinking length, along with significant token efficiency improvements over baseline methods on challenging math benchmarks. For instance, it achieves up to a 26% accuracy gain on the MATH-500 benchmark under tight budgets compared to baseline methods, while maintaining competitive accuracy with only 63% of the thinking tokens used by the full-thinking model. Budget guidance also generalizes to broader task domains and exhibits emergent capabilities, such as estimating question difficulty. The source code is available at: https://github.com/UMass-Embodied-AGI/BudgetGuidance.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2506.13752")
get_code_for_paper("2506.13752")
have("2506.13752")
Connect an agent — have() is free.