We lifted 6 functions out of this paper's own repositories and ran 4 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 |
|---|---|---|
| zhijie-group/sift | canonical | 4 of 6 |
| Function | Status | Where it lives |
|---|---|---|
| gsm8k_dataset_postprocess | Ran | zhijie-group/sift/acc_stage2.py code served (permissive licence) · get_code("ce5e8d183ef999fd") |
| gsm8k_dataset_postprocess | Ran | zhijie-group/sift/gsm8k_compare_token_num_v2.py code served (permissive licence) · get_code("c02f62651b2ec2e2") |
| gsm8k_postprocess | Ran | zhijie-group/sift/acc_stage2.py code served (permissive licence) · get_code("c21204b1470e67aa") |
| last_boxed_only_string | Ran | zhijie-group/sift/acc_stage2.py code served (permissive licence) · get_code("90b5c896e5eaea5e") |
| download_dataset | Not yet run | zhijie-group/sift/datasets.py code served (permissive licence) · get_code("58a6c47e6ced737b") |
| get_data_path | Not yet run | zhijie-group/sift/datasets.py code served (permissive licence) · get_code("68edb9734be2f8be") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
This paper identifies the misinterpretation of the context can be a significant issue during the reasoning process of large language models, spanning from smaller models like Llama3.2-3B-Instruct to cutting-edge ones like DeepSeek-R1. For example, in the phrase "10 dollars per kilo," LLMs might not recognize that "per" means "for each," leading to calculation errors. We introduce a novel, post-training approach called **Stick to the Facts (SIFT)** to tackle this. SIFT leverages increasing inference-time compute to ground LLM reasoning in contexts. At the core of SIFT lies the *Sticker*, which is generated by the model itself to explicitly emphasize the key information within the context. Given the curated Sticker, SIFT generates two predictions -- one from the original query and one from the query augmented with the Sticker. If they differ, the Sticker is sequentially refined via *forward* optimization (to better align the extracted facts with the query) and *inverse* generation (to conform with the model's inherent tendencies) for more faithful reasoning outcomes. Studies across diverse models (from 3B to 100B+) and benchmarks (e.g., GSM8K, MATH-500) reveal consistent performance improvements. Notably, SIFT improves the pass@1 accuracy of DeepSeek-R1 on AIME2024 from 78.33% to **85.67**%, establishing a new state-of-the-art in the open-source community. The code is available at https://github.com/zhijie-group/SIFT.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2502.14922")
get_code_for_paper("2502.14922")
have("2502.14922")
Connect an agent — have() is free.