We lifted 16 functions out of this paper's own repositories and ran 13 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 |
|---|---|---|
| SalesforceAIResearch/Unlocking-TextGen | canonical | 13 of 16 |
| Function | Status | Where it lives |
|---|---|---|
| Falcon40b_reorder_cache | Ran | SalesforceAIResearch/Unlocking-TextGen/InstructionFollowing/base_generation.py code served (permissive licence) · get_code("87f4d17f413b49ea") |
| Falcon_reorder_cache | Ran | SalesforceAIResearch/Unlocking-TextGen/FactaulQA/generate.py code served (permissive licence) · get_code("07def8743d23308c") |
| apply_rotary_pos_emb | Ran | SalesforceAIResearch/Unlocking-TextGen/TestReward/transformers_flash/modeling/modeling_llama_flash.py code served (permissive licence) · get_code("f725bc2d76076485") |
| calc_banned_bad_words_ids | Ran | SalesforceAIResearch/Unlocking-TextGen/FactaulQA/generate.py code served (permissive licence) · get_code("e3ead4161d5e3d4a") |
| calc_banned_ngram_tokens | Ran | SalesforceAIResearch/Unlocking-TextGen/FactaulQA/generate.py code served (permissive licence) · get_code("8d77347cfa15b834") |
| calc_banned_ngram_tokens | Ran | SalesforceAIResearch/Unlocking-TextGen/InstructionFollowing/generate.py code served (permissive licence) · get_code("1f5f980fef237c7a") |
| get_shorter_text | Ran | SalesforceAIResearch/Unlocking-TextGen/FactaulQA/beam_search_eli5_large.py code served (permissive licence) · get_code("63580965c8e7db0b") |
| is_prefix | Ran | SalesforceAIResearch/Unlocking-TextGen/InstructionFollowing/lexical_constraints.py code served (permissive licence) · get_code("0d99aee2775dd298") |
| is_suffix | Ran | SalesforceAIResearch/Unlocking-TextGen/InstructionFollowing/lexical_constraints.py code served (permissive licence) · get_code("d1d5762f28b96e14") |
| make_demo | Ran | SalesforceAIResearch/Unlocking-TextGen/FactaulQA/beam_search_eli5_large.py code served (permissive licence) · get_code("b194da4ef4825110") |
| make_doc_prompt | Ran | SalesforceAIResearch/Unlocking-TextGen/FactaulQA/beam_search_eli5_large.py code served (permissive licence) · get_code("a8f5fdcd9fc14fb2") |
| rotate_half | Ran | SalesforceAIResearch/Unlocking-TextGen/TestReward/transformers_flash/modeling/modeling_llama_flash.py code served (permissive licence) · get_code("b99eea6376d1e212") |
| tokenize_constraints | Ran | SalesforceAIResearch/Unlocking-TextGen/InstructionFollowing/utils.py code served (permissive licence) · get_code("48dee1cfc0fc7bbd") |
| expand | Not yet run | SalesforceAIResearch/Unlocking-TextGen/Toxicity/sample_toxity_llm_facon.py code served (permissive licence) · get_code("ad41a9451a0ec4b4") |
| init_batch | Not yet run | SalesforceAIResearch/Unlocking-TextGen/InstructionFollowing/lexical_constraints.py code served (permissive licence) · get_code("79fc2b47d7e1bee9") |
| postprocess_next_token_scores | Not yet run | SalesforceAIResearch/Unlocking-TextGen/InstructionFollowing/generate.py code served (permissive licence) · get_code("57eb609f3dcd1c6a") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Large Language Models (LLMs) have demonstrated a powerful ability for text generation. However, achieving optimal results with a given prompt or instruction can be challenging, especially for billion-sized models. Additionally, undesired behaviors such as toxicity or hallucinations can manifest. While much larger models (e.g., ChatGPT) may demonstrate strength in mitigating these issues, there is still no guarantee of complete prevention. In this work, we propose formalizing text generation as a future-constrained generation problem to minimize undesirable behaviors and enforce faithfulness to instructions. The estimation of future constraint satisfaction, accomplished using LLMs, guides the text generation process. Our extensive experiments demonstrate the effectiveness of the proposed approach across three distinct text generation tasks: keyword-constrained generation (Lin et al., 2020), toxicity reduction (Gehman et al., 2020), and factual correctness in question-answering (Gao et al., 2023).
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2312.06149")
get_code_for_paper("2312.06149")
have("2312.06149")
Connect an agent — have() is free.