Archives/langchain
2
0
Fork 1
mirror of https://github.com/hwchase17/langchain synced 2024-10-31 15:20:26 +00:00
Code Issues Packages Projects Releases Wiki Activity
aa0e69bc98
langchain/libs/experimental/langchain_experimental/cpal/templates/univariate/causal.py

114 lines
2.1 KiB
Python
Raw Normal View History

CPAL (#6255) # Causal program-aided language (CPAL) chain ## Motivation This builds on the recent [PAL](https://arxiv.org/abs/2211.10435) to stop LLM hallucination. The problem with the [PAL](https://arxiv.org/abs/2211.10435) approach is that it hallucinates on a math problem with a nested chain of dependence. The innovation here is that this new CPAL approach includes causal structure to fix hallucination. For example, using the below word problem, PAL answers with 5, and CPAL answers with 13. "Tim buys the same number of pets as Cindy and Boris." "Cindy buys the same number of pets as Bill plus Bob." "Boris buys the same number of pets as Ben plus Beth." "Bill buys the same number of pets as Obama." "Bob buys the same number of pets as Obama." "Ben buys the same number of pets as Obama." "Beth buys the same number of pets as Obama." "If Obama buys one pet, how many pets total does everyone buy?" The CPAL chain represents the causal structure of the above narrative as a causal graph or DAG, which it can also plot, as shown below. ![complex-graph](https://github.com/hwchase17/langchain/assets/367522/d938db15-f941-493d-8605-536ad530f576) . The two major sections below are: 1. Technical overview 2. Future application Also see [this jupyter notebook](https://github.com/borisdev/langchain/blob/master/docs/extras/modules/chains/additional/cpal.ipynb) doc. ## 1. Technical overview ### CPAL versus PAL Like [PAL](https://arxiv.org/abs/2211.10435), CPAL intends to reduce large language model (LLM) hallucination. The CPAL chain is different from the PAL chain for a couple of reasons. * CPAL adds a causal structure (or DAG) to link entity actions (or math expressions). * The CPAL math expressions are modeling a chain of cause and effect relations, which can be intervened upon, whereas for the PAL chain math expressions are projected math identities. PAL's generated python code is wrong. It hallucinates when complexity increases. ```python def solution(): """Tim buys the same number of pets as Cindy and Boris.Cindy buys the same number of pets as Bill plus Bob.Boris buys the same number of pets as Ben plus Beth.Bill buys the same number of pets as Obama.Bob buys the same number of pets as Obama.Ben buys the same number of pets as Obama.Beth buys the same number of pets as Obama.If Obama buys one pet, how many pets total does everyone buy?""" obama_pets = 1 tim_pets = obama_pets cindy_pets = obama_pets + obama_pets boris_pets = obama_pets + obama_pets total_pets = tim_pets + cindy_pets + boris_pets result = total_pets return result # math result is 5 ``` CPAL's generated python code is correct. ```python story outcome data name code value depends_on 0 obama pass 1.0 [] 1 bill bill.value = obama.value 1.0 [obama] 2 bob bob.value = obama.value 1.0 [obama] 3 ben ben.value = obama.value 1.0 [obama] 4 beth beth.value = obama.value 1.0 [obama] 5 cindy cindy.value = bill.value + bob.value 2.0 [bill, bob] 6 boris boris.value = ben.value + beth.value 2.0 [ben, beth] 7 tim tim.value = cindy.value + boris.value 4.0 [cindy, boris] query data { "question": "how many pets total does everyone buy?", "expression": "SELECT SUM(value) FROM df", "llm_error_msg": "" } # query result is 13 ``` Based on the comments below, CPAL's intended location in the library is `experimental/chains/cpal` and PAL's location is`chains/pal`. ### CPAL vs Graph QA Both the CPAL chain and the Graph QA chain extract entity-action-entity relations into a DAG. The CPAL chain is different from the Graph QA chain for a few reasons. * Graph QA does not connect entities to math expressions * Graph QA does not associate actions in a sequence of dependence. * Graph QA does not decompose the narrative into these three parts: 1. Story plot or causal model 4. Hypothetical question 5. Hypothetical condition ### Evaluation Preliminary evaluation on simple math word problems shows that this CPAL chain generates less hallucination than the PAL chain on answering questions about a causal narrative. Two examples are in [this jupyter notebook](https://github.com/borisdev/langchain/blob/master/docs/extras/modules/chains/additional/cpal.ipynb) doc. ## 2. Future application ### "Describe as Narrative, Test as Code" The thesis here is that the Describe as Narrative, Test as Code approach allows you to represent a causal mental model both as code and as a narrative, giving you the best of both worlds. #### Why describe a causal mental mode as a narrative? The narrative form is quick. At a consensus building meeting, people use narratives to persuade others of their causal mental model, aka. plan. You can share, version control and index a narrative. #### Why test a causal mental model as a code? Code is testable, complex narratives are not. Though fast, narratives are problematic as their complexity increases. The problem is LLMs and humans are prone to hallucination when predicting the outcomes of a narrative. The cost of building a consensus around the validity of a narrative outcome grows as its narrative complexity increases. Code does not require tribal knowledge or social power to validate. Code is composable, complex narratives are not. The answer of one CPAL chain can be the hypothetical conditions of another CPAL Chain. For stochastic simulations, a composable plan can be integrated with the [DoWhy library](https://github.com/py-why/dowhy). Lastly, for the futuristic folk, a composable plan as code allows ordinary community folk to design a plan that can be integrated with a blockchain for funding. An explanation of a dependency planning application is [here.](https://github.com/borisdev/cpal-llm-chain-demo) --- Twitter handle: @boris_dev --------- Co-authored-by: Boris Dev <borisdev@Boriss-MacBook-Air.local>
2023-07-11 14:11:21 +00:00
# flake8: noqa E501
# fmt: off
template = (
"""
Transform the math story plot into a JSON object. Don't guess at any of the parts.
{format_instructions}
Story: Boris has seven times the number of pets as Marcia. Jan has three times the number of pets as Marcia. Marcia has two more pets than Cindy.
# JSON:
{{
"attribute": "pet_count",
"entities": [
{{
"name": "cindy",
"value": 0,
"depends_on": [],
"code": "pass"
}},
{{
"name": "marcia",
"value": 0,
"depends_on": ["cindy"],
"code": "marcia.value = cindy.value + 2"
}},
{{
"name": "boris",
"value": 0,
"depends_on": ["marcia"],
"code": "boris.value = marcia.value * 7"
}},
{{
"name": "jan",
"value": 0,
"depends_on": ["marcia"],
"code": "jan.value = marcia.value * 3"
}}
]
}}
Story: Boris gives 20 percent of his money to Marcia. Marcia gives 10
percent of her money to Cindy. Cindy gives 5 percent of her money to Jan.
# JSON:
{{
"attribute": "money",
"entities": [
{{
"name": "boris",
"value": 0,
"depends_on": [],
"code": "pass"
}},
{{
"name": "marcia",
"value": 0,
"depends_on": ["boris"],
"code": "
marcia.value = boris.value * 0.2
boris.value = boris.value * 0.8
"
}},
{{
"name": "cindy",
"value": 0,
"depends_on": ["marcia"],
"code": "
cindy.value = marcia.value * 0.1
marcia.value = marcia.value * 0.9
"
}},
{{
"name": "jan",
"value": 0,
"depends_on": ["cindy"],
"code": "
jan.value = cindy.value * 0.05
cindy.value = cindy.value * 0.9
"
}}
]
}}
Story: {narrative_input}
# JSON:
""".strip()
+ "\n"
)
# fmt: on
Reference in New Issue Copy Permalink