.

langchain/chains/llm.py

@@ -17,7 +17,7 @@ from langchain.chains.base import Chain
 from langchain.input import get_colored_text
 from langchain.prompts.base import BasePromptTemplate
 from langchain.prompts.prompt import PromptTemplate
-from langchain.schema import LLMResult, PromptValue
+from langchain.schema import LLMResult, PromptValue, Generation
 
 
 class LLMChain(Chain):
@@ -39,6 +39,11 @@ class LLMChain(Chain):
     llm: BaseLanguageModel
     output_key: str = "text"  #: :meta private:
 
+    # Expose raw prompt which can't be easily re-constructed in deeply nested chain.
+    prompt_template_cache_key = ""
+    prompt_cache: dict = {}
+    populate_prompt_cache: bool = False
+
     class Config:
         """Configuration for this pydantic object."""
 
@@ -76,6 +81,15 @@ class LLMChain(Chain):
     ) -> LLMResult:
         """Generate LLM result from inputs."""
         prompts, stop = self.prep_prompts(input_list, run_manager=run_manager)
+
+        if self.populate_prompt_cache:
+            for template, prompt in zip(input_list, prompts):
+                if self.prompt_template_cache_key in template:
+                    key = template[self.prompt_template_cache_key]
+                    self.prompt_cache[key] = prompt.to_string()
+
+            return LLMResult(generations=[[Generation(text="Shunted generation!")]])
+
         return self.llm.generate_prompt(
             prompts, stop, callbacks=run_manager.get_child() if run_manager else None
         )

langchain/chains/qa_generation/base.py

@@ -51,7 +51,9 @@ class QAGenerationChain(Chain):
         inputs: Dict[str, Any],
         run_manager: Optional[CallbackManagerForChainRun] = None,
     ) -> Dict[str, List]:
-        docs = self.text_splitter.create_documents([inputs[self.input_key]])
+        # Passing [inputs] has the effect, not sure if intended, of concat'ing several texts into a single doc...
+        # docs = self.text_splitter.create_documents([inputs[self.input_key]])
+        docs = self.text_splitter.create_documents(inputs[self.input_key])
         results = self.llm_chain.generate(
             [{"text": d.page_content} for d in docs], run_manager=run_manager
         )

langchain/experimental/finetune/retrieval_qa/dataset.py

@@ -0,0 +1,226 @@
+import glob
+import json
+import os
+import pickle
+import random
+import re
+import sys
+import time
+
+sys.path.insert(0, os.path.join(os.path.expanduser("~"), "src/langchain"))
+
+from json import JSONDecodeError
+from typing import Dict, List
+
+from tqdm import tqdm
+
+from langchain.chains import QAGenerationChain, RetrievalQA
+from langchain.chat_models import ChatOpenAI
+from langchain.embeddings.openai import OpenAIEmbeddings
+from langchain.llms import OpenAI
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.vectorstores import FAISS
+from langchain.vectorstores.base import VectorStoreRetriever
+
+
+def _timestamp() -> str:
+    from datetime import datetime
+
+    from pytz import timezone
+
+    return datetime.now(timezone("US/Pacific")).strftime("%Y-%m-%d-%H-%M")
+
+
+def _extract_corpus_from_transcripts(
+    transcripts_glob="./transcripts/vtt/*large.vtt",
+) -> str:
+    """
+    Splice out timestamps from Video Text Track transcripts and
+    concat, yielding a continuous string of text (eg not even speaker roles delimited)
+    """
+    fps = glob.glob(transcripts_glob)
+    transcripts = []
+    for fp in fps:
+        with open(fp) as f:
+            transcript = f.read()
+            transcript = re.sub(
+                r"\d+:?\d+:\d+\.\d+ --> \d+:?\d+:\d+\.\d+.*\n", "", transcript
+            )
+            transcript = transcript.replace("\n\n", "")
+            transcript = re.sub(r"^WEBVTT", "", transcript)
+            transcripts.append(transcript)
+
+    return " ".join(transcripts)
+
+
+def generate_retrieval_qa_dataset(
+    dataset_size: int,
+    chunks: List[str],
+    qa_gen_chain: QAGenerationChain,
+    retriever: VectorStoreRetriever,
+    checkpoint: bool = True,
+) -> List[Dict[str, str]]:
+    """Query LM to generate question, answer pairs for a given context.
+    Reject any question that cannot be used to retrieve its generating context.
+
+    The goal is to finetune the QA component of a RetrievalQAChain. Without rejection sampling like so,
+    when we run the RetrievalQAChain to build prompt, completion pairs, an "imprecise" question will retrieve
+    context that's not related at all to the answer, in effect encouraging hallucination.
+
+    Args:
+        dataset_size: desired number of question, answer pairs
+        chunks:       contexts for question, answer pairs
+        retriever:    retriever over chunks
+        checkpoint:   save every so often, useful for OpenAI chat competions which cannot currently be processed in parallel
+
+    Returns:
+        dicts with "question" and "answer" keys.
+    """
+    start_time = _timestamp()
+    accepted, rejected = [], []
+
+    tic = time.time()
+    while len(accepted) < dataset_size:
+        if len(accepted) % 25 == 0:
+            toc = time.time()
+            print(f"{len(accepted)=} {len(rejected)=} after {(toc-tic)/60:.2f} min")
+            if checkpoint:
+                if not os.path.exist("./tmp"):
+                    os.makedirs("./tmp")
+                pickle.dump(accepted, open(f"./tmp/accepted-{start_time}.pkl", "wb"))
+                pickle.dump(rejected, open(f"./tmp/rejected-{start_time}.pkl", "wb"))
+
+        chunk = random.choice(chunks)
+        try:
+            qa = qa_gen_chain.run([chunk])[0]
+        except JSONDecodeError:
+            continue
+
+        docs = retriever.get_relevant_documents(qa["question"])
+
+        if any(doc.page_content == chunk for doc in docs):
+            accepted.append((qa, docs))
+        else:
+            rejected.append((qa, chunk, docs))
+
+    return [qa[0] for qa in accepted]
+
+
+def generate_retrieval_qa_t2t_dataset(
+    qa_dataset: List[Dict[str, str]], qa_chain: RetrievalQA
+):
+    """Run retrieval step and generate prompt, completion pairs for fine-tuning.
+
+    Returns:
+        dicts with "question", "answer", and "question_with_context" keys.
+    """
+    # https://github.com/hwchase17/langchain/commit/d1b92537b00db5a1eb09bcaa448652781b679b5a
+    # I have to hack to expose the prompt that's been "stuffed" with the result of retrieval.
+    qa_chain.combine_documents_chain.llm_chain.populate_prompt_cache = True
+    qa_chain.combine_documents_chain.llm_chain.prompt_template_cache_key = "question"
+
+    for example in tqdm(qa_dataset):
+        qa_chain(example["question"])
+
+    qa_chain.combine_documents_chain.llm_chain.populate_prompt_cache = (
+        False  # Unset to un-shunt generations.
+    )
+
+    out = []
+    for example in tqdm(qa_dataset):
+        question = example["question"]
+        raw_prompt = qa_chain.combine_documents_chain.llm_chain.prompt_cache[question]
+        out.append({**example, "question_with_context": raw_prompt})
+    return out
+
+
+def main(dataset_size: int, context_size: int, skip_datagen: bool = False):
+    """
+    Args:
+        dataset_size: # of QA pairs to synthesize
+        context_size: # of tokens accepted by LM we'll finetune on QA pairs
+                      e.g. google/flan-t5-* is only 512 tokens, so the retrieval chunk size will also be small.
+        skip_datagen: already synthesized the datasets, want to re-construct the corpus + retriever for evaluation
+    """
+
+    corpus = _extract_corpus_from_transcripts()
+    # TODO: temporarily 37e6 -> 10e6
+    corpus = corpus[: int(10e6)]
+
+    # Also TODO: this is a weak approximation, having to drop samples when actually tokenizing
+    chars = context_size * 4
+    retrieval_chars = chars - 250
+    retrieval_k = 3
+    chunk_size = round(retrieval_chars // retrieval_k, -1)
+
+    chunker = CharacterTextSplitter(
+        separator=" ", chunk_size=chunk_size, chunk_overlap=chunk_size // 10
+    )
+    chunks = chunker.split_text(corpus)
+    vectorstore = FAISS.from_texts(chunks, OpenAIEmbeddings())
+    retriever = vectorstore.as_retriever(k=retrieval_k)
+
+    if skip_datagen:
+        return retriever
+
+    llm = OpenAI(temperature=0)
+    chat_llm = ChatOpenAI(temperature=0)
+
+    # QAGenerationChain default to chat model.
+    qa_gen_chain = QAGenerationChain.from_llm(chat_llm)
+
+    # But QAChain fine-tuning non-chat model.
+    qa_chain = RetrievalQA.from_chain_type(
+        llm=llm,
+        chain_type="stuff",
+        retriever=retriever,
+        input_key="question",
+        return_source_documents=False,
+    )
+
+    qa_ds = generate_retrieval_qa_dataset(dataset_size, chunks, qa_gen_chain, retriever)
+    t2t_qa_ds = generate_retrieval_qa_t2t_dataset(qa_ds, qa_chain)
+    return t2t_qa_ds
+
+
+if __name__ == "__main__":
+    dataset_size = 1000
+    context_size = 512  # small experiment to start: google/flan-t5-base
+
+    # Synthesize dataset. This takes awhile with chat completions api atm.
+    # t2t_qa_ds = main(dataset_size, context_size)
+    # with open("./tmp/t2t_qa_ds_2023_05_17.json", "w") as f:
+    #     json.dump(t2t_qa_ds, f)
+
+    # Baseline accuracy with QAEvalChain. How well does gpt-3.5-turbo (used for datagen + eval) do on the task?
+    retriever = main(dataset_size, context_size, skip_datagen=True)
+    with open("./tmp/t2t_qa_ds_2023_05_17.json") as f:
+        t2t_qa_ds = json.load(f)
+
+    from langchain.evaluation.qa import QAEvalChain
+
+    chat_llm = ChatOpenAI(temperature=0)
+    qa_chain = RetrievalQA.from_chain_type(
+        llm=chat_llm,
+        chain_type="stuff",
+        retriever=retriever,
+        input_key="question",
+        return_source_documents=False,
+    )
+    qa_eval_chain = QAEvalChain.from_llm(chat_llm)
+
+    t2t_qa_ds_subsampled = random.sample(t2t_qa_ds, k=100)
+    results = []
+    for example in tqdm(t2t_qa_ds_subsampled):
+        answer = qa_chain(example["question"])
+        ground_truth = {"question": example["question"], "answer": example["answer"]}
+        prediction = {"result": answer}
+        res = qa_eval_chain.evaluate(
+            [ground_truth], [prediction], question_key="question"
+        )[0]
+        results.append({**ground_truth, **prediction, "evaluation": res["text"]})
+
+    results_correct = sum(
+        result["evaluation"].lower() == "correct" for result in results
+    )
+    print(f"{results_correct=} {len(results)=}")

langchain/experimental/finetune/retrieval_qa/download_transcripts.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+set -ex
+
+url="https://karpathy.ai/lexicap/data.zip"
+filename="data.zip"
+directory="transcripts"
+
+mkdir -p $directory
+curl -o $filename $url
+tar -xf $filename -C $directory
+rm $filename

langchain/experimental/finetune/retrieval_qa/next.txt

@@ -0,0 +1,25 @@
+current:
+- finetune small, easy to operationalize instruction following LM for QA using QAGenerationChain
+- some observations on flan-t5-base, finetuned on ~1000 samples:
+    - dataset:
+        - wrt size, bumping 250->500->1000 samples yield proportional gains in performance
+    - with and without peft/lora
+        - no glaring difference at first glance, tho can easily finetune on collab free
+        - that said, didn't study regression (eg ppl) with full fine-tuning
+    - evaluation
+        - QAEvalChain
+            - have to dig further, sometimes just wrong, sometimes could benefit from more granularity/rubric-style
+        - precision (bleu), recall (rouge) style metrics
+            - quite a bit better on finetune
+        - manual
+            - it's decent, basically seems feasible to adapt a small flan in the style of the gpt-3.5-turbo-synthesized answers
+    
+next:
+- run fine-tune in RetrievalQAChain to do a better side-by-side
+    - there's no HuggingFaceModel?
+    - lora/peft merge
+- some more error analysis, some more scale in data and maybe model (esp. for more realistic retrieval setting, larger chunk size, etc.)
+
+later/not important just noting:
+- LLMChain doesn't expose raw prompt (and in turn any chain wrapping an LLMChain too), so I have some stateful hackery
+- QAGenerationChain is slow => chat completions api has no parallel proc support

langchain/experimental/finetune/retrieval_qa/tmp/rejected_qagen_questions.txt

@@ -0,0 +1,352 @@
+What is the speaker's opinion on acknowledging someone else while doing the thing?
+What is the speaker's opinion on relying on machines to do work?
+What does the speaker think about the possibility of manipulating gravity in our universe?
+What are some of the challenges associated with proof of work mining?
+What does the speaker hope for regarding the actions of leaders in response to COVID?
+What is the topic of the text?
+What is the speaker's opinion on the computational universe?
+What does the author think about death?
+What is the name of the person being interviewed in the text?
+What is one of the things discussed in Deep Work that is important?
+What is one thing we know for sure about the brain?
+What did the speaker learn from watching a video on neural networks?
+What was the original plan for the book mentioned in the text?
+What did the speaker find out by looking at the data?
+What kind of player is he?
+What is the era that the speaker refers to as we are entering?
+What approach did the author's team use for language understanding?
+What is the benefit of automated mechanism design?
+What is the block size debate about?
+What does the speaker compare themselves to?
+What is the bitter lesson according to the text?
+What is the speaker trying to realize and live up to?
+What is the speaker's personal experience with thinking?
+What is the topic of the panel discussion that the speaker saw on YouTube?
+What is the main tension that the speaker is discussing?
+What was the speaker's initial reaction when they heard about the topic?
+What is the speaker referring to when they mention 'the ledger'?
+What is the trope mentioned in the text?
+What is the most important step in starting a business according to the speaker?
+What is the author analyzing in the given text?
+What is the most challenging part of conducting experiments according to the text?
+What was the challenge mentioned in the text?
+What is the beauty of the concept being discussed?
+What does the speaker think is more exciting in science?
+What is the neocortex and how is it structured?
+What is the name of the song the narrator was listening to?
+What does the speaker say about making an individual part perfect?
+What inspires people to come up with new technologies that make a difference?
+What does the speaker think will happen if self-driving vehicles don't happen?
+What does the speaker believe about intelligence?
+What is the difference between science and commerce according to TK?
+What does the speaker suggest is the reason for people's intrinsic goodness?
+What is the author's opinion on having world wars?
+What is the speaker's opinion on the evolution of the universe?
+What does the speaker think about predicting with all the data around today?
+What was one of the hardest things the speaker has ever done?
+What is the author's opinion on the power of culture?
+What is the topic of the podcast mentioned in the text?
+What is the topic of the conversation?
+What is the speaker's advice to the listener regarding their goal?
+What is the author's opinion on SpaceX?
+What is the speaker's opinion on predicting the future?
+What is the range of expertise among the physicists and mathematicians mentioned in the text?
+What does the speaker refer to as 'attachment' in Buddhism?
+What is the topic of discussion in the given text?
+What is the name of the finance app that presents the show?
+What is the speaker's attitude towards partnering with others?
+What is the speaker's opinion of the person who brought evil onto the world?
+What is the basic task of humans according to the text?
+What is the concept of ZK snarks?
+What is the topic of discussion in the given text?
+What event were the speaker and John doing together?
+What is the topic of discussion in the given text?
+What does the speaker think about the scenarios where robots are tethered to something that makes them our slaves?
+What is the author discussing in this passage?
+What is the speaker's opinion on the importance of an interactive component in teaching?
+What is the analogy used to describe Bitcoin in the text?
+What is the Patrushev project that is mentioned in the text?
+What is the speaker's opinion on sex robots?
+What is the human mind doing while a person is driving?
+What is the speaker's opinion on the violence that occurred during wars in the past?
+What advice does the speaker give to avoid boredom?
+What is Novid and what kind of work have they been doing?
+What did the speaker learn about during the conversation?
+What is the theory being discussed in the text?
+What question does the speaker ask the author?
+What are some benefits of using a faster network?
+What is the speaker's advice for finding self-love?
+What is the West and how is it defined?
+What is the middle camp's view on consciousness and information processing?
+What happened to the virus over time?
+What is Numba and why was it created?
+What is the speaker's favorite set of games?
+What is the ethical question posed by the author?
+What topic is the speaker interested in after buying three books on boredom?
+What does the speaker mean when they say they are 'playing up a side of yourself'?
+What is the daily routine that the speaker follows?
+What is the speaker's vision for using AI in language translation?
+What is the author's opinion about the possibility of aliens existing?
+What is the quote that the speaker keeps on their phone?
+What is the question that the speaker asks the listener?
+What is the remarkable achievement mentioned in the text?
+What are the names of the sponsors mentioned in the podcast?
+What is an example of an interaction where AI should not have a conversation with you?
+What does first principles thinking require?
+What is the author's opinion on the current understanding of the human body?
+What is the author's opinion on the use of the terms 'machine learning' and 'deep learning'?
+What is the magic of deep learning?
+What does the speaker think about expertise?
+What does the author think about the ability of a group to make ethical decisions?
+What is the author describing in terms of their experience with the Olympic games?
+What is the benefit of bypassing the image processor?
+What is the comparison made between becoming a great comic and Jiu Jitsu?
+What does the author believe about the idea that technology is values agnostic?
+What is the speaker's motivation for working on AI?
+What is the name of the book that the speaker mentions as one of their favorites?
+Why is experiencing wins and losses important?
+What is the potential danger of people who think it's the end times?
+What does the speaker think about the idea of living forever?
+What did the speaker learn from the person mentioned in the text?
+What is the author's opinion on teaching math?
+What is the topic of discussion in this text?
+What is one of the things that is important to people on a day to day basis?
+What is the speaker going to argue against?
+What is the author's opinion on the importance of captivation?
+What is the main concern regarding the use of data mentioned in the text?
+What feature did they recently launch and why?
+What is the neural network being compared to in this text?
+What is the speaker's main concern when it comes to their job?
+What is the initial focus of the project?
+Who is the author dedicating their book to?
+What is the poker community's opinion of the player being discussed?
+What did the speaker switch to from experimental physics?
+What does the author think it takes to make the world a better place?
+What is the speaker's background in computer science?
+What is the speaker's opinion about the person they are talking to?
+What is the speaker's opinion on discovering joy?
+What is the speaker's opinion on journals?
+What is the speaker's opinion on making major irreversible life decisions?
+What company is being discussed in the text?
+What is the author's opinion on the current systems?
+What is the author's opinion on the potential of Librex?
+What did David give to the speaker when they visited his lab at MIT?
+What is the bigger problem in capitalism according to the speaker?
+What is the author suggesting we should do more of?
+What was the mechanism that helped the speaker start listening to their body?
+What is the phenomenon that the speaker is describing?
+What was the conversation the speaker was calling for?
+What did the speaker work on in the 1980s?
+What is the brain doing when it makes predictions?
+What is the speaker's concern regarding the current situation?
+What are some ways to support the podcast mentioned in the text?
+What is the speaker's opinion on the program's information?
+What is the speaker's concern about the system?
+What odds did the speaker put on P equaling NP?
+What is the conversation the speaker had with Dan Carlin?
+What does the speaker find fascinating?
+What is the main difference between this game and chess or turn-based strategy games?
+What is the speaker's theory related to quantum?
+What is the breakthrough prize awarded for?
+What is the significance of the people mentioned in the text?
+What is the network sometimes called that represents the spread of a virus?
+What is the example given in the text of a scientist who was trying to understand a complex problem for months?
+What does the speaker mean when they say 'putting a new cartridge in your brain'?
+What is quantum field theory?
+What areas of research require massive funding according to the text?
+What is becoming more common these days when a person passes away?
+What is the suggested method for efficiently transmitting information across the network?
+What did the woman ask the author when she expressed interest in his book?
+What is the system that the speaker is referring to and how has it been trained?
+What analogy does the speaker use to describe the process of consciousness being generated?
+What is the speaker's opinion on using only neural networks for problem-solving?
+What is the 'alignment problem' referred to in the text?
+What is one of the things that is important to people on a day to day basis?
+What is the text discussing?
+What did Trump tap into according to the text?
+What is the message that people have been hearing over and over throughout human history?
+What is the David Goggins model?
+What is neuromorphic computing?
+What are some of the books mentioned in the text?
+What are the problems that you have to attack?
+What is an example of a problem in NP?
+What is the joke that the speaker and their friend share?
+What are the two psychological reasons mentioned by the speaker?
+What is the name of the podcast mentioned in the text?
+What does the speaker love about disagreements?
+What is the author's question regarding the vaccine and their mind?
+What is the speaker's experience with sauna?
+What is the author disappointed by in the community?
+What is the concept described in the text?
+What is the paradigm that was tested in a very interesting way?
+What is the speaker's opinion on the possibility of aliens showing up?
+What is the task that may not seem trivial?
+What is the speaker's opinion about the task at hand?
+What are some advantages of using Ethereum instead of Bitcoin?
+What is the author questioning the possibility of?
+What is the speaker's opinion on the interaction between machines and humans?
+What is the speaker pondering about in regards to a petabyte of memory?
+What does the speaker believe is necessary for longterm growth?
+What is the importance of consciousness according to the text?
+What is the trade off being discussed in this text?
+What is the dilemma that the speaker is referring to?
+What is the purpose of the dojo system?
+What is the analogy used to describe emergent patterns?
+What is the music video that the speaker mentions in the text?
+What advice does the speaker ask for from the person being addressed?
+What is the challenge mentioned in the text?
+What is the author's opinion on the concept of life being short?
+What did the speaker learn from the language they were using?
+What is the set theoretic hierarchy?
+What is the big discovery of the principle mentioned in the text?
+What is the libertarian perspective on government and military?
+Why does the speaker feel the need to be sharp when acting on screen?
+What is Edward?
+What is the question in the space of all possible intelligences?
+What is the author discussing in this passage?
+What does the speaker credit for their success?
+What is the topic of discussion in the text?
+What is the main problem the speaker is facing?
+What was Dick Feynman's opinion on quantum mechanics?
+What is the author disappointed in regarding expertise in science?
+What is the speaker's opinion on the Holocaust?
+What is an organoid and can it mimic different regions of the brain?
+What are the two deeply human aspects mentioned in the text?
+What are the two fields that the speaker particularly focuses on?
+What is the speaker's opinion on Bitcoin?
+What is the author's point about our dependence on supply chains?
+What is the author discussing in this text?
+What is the topic of discussion in this text?
+What is the purpose of the author's writing style in 'just my stuff'?
+What have they been talking about in regards to robots?
+What is the town square test and who proposed it?
+What is the fallacy that the speaker talks about in the given text?
+What experimental test did the speaker propose based on the professor's statement?
+What is the theme of exponential growth mentioned in the text?
+What is the author discussing in this passage?
+What was the author's initial area of study?
+What is the idea behind the Trojan horse allegory mentioned in the text?
+What is the ultimate AI problem discussed in the text?
+What is the speaker doing while looking at the monitor?
+What does the author love?
+What is the speaker's reason for not going to Burning Man?
+What is the author's point of view on the idea of unification?
+What are the two main problems that need to be solved in robotics?
+What is the goal of the process mentioned in the text?
+What does the speaker say about the issues being discussed?
+What does the speaker like doing?
+Why shouldn't you go to bed any earlier?
+What is the human body compared to in the text?
+What is the comparison made between the trailer drivers and Uber drivers?
+What is the game that the speaker is referring to?
+What is the benefit of a process where it's good for the individual and good for the whole?
+What is the speaker's opinion on the meaning of life?
+What did the author fear the most about COVID?
+What is the difference between humans and non-human animals?
+What is the author's opinion on the potential for exponential growth in technology?
+What is the author discussing in this passage?
+What was the author's initial idea about programming languages?
+What are some possible explanations for why we haven't heard from extraterrestrial life?
+What is the 20 minute challenge?
+What is the author's opinion on the idea that machines can't have experience?
+What is the topic of the text?
+What is the speaker discussing in regards to numbers?
+What was the goal of the research mentioned in the text?
+What is the speaker's view of Russia?
+What is the benefit of using the code LEX PODCAST when downloading Cash App?
+Who did the speaker text before going on Tucker's show?
+What does the speaker urge people not to get into?
+What is the most compelling question to the speaker?
+What is the conversation the speaker had with Dan Carlin?
+What organization will receive a donation if you use the code LEXPODCAST on Cash App?
+What is the innate desire that humans have according to the text?
+What was the name of the first robot the speaker worked with?
+What is the speaker's opinion on the size of the list?
+What is the ethical question posed by the author?
+What does the author love about fighting?
+Who did the speaker and their partner go to for therapy?
+What does the author believe about the idea that technology is values agnostic?
+What is the beauty of music according to the speaker?
+What advice did the speaker give to students?
+What is the scientist's goal in sharing information about the science of sleep?
+What is the trend that the speaker sees when looking back at history?
+What is the speaker's opinion on the argument being discussed?
+What was the name of the popular forum that the speaker used to ask questions on?
+What is the topic of discussion in the given text?
+What technology does the speaker rely on for intuition?
+What was the speaker's profession and where did they work?
+What is the speaker's background that gives them an advantage as an ethnographer?
+What is the problem with the Olympics according to the text?
+What is the speaker excited about in the future?
+What topics are currently in the realm of philosophy?
+What is the big discovery of the principle mentioned in the text?
+What is the topic of the panel discussion that the speaker saw on YouTube?
+What is the best way to learn how to do something?
+What book was the speaker listening to while running?
+What is the speaker's opinion on building an AI that can detect nuanced language in different languages?
+What is the anonymous discussion feed for Yale campus?
+What is the speaker's opinion on the dark aspects of human nature?
+What are the names of the two startups mentioned in the text?
+What does the speaker do to expand their brain's capacity to predict more effectively?
+What is the book about that is being discussed?
+What is the topic of discussion in the given text?
+What game did the speaker hate playing in high school?
+What does the author think about the use of social media?
+What is the author discussing in this text?
+What is radar jamming?
+What is the author's perspective on human history?
+What is the speaker's opinion on human level intelligence?
+What is the speaker's concern about machine learning?
+What types of things can be found on Wikipedia according to the text?
+What was the name of the popular forum that the speaker used to ask questions on?
+What is the significance of the Winter War mentioned in the text?
+What is the author trying to say about our experiences?
+What is the author's opinion on the potential of Librex?
+What is the 'lockdown harm' referred to in the text?
+What is the open discussion that the author suggests we are still having?
+What is the advantage of technology according to the text?
+What is Dan Gable known for?
+What did the researchers discover during phase two of MAPS?
+What is the topic of discussion in the given text?
+What is the speaker's opinion on alien technology?
+What is the author's opinion on traveling faster than the speed of light?
+What is the topic of discussion in the given text?
+What is the benefit of using the code LEXPODCAST when downloading Cash App?
+What is the difference between when the speaker was talking to people after Deep Work and now, five years later?
+What does the author mean by 'skin in the game'?
+What is the speaker's initial fascination with math and physics?
+What advice would George give to a young person about life?
+According to the panpsychist, what is all there is?
+What is the focus of volume one of the book?
+What is the suggestion made by the speaker to solve the email problem?
+What is the speaker's opinion about the potential for their product to be useful?
+What are some possible explanations for why we haven't heard a big, loud signal from other intelligent life forms?
+What is the speaker referring to as an 'intractable problem'?
+What was the important thing to take into account when considering the size of the armies?
+What is the speaker's opinion on the importance of understanding a person's thinking and actions?
+What is the speaker's question regarding understanding something?
+What is the author's opinion on the ability of neural networks?
+What is the problem with the COVID virus?
+What is the speaker's profession or occupation?
+What are the names of the sponsors mentioned in the conversation?
+What is the purpose of science according to the speaker?
+What is one thing that the speaker mentions is missing in Alexa?
+What was the author studying and why?
+What was one of the really stressful parts of the situation?
+What is the speaker's opinion on Bitcoin as a mechanism for resisting authoritarianism?
+What is the speaker's fundamental belief about people?
+What is the example given in the text to explain a distributed algorithm?
+What kind of problems should you pick according to the speaker?
+What is the open scientific problem that the speaker is referring to?
+What is the author's opinion on the unexpected result?
+What was the speaker's reaction to the object they saw?
+What is ultimately ending COVID?
+What is the benefit of using ExpressVPN?
+What is the author's opinion on integrating compound semiconductors with silicon?
+What is the speaker going to talk about?
+What is the concept of complex exponentiation trying to capture?
+What was the reason for the speaker's need for medication?
+What does the author wonder about physicists and their ability to understand a phenomenon?
+What is the author's opinion on the trend of history?
+What is the difference between closed end problems and reading?
+What are the two things involved in creating a hypergraph?

langchain/vectorstores/base.py

@@ -330,7 +330,7 @@ class VectorStore(ABC):
         raise NotImplementedError
 
     def as_retriever(self, **kwargs: Any) -> BaseRetriever:
-        return VectorStoreRetriever(vectorstore=self, **kwargs)
+        return VectorStoreRetriever(vectorstore=self, search_kwargs=kwargs)
 
 
 class VectorStoreRetriever(BaseRetriever, BaseModel):