diff --git a/Makefile b/Makefile new file mode 100644 index 00000000..eabd8090 --- /dev/null +++ b/Makefile @@ -0,0 +1,24 @@ +install-vercel-deps: + yum -y update + yum install gcc bzip2-devel libffi-devel zlib-devel wget tar gzip rsync -y + +PYTHON = .venv/bin/python + +build-api-ref: + git clone --depth=1 https://github.com/langchain-ai/langsmith-sdk.git + python3 -m venv .venv + . .venv/bin/activate + $(PYTHON) -m pip install --upgrade pip + $(PYTHON) -m pip install --upgrade uv + cd langsmith-sdk && ../$(PYTHON) -m uv pip install -r python/docs/requirements.txt + $(PYTHON) langsmith-sdk/python/docs/create_api_rst.py + LC_ALL=C $(PYTHON) -m sphinx -T -E -b html -d langsmith-sdk/python/docs/_build/doctrees -c langsmith-sdk/python/docs langsmith-sdk/python/docs langsmith-sdk/python/docs/_build/html -j auto + $(PYTHON) langsmith-sdk/python/docs/scripts/custom_formatter.py langsmith-sdk/docs/_build/html/ + + +vercel-build: install-vercel-deps build-api-ref + mkdir -p static/reference/python + mv langsmith-sdk/python/docs/_build/html/* static/reference/python/ + rm -rf langsmith-sdk + NODE_OPTIONS="--max-old-space-size=5000" yarn run docusaurus build + diff --git a/docs/evaluation/how_to_guides/async.mdx b/docs/evaluation/how_to_guides/async.mdx index facb4b81..9c35a0c4 100644 --- a/docs/evaluation/how_to_guides/async.mdx +++ b/docs/evaluation/how_to_guides/async.mdx @@ -8,8 +8,8 @@ import { CodeTabs, python } from "@site/src/components/InstructionsWithCode"; ::: -We can run evaluations asynchronously via the SDK using [aevaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._arunner.aevaluate.html), -which accepts all of the same arguments as [evaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._runner.evaluate.html) but expects the application function to be asynchronous. +We can run evaluations asynchronously via the SDK using [aevaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._arunner.aevaluate), +which accepts all of the same arguments as [evaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._runner.evaluate) but expects the application function to be asynchronous. You can learn more about how to use the `evaluate()` function [here](./evaluate_llm_application). :::info Python only @@ -25,8 +25,8 @@ You can see how to use it [here](./evaluate_llm_application). =0.1.145`"})` - from langsmith import aevaluate, wrappers, Client + python({caption: "Requires `langsmith>=0.2.0`"})` + from langsmith import wrappers, Client from openai import AsyncOpenAI # Optionally wrap the OpenAI client to trace all model calls. @@ -61,12 +61,15 @@ list 5 concrete questions that should be investigated to determine if the idea i inputs=[{"idea": e} for e in examples, ) - results = await aevaluate( + # Can equivalently use the 'aevaluate' function directly: + # from langsmith import aevaluate + # await aevaluate(...) + results = await ls_client.aevaluate( researcher_app, data=dataset, evaluators=[concise], - # Optional, no max_concurrency by default but it is recommended to set one. - max_concurrency=2, + # Optional, add concurrency. + max_concurrency=2, # Optional, add concurrency. experiment_prefix="gpt-4o-mini-baseline" # Optional, random by default. ) `, diff --git a/docs/evaluation/how_to_guides/create_few_shot_evaluators.mdx b/docs/evaluation/how_to_guides/create_few_shot_evaluators.mdx index 4bf8f696..ecdbbb78 100644 --- a/docs/evaluation/how_to_guides/create_few_shot_evaluators.mdx +++ b/docs/evaluation/how_to_guides/create_few_shot_evaluators.mdx @@ -2,7 +2,7 @@ sidebar_position: 10 --- -How to create few-shot evaluators +# How to create few-shot evaluators Using LLM-as-a-Judge evaluators can be very helpful when you can't evaluate your system programmatically. However, improving/iterating on these prompts can add unnecessary overhead to the development process of an LLM-based application - you now need to maintain both your application **and** your evaluators. To make this process easier, LangSmith allows diff --git a/docs/evaluation/how_to_guides/custom_evaluator.mdx b/docs/evaluation/how_to_guides/custom_evaluator.mdx index 93ac7ef0..de567cae 100644 --- a/docs/evaluation/how_to_guides/custom_evaluator.mdx +++ b/docs/evaluation/how_to_guides/custom_evaluator.mdx @@ -13,14 +13,14 @@ import { ::: Custom evaluators are just functions that take a dataset example and the resulting application output, and return one or more metrics. -These functions can be passed directly into [evaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._runner.evaluate.html) / [aevaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._arunner.aevaluate.html). +These functions can be passed directly into [evaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._runner.evaluate) / [aevaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._arunner.aevaluate). ## Basic example =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` from langsmith import evaluate def correct(outputs: dict, reference_outputs: dict) -> bool: @@ -36,12 +36,14 @@ These functions can be passed directly into [evaluate()](https://langsmith-sdk.r evaluators=[correct] ) `, - typescript` + typescript({caption: "Requires `langsmith>=0.2.9`"})` import type { EvaluationResult } from "langsmith/evaluation"; - import type { Run, Example } from "langsmith/schemas"; - function correct(run: Run, example: Example): EvaluationResult { - const score = run.outputs?.output === example.outputs?.output; + const correct = async ({ outputs, referenceOutputs }: { + outputs: Record; + referenceOutputs?: Record; + }): Promise => { + const score = outputs?.answer === referenceOutputs?.answer; return { key: "correct", score }; } `, @@ -53,19 +55,16 @@ These functions can be passed directly into [evaluate()](https://langsmith-sdk.r Custom evaluator functions must have specific argument names. They can take any subset of the following arguments: -Python and JS/TS - -- `run: langsmith.schemas.Run`: The full Run object generated by the application on the given example. -- `example: langsmith.schemas.Example`: The full dataset Example, including the example inputs, outputs (if available), and metdata (if available). - -Currently Python only - +- `run: Run`: The full [Run](/reference/data_formats/run_data_format) object generated by the application on the given example. +- `example: Example`: The full dataset [Example](/reference/data_formats/example_data_format), including the example inputs, outputs (if available), and metdata (if available). - `inputs: dict`: A dictionary of the inputs corresponding to a single example in a dataset. - `outputs: dict`: A dictionary of the outputs generated by the application on the given `inputs`. -- `reference_outputs: dict`: A dictionary of the reference outputs associated with the example, if available. +- `reference_outputs/referenceOutputs: dict`: A dictionary of the reference outputs associated with the example, if available. For most use cases you'll only need `inputs`, `outputs`, and `reference_outputs`. `run` and `example` are useful only if you need some extra trace or example metadata outside of the actual inputs and outputs of the application. +When using JS/TS these should all be passed in as part of a single object argument. + ## Evaluator output Custom evaluators are expected to return one of the following types: @@ -85,16 +84,17 @@ Currently Python only =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` from langsmith import evaluate, wrappers + from langsmith.schemas import Run, Example from openai import AsyncOpenAI # Assumes you've installed pydantic. from pydantic import BaseModel - # Compare actual and reference outputs - def correct(outputs: dict, reference_outputs: dict) -> bool: + # We can still pass in Run and Example objects if we'd like + def correct_old_signature(run: Run, example: Example) -> dict: """Check if the answer exactly matches the expected answer.""" - return outputs["answer"] == reference_outputs["answer"] + return {"key": "correct", "score": run.outputs["answer"] == example.outputs["answer"]} # Just evaluate actual outputs def concision(outputs: dict) -> int: @@ -129,9 +129,105 @@ answer is logically valid and consistent with question and the answer.""" results = evaluate( dummy_app, data="dataset_name", - evaluators=[correct, concision, valid_reasoning] + evaluators=[correct_old_signature, concision, valid_reasoning] ) `, + typescript` + import { Client } from "langsmith"; + import { evaluate } from "langsmith/evaluation"; + import { Run, Example } from "langsmith/schemas"; + import OpenAI from "openai"; + + // Type definitions + interface AppInputs { + question: string; + } + + interface AppOutputs { + answer: string; + reasoning: string; + } + + interface Response { + reasoning_is_valid: boolean; + } + + // Old signature evaluator + function correctOldSignature(run: Run, example: Example) { + return { + key: "correct", + score: run.outputs?.["answer"] === example.outputs?.["answer"], + }; + } + + // Output-only evaluator + function concision({ outputs }: { outputs: AppOutputs }) { + return { + key: "concision", + score: Math.min(Math.floor(outputs.answer.length / 1000), 4) + 1, + }; + } + + // LLM-as-judge evaluator + const openai = new OpenAI(); + + async function validReasoning({ + inputs, + outputs + }: { + inputs: AppInputs; + outputs: AppOutputs; + }) { + const instructions = \`\ + Given the following question, answer, and reasoning, determine if the reasoning for the \ + answer is logically valid and consistent with question and the answer.\`; + + const msg = \`Question: \${inputs.question}\nAnswer: \${outputs.answer}\\nReasoning: \${outputs.reasoning}\`; + + const response = await openai.chat.completions.create({ + model: "gpt-4", + messages: [ + { role: "system", content: instructions }, + { role: "user", content: msg } + ], + response_format: { type: "json_object" }, + functions: [{ + name: "parse_response", + parameters: { + type: "object", + properties: { + reasoning_is_valid: { + type: "boolean", + description: "Whether the reasoning is valid" + } + }, + required: ["reasoning_is_valid"] + } + }] + }); + + const parsed = JSON.parse(response.choices[0].message.content ?? "{}") as Response; + + return { + key: "valid_reasoning", + score: parsed.reasoning_is_valid ? 1 : 0 + }; + } + + // Example application + function dummyApp(inputs: AppInputs): AppOutputs { + return { + answer: "hmm i'm not sure", + reasoning: "i didn't understand the question" + }; + } + + const results = await evaluate(dummyApp, { + data: "dataset_name", + evaluators: [correctOldSignature, concision, validReasoning], + client: new Client() + }); + ` ]} /> diff --git a/docs/evaluation/how_to_guides/dataset_subset.mdx b/docs/evaluation/how_to_guides/dataset_subset.mdx index efc914c9..8aa2babf 100644 --- a/docs/evaluation/how_to_guides/dataset_subset.mdx +++ b/docs/evaluation/how_to_guides/dataset_subset.mdx @@ -85,4 +85,4 @@ You can use the `list_examples` / `listExamples` method to evaluate on one or mu ## Related -- More on [how to filter datasets](./manage_datasets_programmatically#list-examples-by-structured-filter) +- Learn more about how to fetch views of a dataset [here](./manage_datasets_programmatically#fetch-datasets) diff --git a/docs/evaluation/how_to_guides/dataset_version.mdx b/docs/evaluation/how_to_guides/dataset_version.mdx index 564c1295..791b879a 100644 --- a/docs/evaluation/how_to_guides/dataset_version.mdx +++ b/docs/evaluation/how_to_guides/dataset_version.mdx @@ -13,35 +13,53 @@ Additionally, it might be helpful to read the [guide on fetching examples](./man ::: -You can take advantage of the fact that `evaluate` allows passing in an iterable of examples to evaluate on a particular version of a dataset. -Simply use `list_examples` / `listExamples` to fetch examples from a particular version tag using `as_of` / `asOf`. +## Using `list_examples` + +You can take advantage of the fact that `evaluate` / `aevaluate` allows passing in an iterable of examples to evaluate on a particular version of a dataset. +Simply use `list_examples` / `listExamples` to fetch examples from a particular version tag using `as_of` / `asOf` and pass that in to the `data` argument. bool: + return outputs["class"] == reference_outputs["label"] + + results = ls_client.evaluate( + lambda inputs: {"class": "Not toxic"}, + # Pass in filtered data here: + # highlight-next-line + data=ls_client.list_examples( + # highlight-next-line + dataset_name="Toxic Queries", + # highlight-next-line + as_of="latest", # specify version here + # highlight-next-line + ), + evaluators=[correct], ) `, typescript` import { evaluate } from "langsmith/evaluation"; - + await evaluate((inputs) => labelText(inputs["input"]), { data: langsmith.listExamples({ datasetName: datasetName, asOf: "latest", }), evaluators: [correctLabel], - experimentPrefix: "Toxic Queries", }); `, - ]} + +]} /> + +## Related + +- Learn more about how to fetch views of a dataset [here](./manage_datasets_programmatically#fetch-datasets) diff --git a/docs/evaluation/how_to_guides/evaluate_existing_experiment.mdx b/docs/evaluation/how_to_guides/evaluate_existing_experiment.mdx index 7c7ac33f..489aa6c2 100644 --- a/docs/evaluation/how_to_guides/evaluate_existing_experiment.mdx +++ b/docs/evaluation/how_to_guides/evaluate_existing_experiment.mdx @@ -4,60 +4,20 @@ sidebar_position: 6 # How to evaluate an existing experiment (Python only) -:::note -Currently, `evaluate_existing` is only supported in the Python SDK. +:::inof +Evaluation of existing experiments is currently only supported in the Python SDK. ::: If you have already run an experiment and want to add additional evaluation metrics, you -can apply any evaluators to the experiment using the `evaluate_existing` method. - -```python -from langsmith import evaluate_existing - -def always_half(run, example): - return {"score": 0.5} - -experiment_name = "my-experiment:abcd123" # Replace with an actual experiment name or ID -evaluate_existing(experiment_name, evaluators=[always_half]) -``` - -## Example - -Suppose you are evaluating a semantic router. You may first run an experiment: +can apply any evaluators to the experiment using the `evaluate()` / `aevaluate()` methods as before. +Just pass in the experiment name / ID instead of a target function: ```python from langsmith import evaluate -def semantic_router(inputs: dict): - return {"class": 1} - -def accuracy(run, example): - prediction = run.outputs["class"] - expected = example.outputs["label"] - return {"score": prediction == expected} -results = evaluate(semantic_router, data="Router Classification Dataset", evaluators=[accuracy]) -experiment_name = results.experiment_name -``` - -Later, you realize you want to add precision and recall summary metrics. The `evaluate_existing` method accepts the same arguments as the `evaluate` method, replacing the `target` system with the `experiment` you wish to add metrics to, meaning -you can add both instance-level `evaluator`'s and aggregate `summary_evaluator`'s. - -```python -from langsmith import evaluate_existing +def always_half(inputs: dict, outputs: dict) -> float: + return 0.5 -def precision(runs: list, examples: list): - true_positives = sum([1 for run, example in zip(runs, examples) if run.outputs["class"] == example.outputs["label"]]) - false_positives = sum([1 for run, example in zip(runs, examples) if run.outputs["class"] != example.outputs["label"]]) - return {"score": true_positives / (true_positives + false_positives)} - -def recall(runs: list, examples: list): - true_positives = sum([1 for run, example in zip(runs, examples) if run.outputs["class"] == example.outputs["label"]]) - false_negatives = sum([1 for run, example in zip(runs, examples) if run.outputs["class"] != example.outputs["label"]]) - return {"score": true_positives / (true_positives + false_negatives)} - -evaluate_existing(experiment_name, summary_evaluators=[precision, recall]) +experiment_name = "my-experiment:abc" # Replace with an actual experiment name or ID +evaluate(experiment_name, evaluators=[always_half]) ``` - -The precision and recall metrics will now be available in the LangSmith UI for the `experiment_name` experiment. - -As is the case with the `evaluate` function, there is an identical, asynchronous `aevaluate_existing` function that can be used to evaluate experiments asynchronously. diff --git a/docs/evaluation/how_to_guides/evaluate_llm_application.mdx b/docs/evaluation/how_to_guides/evaluate_llm_application.mdx index 84a8cff5..e8ac1937 100644 --- a/docs/evaluation/how_to_guides/evaluate_llm_application.mdx +++ b/docs/evaluation/how_to_guides/evaluate_llm_application.mdx @@ -16,13 +16,17 @@ import { ::: -In this guide we'll go over how to evaluate an application using the [evaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._runner.evaluate.html) method in the LangSmith SDK. +In this guide we'll go over how to evaluate an application using the [evaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._runner.evaluate) method in the LangSmith SDK. -:::tip +:::tip Running large jobs -For larger evaluation jobs in Python we recommend using [aevaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._arunner.aevaluate.html), the asynchronous version of `evaluate()`. -It is still worthwhile to read this guide first, as the two have nearly identical interfaces, -and then read the how-to guide on [running an evaluation asynchronously](./async). +For larger evaluation jobs in Python we recommend using [aevaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._arunner.aevaluate), the asynchronous version of [evaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._runner.evaluate). +It is still worthwhile to read this guide first, as the two have identical interfaces, before reading the how-to guide on [running an evaluation asynchronously](./async). + +In JS/TS evaluate() is already asynchronous so no separate method is needed. + +It is also important to configure the `max_concurrency`/`maxConcurrency` arg when running large jobs. +This parallelizes evaluation by effectively splitting the dataset across threads. ::: @@ -42,7 +46,7 @@ First we need an application to evaluate. Let's create a simple toxicity classif # Optionally add the 'traceable' decorator to trace the inputs/outputs of this function. @traceable - def toxicity_classifier(inputs: dict) -> str: + def toxicity_classifier(inputs: dict) -> dict: instructions = ( "Please review the user query below and determine if it contains any form of toxic behavior, " "such as insults, threats, or highly negative comments. Respond with 'Toxic' if it does " @@ -55,7 +59,7 @@ First we need an application to evaluate. Let's create a simple toxicity classif result = oai_client.chat.completions.create( messages=messages, model="gpt-4o-mini", temperature=0 ) - return result.choices[0].message.content + return {"class": result.choices[0].message.content} `, typescript` import { OpenAI } from "openai"; @@ -113,10 +117,11 @@ We need a [Dataset](../concepts#datasets) to evaluate our application on. Our da dataset_name = "Toxic Queries" dataset = ls_client.create_dataset(dataset_name=dataset_name) - inputs, outputs = zip( - *[({"text": text}, {"label": label}) for text, label in labeled_texts] + ls_client.create_examples( + inputs=[{"text": text} for text, _ in labeled_texts], + outputs=[{"label": label} for _, label in labeled_texts], + dataset_id=dataset.id, ) - ls_client.create_examples(inputs=inputs, outputs=outputs, dataset_id=dataset.id) `, typescript` import { Client } from "langsmith"; @@ -160,16 +165,21 @@ Since we have labels for this task, our evaluator can directly check if the actu =0.1.145`" })` + python({ caption: "Requires `langsmith>=0.2.0`" })` def correct(inputs: dict, outputs: dict, reference_outputs: dict) -> bool: - return outputs["output"] == reference_outputs["label"] + return outputs["class"] == reference_outputs["label"] `, - typescript` + typescript({ caption: "Requires `langsmith>=0.2.9`" })` import type { EvaluationResult } from "langsmith/evaluation"; - import type { Run, Example } from "langsmith/schemas"; - function correct(rootRun: Run, example: Example): EvaluationResult { - const score = rootRun.outputs?.output === example.outputs?.outputs; + function correct({ + outputs, + referenceOutputs, + }: { + outputs: Record; + referenceOutputs?: Record; + }): EvaluationResult { + const score = outputs.output === referenceOutputs?.outputs; return { key: "correct", score }; } `, @@ -180,26 +190,27 @@ See [here](.#define-an-evaluator) for more on how to define evaluators. ## Run the evaluation -We'll use the [evaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._runner.evaluate.html) / [aevaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._arunner.aevaluate.html) methods to run the evaluation. +We'll use the [evaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._runner.evaluate) / [aevaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._arunner.aevaluate) methods to run the evaluation. The key arguments are: -- a function that takes an input dictionary and returns an output dictionary or object +- a target function that takes an input dictionary and returns an output dictionary. The `example.inputs` field of each [Example](/reference/data_formats/example_data_format) is what gets passed to the target function. In this case our `toxicity_classifier` is already set up to take in example inputs so we can use it directly. - `data` - the name OR UUID of the LangSmith dataset to evaluate on, or an iterator of examples - `evaluators` - a list of evaluators to score the outputs of the function =0.1.145`" })` - from langsmith import evaluate - - results = evaluate( + python({ caption: "Requires `langsmith>=0.2.0`" })` + # Can equivalently use the 'evaluate' function directly: + # from langsmith import evaluate; evaluate(...) + results = ls_client.evaluate( toxicity_classifier, data=dataset_name, evaluators=[correct], experiment_prefix="gpt-4o-mini, baseline", # optional, experiment name prefix description="Testing the baseline system.", # optional, experiment description + max_concurrency=4, # optional, add concurrency ) `, typescript` @@ -209,6 +220,7 @@ The key arguments are: data: datasetName, evaluators: [correct], experimentPrefix: "gpt-4o-mini, baseline", // optional, experiment name prefix + maxConcurrency: 4, // optional, add concurrency }); `, ]} @@ -232,8 +244,8 @@ _If you've annotated your code for tracing, you can open the trace of each row i =0.1.145`"})` - from langsmith import Client, evaluate, traceable, wrappers + python({caption: "Requires `langsmith>=0.2.0`"})` + from langsmith import Client, traceable, wrappers from openai import OpenAI # Step 1. Define an application @@ -269,22 +281,25 @@ _If you've annotated your code for tracing, you can open the trace of each row i dataset_name = "Toxic Queries" dataset = ls_client.create_dataset(dataset_name=dataset_name) - inputs, outputs = zip( - *[({"text": text}, {"label": label}) for text, label in labeled_texts] + ls_client.create_examples( + inputs=[{"text": text} for text, _ in labeled_texts], + outputs=[{"label": label} for _, label in labeled_texts], + dataset_id=dataset.id, ) - ls_client.create_examples(inputs=inputs, outputs=outputs, dataset_id=dataset.id) # Step 3. Define an evaluator def correct(inputs: dict, outputs: dict, reference_outputs: dict) -> bool: return outputs["output"] == reference_outputs["label"] # Step 4. Run the evaluation - results = evaluate( + # Client.evaluate() and evaluate() behave the same. + results = ls_client.evaluate( toxicity_classifier, data=dataset_name, evaluators=[correct], experiment_prefix="gpt-4o-mini, simple", # optional, experiment name prefix description="Testing the baseline system.", # optional, experiment description + max_concurrency=4, # optional, add concurrency ) `, typescript` @@ -344,8 +359,14 @@ _If you've annotated your code for tracing, you can open the trace of each row i await langsmith.createExamples({ inputs, outputs, datasetId: toxicDataset.id }); // Row-level evaluator - function correct(rootRun: Run, example: Example): EvaluationResult { - const score = rootRun.outputs?.output === example.outputs?.outputs; + function correct({ + outputs, + referenceOutputs, + }: { + outputs: Record; + referenceOutputs?: Record; + }): EvaluationResult { + const score = outputs.output === referenceOutputs?.outputs; return { key: "correct", score }; } @@ -353,7 +374,7 @@ _If you've annotated your code for tracing, you can open the trace of each row i data: datasetName, evaluators: [correct], experimentPrefix: "gpt-4o-mini, simple", // optional, experiment name prefix - + maxConcurrency: 4, // optional, add concurrency }); `, diff --git a/docs/evaluation/how_to_guides/evaluate_on_intermediate_steps.mdx b/docs/evaluation/how_to_guides/evaluate_on_intermediate_steps.mdx index a22f5df1..772c5976 100644 --- a/docs/evaluation/how_to_guides/evaluate_on_intermediate_steps.mdx +++ b/docs/evaluation/how_to_guides/evaluate_on_intermediate_steps.mdx @@ -10,7 +10,7 @@ import { typescript, } from "@site/src/components/InstructionsWithCode"; -# How to evaluate intermediate steps +# How to evaluate an application's intermediate steps While, in many scenarios, it is sufficient to evaluate the final output of your task, in some cases you might want to evaluate the intermediate steps of your pipeline. @@ -21,7 +21,7 @@ For example, for retrieval-augmented generation (RAG), you might want to In this guide, we will use a simple, fully-custom evaluator for evaluating criteria 1 and an LLM-based evaluator for evaluating criteria 2 to highlight both scenarios. -In order to evaluate the intermediate steps of your pipeline, your evaluator function should traverse and process the `root_run`/`rootRun` argument, which is a `Run` object that contains the intermediate steps of your pipeline. +In order to evaluate the intermediate steps of your pipeline, your evaluator function should traverse and process the `run`/`rootRun` argument, which is a `Run` object that contains the intermediate steps of your pipeline. ## 1. Define your LLM pipeline @@ -29,57 +29,97 @@ The below RAG pipeline consists of 1) generating a Wikipedia query given the inp = 2: - return results\n -@traceable -def generate_answer(question, context): - messages = [ - {"role": "system", "content": f"Answer the user's question based ONLY on the content below:\\n\\n{context}"}, - {"role": "user", "content": question} - ] - result = openai.chat.completions.create(messages=messages, model="gpt-4o-mini", temperature=0) - return result.choices[0].message.content\n -@traceable -def rag_pipeline(question): - query = generate_wiki_search(question) - context = "\\n\\n".join([doc["page_content"] for doc in retrieve(query)]) - answer = generate_answer(question, context) - return answer`), + { + value: "python", + label: "Python", + language: "bash", + content: `pip install -U langsmith langchain langchain-openai wikipedia`, + }, + { + value: "typescript", + label: "TypeScript", + language: "bash", + content: `yarn add langsmith langchain @langchain/openai wikipedia`, + }, + ]} + groupId="client-language" +/> + +=0.2.0`' })` + import wikipedia as wp + from openai import OpenAI + from langsmith import traceable, wrappers + + oai_client = wrappers.wrap_openai(OpenAI()) + + @traceable + def generate_wiki_search(question: str) -> str: + """Generate the query to search in wikipedia.""" + instructions = ( + "Generate a search query to pass into wikipedia to answer the user's question. " + "Return only the search query and nothing more. " + "This will passed in directly to the wikipedia search engine." + ) + messages = [ + {"role": "system", "content": instructions}, + {"role": "user", "content": question} + ] + result = oai_client.chat.completions.create( + messages=messages, + model="gpt-4o-mini", + temperature=0, + ) + return result.choices[0].message.content + + @traceable(run_type="retriever") + def retrieve(query: str) -> list: + """Get up to two search wikipedia results.""" + results = [] + for term in wp.search(query, results = 10): + try: + page = wp.page(term, auto_suggest=False) + results.append({ + "page_content": page.summary, + "type": "Document", + "metadata": {"url": page.url} + }) + except wp.DisambiguationError: + pass + if len(results) >= 2: + return results + + @traceable + def generate_answer(question: str, context: str) -> str: + """Answer the question based on the retrieved information.""" + instructions = f"Answer the user's question based ONLY on the content below:\\n\\n{context}" + messages = [ + {"role": "system", "content": instructions}, + {"role": "user", "content": question} + ] + result = oai_client.chat.completions.create( + messages=messages, + model="gpt-4o-mini", + temperature=0 + ) + return result.choices[0].message.content + + @traceable + def qa_pipeline(question: str) -> str: + """The full pipeline.""" + query = generate_wiki_search(question) + context = "\\n\\n".join([doc["page_content"] for doc in retrieve(query)]) + return generate_answer(question, context)`, typescript` import OpenAI from "openai"; import wiki from "wikipedia"; import { Client } from "langsmith"; import { traceable } from "langsmith/traceable"; import { wrapOpenAI } from "langsmith/wrappers"; - + const openai = wrapOpenAI(new OpenAI()); - + const generateWikiSearch = traceable( async (input: { question: string }) => { const messages = [ @@ -90,18 +130,18 @@ def rag_pipeline(question): }, { role: "user" as const, content: input.question }, ]; - + const chatCompletion = await openai.chat.completions.create({ model: "gpt-4o-mini", messages: messages, temperature: 0, }); - + return chatCompletion.choices[0].message.content ?? ""; }, { name: "generateWikiSearch" } ); - + const retrieve = traceable( async (input: { query: string; numDocuments: number }) => { const { results } = await wiki.search(input.query, { limit: 10 }); @@ -110,7 +150,7 @@ def rag_pipeline(question): type: "Document"; metadata: { url: string }; }> = []; - + for (const result of results) { if (finalResults.length >= input.numDocuments) { // Just return the top 2 pages for now @@ -124,12 +164,12 @@ def rag_pipeline(question): metadata: { url: page.fullurl }, }); } - + return finalResults; }, { name: "retrieve", run_type: "retriever" } ); - + const generateAnswer = traceable( async (input: { question: string; context: string }) => { const messages = [ @@ -139,7 +179,7 @@ def rag_pipeline(question): }, { role: "user" as const, content: input.question }, ]; - + const chatCompletion = await openai.chat.completions.create({ model: "gpt-4o-mini", messages: messages, @@ -149,7 +189,7 @@ def rag_pipeline(question): }, { name: "generateAnswer" } ); - + const ragPipeline = traceable( async ({ question }: { question: string }, numDocuments: number = 2) => { const query = await generateWikiSearch({ question }); @@ -162,8 +202,9 @@ def rag_pipeline(question): }, { name: "ragPipeline" } );`, - ]} - groupId="client-language" + +]} +groupId="client-language" /> This pipeline will produce a trace that looks something like: @@ -178,25 +219,23 @@ We are building a very simple dataset with a couple of examples to evaluate the python` from langsmith import Client - client = Client() - - examples = [ - ("What is LangChain?", "LangChain is an open-source framework for building applications using large language models."), - ("What is LangSmith?", "LangSmith is an observability and evaluation tool for LLM products, built by LangChain Inc.") - ] - + ls_client = Client() dataset_name = "Wikipedia RAG" - if not client.has_dataset(dataset_name=dataset_name): - dataset = client.create_dataset(dataset_name=dataset_name) - inputs, outputs = zip( - *[({"input": input}, {"expected": expected}) for input, expected in examples] + + if not ls_client.has_dataset(dataset_name=dataset_name): + dataset = ls_client.create_dataset(dataset_name=dataset_name) + ls_client.create_examples( + inputs=[ + {"question": "What is LangChain?"}, + {"question": "What is LangSmith?"}, + ], + dataset_id=dataset.id ) - client.create_examples(inputs=inputs, outputs=outputs, dataset_id=dataset.id) `, typescript` import { Client } from "langsmith"; const client = new Client(); - + const examples = [ [ "What is LangChain?", @@ -207,17 +246,18 @@ We are building a very simple dataset with a couple of examples to evaluate the "LangSmith is an observability and evaluation tool for LLM products, built by LangChain Inc.", ], ]; - + const datasetName = "Wikipedia RAG"; - + const inputs = examples.map(([input, _]) => ({ input })); const outputs = examples.map(([_, expected]) => ({ expected })); - + const dataset = await client.createDataset(datasetName); await client.createExamples({ datasetId: dataset.id, inputs, outputs }); `, - ]} - groupId="client-language" + +]} +groupId="client-language" /> ## 3. Define your custom evaluators @@ -225,61 +265,81 @@ We are building a very simple dataset with a couple of examples to evaluate the As mentioned above, we will define two evaluators: one that evaluates the relevance of the retrieved documents w.r.t the input query and another that evaluates the hallucination of the generated answer w.r.t the retrieved documents. We will be using LangChain LLM wrappers, along with [`with_structured_output`](https://python.langchain.com/v0.1/docs/modules/model_io/chat/structured_output/) to define the evaluator for hallucination. -The key here is that the evaluator function should traverse the `root_run` / `rootRun` argument to access the intermediate steps of the pipeline. The evaluator can then process the inputs and outputs of the intermediate steps to evaluate according to the desired criteria. +The key here is that the evaluator function should traverse the `run` / `rootRun` argument to access the intermediate steps of the pipeline. The evaluator can then process the inputs and outputs of the intermediate steps to evaluate according to the desired criteria. dict: - """ - A very simple evaluator that checks to see if the input of the retrieval step exists - in the retrieved docs. - """ - rag_pipeline_run = next(run for run in root_run.child_runs if run.name == "rag_pipeline") - retrieve_run = next(run for run in rag_pipeline_run.child_runs if run.name == "retrieve") - page_contents = "\\n\\n".join(doc["page_content"] for doc in retrieve_run.outputs["output"]) - score = retrieve_run.inputs["query"] in page_contents - return {"key": "simple_document_relevance", "score": score}\n -def hallucination(root_run: Run, example: Example) -> dict: - """ - A simple evaluator that checks to see the answer is grounded in the documents - """ - # Get documents and answer - rag_pipeline_run = next(run for run in root_run.child_runs if run.name == "rag_pipeline") - retrieve_run = next(run for run in rag_pipeline_run.child_runs if run.name == "retrieve") - page_contents = "\\n\\n".join(doc["page_content"] for doc in retrieve_run.outputs["output"]) - generation = rag_pipeline_run.outputs["output"]\n + python({ caption: "Example uses `langchain` for convenience, this is not required." })` + from langchain.chat_models import init_chat_model + from langsmith.schemas import Run + from pydantic import BaseModel, Field + + def document_relevance(run: Run) -> bool: + """Checks if retriever input exists in the retrieved docs.""" + qa_pipeline_run = next( + r for run in run.child_runs if r.name == "qa_pipeline" + ) + retrieve_run = next( + r for run in qa_pipeline_run.child_runs if r.name == "retrieve" + ) + page_contents = "\\n\\n".join( + doc["page_content"] for doc in retrieve_run.outputs["output"] + ) + return retrieve_run.inputs["query"] in page_contents + + # Data model class GradeHallucinations(BaseModel): - """Binary score for hallucination present in generation answer."""\n - binary_score: int = Field(description="Answer is grounded in the facts, 1 or 0")\n - # LLM with function call - llm = ChatOpenAI(model="gpt-4o-mini", temperature=0) - structured_llm_grader = llm.with_structured_output(GradeHallucinations)\n - # Prompt - system = """You are a grader assessing whether an LLM generation is grounded in / supported by a set of retrieved facts. \\n - Give a binary score 1 or 0, where 1 means that the answer is grounded in / supported by the set of facts.""" - hallucination_prompt = ChatPromptTemplate.from_messages( - [ - ("system", system), - ("human", "Set of facts: \\n\\n {documents} \\n\\n LLM generation: {generation}"), + """Binary score for hallucination present in generation answer.""" + + is_grounded: bool = Field(..., description="True if the answer is grounded in the facts, False otherwise.") + + # LLM with structured outputs for grading hallucinations + # For more see: https://python.langchain.com/docs/how_to/structured_output/ + grader_llm= init_chat_model("gpt-4o-mini", temperature=0).with_structured_output( + GradeHallucinations, + method="json_schema", + strict=True, + ) + + def no_hallucination(run: Run) -> bool: + """Check if the answer is grounded in the documents. + + Return True if there is no hallucination, False otherwise. + """ + # Get documents and answer + qa_pipeline_run = next( + r for r in run.child_runs if r.name == "qa_pipeline" + ) + retrieve_run = next( + r for r in qa_pipeline_run.child_runs if r.name == "retrieve" + ) + retrieved_content = "\\n\\n".join( + doc["page_content"] for doc in retrieve_run.outputs["output"] + ) + + # Construct prompt + instructions = ( + "You are a grader assessing whether an LLM generation is grounded in / " + "supported by a set of retrieved facts. Give a binary score 1 or 0, " + "where 1 means that the answer is grounded in / supported by the set of facts." + ) + + messages = [ + {"role": "system", "content": instructions}, + {"role": "user", "content": f"Set of facts:\\n{retrieved_content}\\n\\nLLM generation: {run.outputs['answer']}"}, ] - )\n - hallucination_grader = hallucination_prompt | structured_llm_grader - score = hallucination_grader.invoke({"documents": page_contents, "generation": generation}) - return {"key": "answer_hallucination", "score": int(score.binary_score)}`), + + grade = grader_llm.invoke(messages) + return grade.is_grounded`, typescript` import { EvaluationResult } from "langsmith/evaluation"; import { Run, Example } from "langsmith/schemas"; - + import { ChatPromptTemplate } from "@langchain/core/prompts"; import { ChatOpenAI } from "@langchain/openai"; import { z } from "zod"; - + function findNestedRun(run: Run, search: (run: Run) => boolean): Run | null { const queue: Run[] = [run]; while (queue.length > 0) { @@ -289,31 +349,31 @@ def hallucination(root_run: Run, example: Example) -> dict: } return null; } - + // A very simple evaluator that checks to see if the input of the retrieval step exists // in the retrieved docs. function documentRelevance(rootRun: Run, example: Example): EvaluationResult { const retrieveRun = findNestedRun(rootRun, (run) => run.name === "retrieve"); const docs: Array<{ page_content: string }> | undefined = retrieveRun.outputs?.outputs; - + const pageContents = docs?.map((doc) => doc.page_content).join("\\n\\n"); const score = pageContents.includes(retrieveRun.inputs?.query); return { key: "simple_document_relevance", score }; } - + async function hallucination( rootRun: Run, example: Example ): Promise { const rag = findNestedRun(rootRun, (run) => run.name === "ragPipeline"); const retrieve = findNestedRun(rootRun, (run) => run.name === "retrieve"); - + const docs: Array<{ page_content: string }> | undefined = retrieve.outputs?.outputs; - + const documents = docs?.map((doc) => doc.page_content).join("\\n\\n"); - + const prompt = ChatPromptTemplate.fromMessages<{ documents: string; generation: string; @@ -330,7 +390,7 @@ def hallucination(root_run: Run, example: Example) -> dict: "Set of facts: \\n\\n {documents} \\n\\n LLM generation: {generation}", ], ]); - + const llm = new ChatOpenAI({ model: "gpt-4o-mini", temperature: 0, @@ -343,18 +403,19 @@ def hallucination(root_run: Run, example: Example) -> dict: }) .describe("Binary score for hallucination present in generation answer.") ); - + const grader = prompt.pipe(llm); const score = await grader.invoke({ documents, generation: rag.outputs?.outputs, }); - + return { key: "answer_hallucination", score: score.binary_score }; } `, - ]} - groupId="client-language" + +]} +groupId="client-language" /> ## 4. Evaluate the pipeline @@ -364,26 +425,29 @@ Finally, we'll run `evaluate` with the custom evaluators defined above. dict: + """Wrap the qa_pipeline so it can accept the Example.inputs dict as input.""" + return {"answer": qa_pipeline(inputs["question"])} + + experiment_results = ls_client.evaluate( + qa_wrapper, data=dataset_name, - evaluators=[document_relevance, hallucination], + evaluators=[document_relevance, no_hallucination], experiment_prefix="rag-wiki-oai" ) `, typescript` import { evaluate } from "langsmith/evaluation"; - + await evaluate((inputs) => ragPipeline({ question: inputs.input }), { data: datasetName, evaluators: [hallucination, documentRelevance], experimentPrefix: "rag-wiki-oai", }); `, - ]} - groupId="client-language" + +]} +groupId="client-language" /> The experiment will contain the results of the evaluation, including the scores and comments from the evaluators: diff --git a/docs/evaluation/how_to_guides/evaluate_pairwise.mdx b/docs/evaluation/how_to_guides/evaluate_pairwise.mdx index f21ff146..e487393c 100644 --- a/docs/evaluation/how_to_guides/evaluate_pairwise.mdx +++ b/docs/evaluation/how_to_guides/evaluate_pairwise.mdx @@ -20,22 +20,22 @@ import { LangSmith supports evaluating **existing** experiments in a comparative manner. This allows you to score the outputs from multiple experiments against each other, rather than being confined to evaluating outputs one at a time. Think [LMSYS Chatbot Arena](https://chat.lmsys.org/) - this is the same concept! -To do this, use the [evaluate_comparative](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._runner.evaluate_comparative.html) / `evaluateComparative` function with two existing experiments. +To do this, use the [evaluate()](https://langsmith-docs-git-bagatur-rfcbuiltinsdkref-langchain.vercel.app/reference/python/evaluation/langsmith.evaluation._runner.evaluate) function with two existing experiments. If you haven't already created experiments to compare, check out our [quick start](../) or our [how-to guide](./evaluate_llm_application) to get started with evaluations. -## `evaluate_comparative` args +## `evaluate()` comparative args -:::note -Pairwise evaluations require `langsmith` SDK Python version `>=0.1.145` or JS version `>=0.1.24`. +:::info +This guide requires `langsmith` Python version `>=0.2.0` or JS version `>=0.2.9`. ::: -At its simplest, `evaluate_comparative` / `evaluateComparative` function takes the following arguments: +At its simplest, `evaluate` / `aevaluate` function takes the following arguments: -| Argument | Description | -| ------------- | ---------------------------------------------------------------------------------------------------------------------------------- | -| `experiments` | A list of the two **existing experiments** you would like to evaluate against each other. These can be uuids or experiment names. | -| `evaluators` | A list of the pairwise evaluators that you would like to attach to this evaluation. See the section below for how to define these. | +| Argument | Description | +| ------------ | ---------------------------------------------------------------------------------------------------------------------------------- | +| `target` | A list of the two **existing experiments** you would like to evaluate against each other. These can be uuids or experiment names. | +| `evaluators` | A list of the pairwise evaluators that you would like to attach to this evaluation. See the section below for how to define these. | Along with these, you can also pass in the following optional args: @@ -57,18 +57,13 @@ Pairwise evaluators are just functions with an expected signature. Custom evaluator functions must have specific argument names. They can take any subset of the following arguments: -Python and JS/TS - -- `runs: list[langsmith.schemas.Run]`: A two-item list of the full Run objects generated by the two experiments on the given example. Use this if you need access to intermediate steps or metadata about each run. -- `example: langsmith.schemas.Example`: The full dataset Example, including the example inputs, outputs (if available), and metdata (if available). - -Currently Python only - - `inputs: dict`: A dictionary of the inputs corresponding to a single example in a dataset. -- `outputs: list[dict]`: A two-item list of the outputs produced by each experiment on the given inputs. -- `reference_outputs: dict`: A dictionary of the reference outputs associated with the example, if available. +- `outputs: list[dict]`: A two-item list of the dict outputs produced by each experiment on the given inputs. +- `reference_outputs` / `referenceOutputs: dict`: A dictionary of the reference outputs associated with the example, if available. +- `runs: list[Run]`: A two-item list of the full [Run](/reference/data_formats/run_data_format) objects generated by the two experiments on the given example. Use this if you need access to intermediate steps or metadata about each run. +- `example: Example`: The full dataset [Example](/reference/data_formats/example_data_format), including the example inputs, outputs (if available), and metdata (if available). -For most Python use cases you'll only need `inputs`, `outputs`, and `reference_outputs`. `run` and `example` are useful only if you need some extra trace or example metadata outside of the actual inputs and outputs of the application. +For most use cases you'll only need `inputs`, `outputs`, and `reference_outputs` / `referenceOutputs`. `run` and `example` are useful only if you need some extra trace or example metadata outside of the actual inputs and outputs of the application. ### Evaluator output @@ -101,57 +96,57 @@ In the Python example below, we are pulling [this structured prompt](https://smi =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` from langchain import hub from langchain.chat_models import init_chat_model - from langsmith import evaluate_comparative + from langsmith import evaluate + # See the prompt: https://smith.langchain.com/hub/langchain-ai/pairwise-evaluation-2 prompt = hub.pull("langchain-ai/pairwise-evaluation-2") model = init_chat_model("gpt-4o") chain = prompt | model def ranked_preference(inputs: dict, outputs: list[dict]) -> list: + # Assumes example inputs have a 'question' key and experiment + # outputs have an 'answer' key. response = chain.invoke({ "question": inputs["question"], - "answer_a": outputs[0].get("output", "N/A"), - "answer_b": outputs[1].get("output", "N/A"), + "answer_a": outputs[0].get("answer", "N/A"), + "answer_b": outputs[1].get("answer", "N/A"), }) - preference = response["Preference"] - if preference == 1: + if response["Preference"] == 1: scores = [1, 0] - elif preference == 2: + elif response["Preference"] == 2: scores = [0, 1] else: scores = [0, 0] return scores - evaluate_comparative( - # Replace the following array with the names or IDs of your experiments - ["my-experiment-name-1", "my-experiment-name-2"], + evaluate( + ["experiment-1", "experiment-2"], # Replace with the names/IDs of your experiments evaluators=[ranked_preference], + randomize_order=True, + max_concurrency=4, ) `, - typescript({ - caption: - "Note: LangChain support inside `evaluate` / `evaluateComparative` is not supported yet. See [this issue](https://github.com/langchain-ai/langsmith-sdk/issues/598) for more details.", - })` - import type { Run, Example } from "langsmith"; - import { evaluateComparative } from "langsmith/evaluation"; + typescript({caption: "Requires `langsmith>=0.2.9`"})` + import { evaluate} from "langsmith/evaluation"; + import { Run } from "langsmith/schemas"; import { wrapOpenAI } from "langsmith/wrappers"; import OpenAI from "openai"; + import { z } from "zod"; const openai = wrapOpenAI(new OpenAI()); - import { z } from "zod"; - async function evaluatePairwise(runs: Run[], example: Example) { + async function rankedPreference({ inputs, runs }: { inputs: Record, runs: Run[] }) { const scores: Record = {}; const [runA, runB] = runs; if (!runA || !runB) throw new Error("Expected at least two runs"); const payload = { - question: example.inputs?.question, + question: inputs.question, answer_a: runA?.outputs?.output ?? "N/A", answer_b: runB?.outputs?.output ?? "N/A", }; @@ -228,8 +223,8 @@ In the Python example below, we are pulling [this structured prompt](https://smi return { key: "ranked_preference", scores }; } - await evaluateComparative(["earnest-name-40", "reflecting-pump-91"], { - evaluators: [evaluatePairwise], + await evaluate(["earnest-name-40", "reflecting-pump-91"], { + evaluators: [rankedPreference], }); `, diff --git a/docs/evaluation/how_to_guides/index.md b/docs/evaluation/how_to_guides/index.md index 06c671c0..120ee0c1 100644 --- a/docs/evaluation/how_to_guides/index.md +++ b/docs/evaluation/how_to_guides/index.md @@ -12,14 +12,14 @@ Evaluate and improve your application before deploying it. ### Run an evaluation -- [Run an evaluation](./how_to_guides/evaluate_llm_application) +- [Run an evaluation with the SDK](./how_to_guides/evaluate_llm_application) - [Run an evaluation asynchronously](./how_to_guides/async) - [Run an evaluation comparing two experiments](./how_to_guides/evaluate_pairwise) - [Evaluate a `langchain` runnable](./how_to_guides/langchain_runnable) - [Evaluate a `langgraph` graph](./how_to_guides/langgraph) - [Evaluate an existing experiment (Python only)](./how_to_guides/evaluate_existing_experiment) -- [Run an evaluation via the REST API](./how_to_guides/run_evals_api_only) - [Run an evaluation from the UI](./how_to_guides/run_evaluation_from_prompt_playground) +- [Run an evaluation via the REST API](./how_to_guides/run_evals_api_only) ### Define an evaluator @@ -28,7 +28,7 @@ Evaluate and improve your application before deploying it. - [Define a pairwise evaluator](./how_to_guides/evaluate_pairwise) - [Define a summary evaluator](./how_to_guides/summary) - [Use an off-the-shelf evaluator via the SDK (Python only)](./how_to_guides/use_langchain_off_the_shelf_evaluators) -- [Evaluate intermediate steps](./how_to_guides/evaluate_on_intermediate_steps) +- [Evaluate an application's intermediate steps](./how_to_guides/evaluate_on_intermediate_steps) - [Return multiple metrics in one evaluator](./how_to_guides/multiple_scores) - [Return categorical vs numerical metrics](./how_to_guides/metric_type) @@ -82,9 +82,9 @@ Manage datasets in LangSmith used by your evaluations. - [Export a dataset from the UI](./how_to_guides/manage_datasets_in_application#export-a-dataset) - [Create a dataset split from the UI](./how_to_guides/manage_datasets_in_application#create-and-manage-dataset-splits) - [Filter examples from the UI](./how_to_guides/manage_datasets_in_application#filter-examples) -- [Create a dataset via the SDK](./how_to_guides/manage_datasets_programmatically#create-a-dataset) -- [Fetch a dataset via the SDK](./how_to_guides/manage_datasets_programmatically#fetch-datasets) -- [Update a dataset via the SDK](./how_to_guides/manage_datasets_programmatically#update-examples) +- [Create a dataset with the SDK](./how_to_guides/manage_datasets_programmatically#create-a-dataset) +- [Fetch a dataset with the SDK](./how_to_guides/manage_datasets_programmatically#fetch-datasets) +- [Update a dataset with the SDK](./how_to_guides/manage_datasets_programmatically#update-examples) - [Version a dataset](./how_to_guides/version_datasets) - [Share/unshare a dataset publicly](./how_to_guides/share_dataset) - [Export filtered traces from an experiment to a dataset](./how_to_guides/export_filtered_traces_to_dataset) diff --git a/docs/evaluation/how_to_guides/langchain_runnable.mdx b/docs/evaluation/how_to_guides/langchain_runnable.mdx index faf2b216..54e2ff91 100644 --- a/docs/evaluation/how_to_guides/langchain_runnable.mdx +++ b/docs/evaluation/how_to_guides/langchain_runnable.mdx @@ -85,7 +85,7 @@ To evaluate our chain we can pass it directly to the `evaluate()` / `aevaluate() =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` from langsmith import aevaluate, Client client = Client() diff --git a/docs/evaluation/how_to_guides/langgraph.mdx b/docs/evaluation/how_to_guides/langgraph.mdx index 557c2a6c..66f791b2 100644 --- a/docs/evaluation/how_to_guides/langgraph.mdx +++ b/docs/evaluation/how_to_guides/langgraph.mdx @@ -136,7 +136,7 @@ And a simple evaluator: =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` judge_llm = init_chat_model("gpt-4o") @@ -184,7 +184,7 @@ If any of you nodes are defined as async, you'll need to use `aevaluate` =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` from langsmith import aevaluate def example_to_state(inputs: dict) -> dict: @@ -217,7 +217,7 @@ For example, we can look at the messages to assert that the model invoked the 's =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` def right_tool(outputs: dict) -> bool: tool_calls = outputs["messages"][1].tool_calls diff --git a/docs/evaluation/how_to_guides/llm_as_judge.mdx b/docs/evaluation/how_to_guides/llm_as_judge.mdx index b4d7ba8a..5a572c24 100644 --- a/docs/evaluation/how_to_guides/llm_as_judge.mdx +++ b/docs/evaluation/how_to_guides/llm_as_judge.mdx @@ -23,7 +23,7 @@ For maximal control of evaluator logic, we can write a custom evaluator and run =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` from langsmith import evaluate, traceable, wrappers, Client from openai import OpenAI # Assumes you've installed pydantic diff --git a/docs/evaluation/how_to_guides/local.mdx b/docs/evaluation/how_to_guides/local.mdx index 8ab10d62..5ee3f6bd 100644 --- a/docs/evaluation/how_to_guides/local.mdx +++ b/docs/evaluation/how_to_guides/local.mdx @@ -22,12 +22,12 @@ Let's take a look at an example: =0.1.145`. Example also uses `pandas`."})` - from langsmith import evaluate, Client + python({caption: "Requires `langsmith>=0.2.0`. Example also uses `pandas`."})` + from langsmith import Client # 1. Create and/or select your dataset - client = Client() - dataset = client.clone_public_dataset( + ls_client = Client() + dataset = ls_client.clone_public_dataset( "https://smith.langchain.com/public/a63525f9-bdf2-4512-83e3-077dc9417f96/d" ) @@ -40,7 +40,7 @@ Let's take a look at an example: return {"answer": inputs["question"] + " is a good question. I don't know the answer."} # 4. Run an evaluation - experiment = evaluate( + experiment = ls_client.evaluate( chatbot, data=dataset, evaluators=[is_concise], diff --git a/docs/evaluation/how_to_guides/metric_type.mdx b/docs/evaluation/how_to_guides/metric_type.mdx index 68610753..93ad9361 100644 --- a/docs/evaluation/how_to_guides/metric_type.mdx +++ b/docs/evaluation/how_to_guides/metric_type.mdx @@ -23,7 +23,7 @@ Here are some examples: =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` def numerical_metric(inputs: dict, outputs: dict, reference_outputs: dict) -> float: # Evaluation logic... diff --git a/docs/evaluation/how_to_guides/multiple_scores.mdx b/docs/evaluation/how_to_guides/multiple_scores.mdx index 17f3fb9d..e42cc175 100644 --- a/docs/evaluation/how_to_guides/multiple_scores.mdx +++ b/docs/evaluation/how_to_guides/multiple_scores.mdx @@ -35,7 +35,7 @@ Example: =0.1.145`"})` + python({caption: "Requires `langsmith>=0.2.0`"})` def multiple_scores(outputs: dict, reference_outputs: dict) -> list[dict]: # Replace with real evaluation logic. precision = 0.8 diff --git a/docs/evaluation/how_to_guides/rate_limiting.mdx b/docs/evaluation/how_to_guides/rate_limiting.mdx index 240bf482..41fda7f7 100644 --- a/docs/evaluation/how_to_guides/rate_limiting.mdx +++ b/docs/evaluation/how_to_guides/rate_limiting.mdx @@ -83,7 +83,8 @@ See some examples of how to do this in the [OpenAI docs](https://platform.openai ## Limiting max_concurrency Limiting the number of concurrent calls you're making to your application and evaluators is another way to decrease the frequency of model calls you're making, and in that way avoid rate limit errors. -`max_concurrency` can be set directly on the [evaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._runner.evaluate.html) / [aevaluate()](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._arunner.aevaluate.html) functions. +`max_concurrency` can be set directly on the [evaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._runner.evaluate) / [aevaluate()](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._arunner.aevaluate) functions. +This parallelizes evaluation by effectively splitting the dataset across threads. dict: - correct = 0 - for i, run in enumerate(runs): - if run.outputs["output"] == examples[i].outputs["label"]: - correct += 1 - if correct / len(runs) > 0.5: - return {"key": "pass", "score": True} - else: - return {"key": "pass", "score": False} + def pass_50(outputs: list[dict], reference_outputs: list[dict]) -> bool: + """Pass if >50% of all results are correct.""" + correct = sum([out["class"] == ref["label"] for out, ref in zip(outputs, reference_outputs)]) + return correct / len(outputs) > 0.5 `, typescript` - import { Run, Example } from "langsmith/schemas"; - - function summaryEval(runs: Run[], examples: Example[]) { + function summaryEval({ outputs, referenceOutputs }: { outputs: Record[], referenceOutputs?: Record[]}) { let correct = 0; - for (let i = 0; i < runs.length; i++) { - if (runs[i].outputs["output"] === examples[i].outputs["label"]) { + for (let i = 0; i < outputs.length; i++) { + if (outputs[i]["output"] === referenceOutputs[i]["label"]) { correct += 1; } } - return { key: "pass", score: correct / runs.length > 0.5 }; + return { key: "pass", score: correct / outputs.length > 0.5 }; } `, ]} @@ -52,23 +43,58 @@ You can then pass this evaluator to the `evaluate` method as follows: groupId="client-language" tabs={[ python` - results = evaluate( - lambda inputs: label_query(inputs["text"]), - data=dataset_name, - evaluators=[correct_label], - summary_evaluators=[summary_eval], - experiment_prefix="Toxic Queries", + from langsmith import Client + + ls_client = Client() + dataset = ls_client.clone_public_dataset( + "https://smith.langchain.com/public/3d6831e6-1680-4c88-94df-618c8e01fc55/d + ) + + def bad_classifier(inputs: dict) -> dict: + return {"class": "Not toxic"} + + def correct(outputs: dict, reference_outputs: dict) -> bool: + """Row-level correctness evaluator.""" + return outputs["class"] == reference_outputs["label"] + + results = ls_client.evaluate( + bad_classified, + data=dataset, + evaluators=[correct], + summary_evaluators=[pass_50], ) `, typescript` - await evaluate((inputs) => labelQuery(inputs["input"]), { + + import { Client } from "langsmith"; + import { evaluate } from "langsmith/evaluation"; + import type { EvaluationResult } from "langsmith/evaluation"; + + const client = new Client(); + const datasetName = "Toxic queries"; + const dataset = await client.clonePublicDataset( + "https://smith.langchain.com/public/3d6831e6-1680-4c88-94df-618c8e01fc55/d, + { datasetName: datasetName } + ); + + function correct({ outputs, referenceOutputs }: { outputs: Record, referenceOutputs?: Record }): EvaluationResult { + const score = outputs["class"] === referenceOutputs?["label"]; + return { key: "correct", score }; + } + + function badClassifier(inputs: Record): { class: string } { + return { class: "Not toxic" }; + } + + await evaluate(badClassifier, { data: datasetName, - evaluators: [correctLabel], + evaluators: [correct], summaryEvaluators: [summaryEval], experimentPrefix: "Toxic Queries", }); `, - ]} + +]} /> In the LangSmith UI, you'll the summary evaluator's score displayed with the corresponding key. diff --git a/docs/evaluation/index.mdx b/docs/evaluation/index.mdx index f8d368a3..c119cfc7 100644 --- a/docs/evaluation/index.mdx +++ b/docs/evaluation/index.mdx @@ -54,8 +54,8 @@ export LANGCHAIN_API_KEY=`), =0.1.145`"})` - from langsmith import evaluate, Client + python({caption: "Requires `langsmith>=0.2.0`"})` + from langsmith import Client # 1. Create and/or select your dataset client = Client() @@ -72,18 +72,20 @@ export LANGCHAIN_API_KEY=`), return {"answer": inputs["question"] + " is a good question. I don't know the answer."} # 4. Run an evaluation - evaluate( + experiment_results = client.evaluate( chatbot, data=dataset, evaluators=[is_concise], - experiment_prefix="my first experiment " + experiment_prefix="my first experiment ", + max_concurrency=4, ) `, - typescript`import { Client } from "langsmith"; + typescript({caption: "Requires `langsmith>=0.2.9`"})` +import { Client } from "langsmith"; import { evaluate } from "langsmith/evaluation"; import type { EvaluationResult } from "langsmith/evaluation"; -import type { Run, Example } from "langsmith/schemas";\n + // 1. Define a dataset const client = new Client(); const datasetName = "my first dataset" @@ -92,21 +94,21 @@ const dataset = await client.clonePublicDataset( { datasetName: datasetName } )\n // 2. Define an evaluator -function isConcise(rootRun: Run, example: Example): EvaluationResult { -const score = rootRun.outputs?.outputs.length < 3 \* example.outputs?.answer.length; +function isConcise({ outputs, referenceOutputs }: { outputs?: Record, referenceOutputs?: Record }): EvaluationResult { +const score = outputs?.answer.length < 3 \* referenceOutputs?.answer.length; return { key: "is_concise", score: score }; }\n // 3. Run an evaluation -// For more info on evaluators, see: https://docs.smith.langchain.com/concepts/evaluation#evaluators await evaluate( -(exampleInput) => { +(inputs: { question: string }) => { return { -answer: exampleInput.question + " Good question. I don't know the answer" +answer: inputs.question + " Good question. I don't know the answer" }; }, { data: datasetName, evaluators: [isConcise], experimentPrefix: "my first experiment ", +maxConcurrency: 4, });`, ]} groupId="client-language" diff --git a/docs/evaluation/tutorials/evaluation.mdx b/docs/evaluation/tutorials/evaluation.mdx index f10acb3a..31f7e573 100644 --- a/docs/evaluation/tutorials/evaluation.mdx +++ b/docs/evaluation/tutorials/evaluation.mdx @@ -19,7 +19,7 @@ At a high level, in this tutorial we will go over how to: - _Track results over time_ - _Set up automated testing to run in CI/CD_ -For more information on the evaluation workflows LangSmith supports, check out the [how-to guides](../../evaluation/how_to_guides), or see the reference docs for [evaluate](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._runner.evaluate.html) and its asynchronous [aevaluate](https://langsmith-sdk.readthedocs.io/en/latest/evaluation/langsmith.evaluation._arunner.aevaluate.html) counterpart. +For more information on the evaluation workflows LangSmith supports, check out the [how-to guides](../../evaluation/how_to_guides), or see the reference docs for [evaluate](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._runner.evaluate) and its asynchronous [aevaluate](https://docs.smith.langchain.com/reference/python/evaluation/langsmith.evaluation._arunner.aevaluate) counterpart. Lots to cover, let's dive in! diff --git a/docs/index.mdx b/docs/index.mdx index 8ad45eab..bb870143 100644 --- a/docs/index.mdx +++ b/docs/index.mdx @@ -8,9 +8,8 @@ import Tabs from "@theme/Tabs"; import CodeBlock from "@theme/CodeBlock"; import { CodeTabs, - PythonBlock, - TypeScriptBlock, typescript, + python, } from "@site/src/components/InstructionsWithCode"; import { LangChainInstallationCodeTabs, @@ -27,7 +26,9 @@ import { RegionalUrl } from "@site/src/components/RegionalUrls"; # Get started with LangSmith -**LangSmith** is a platform for building production-grade LLM applications. It allows you to closely monitor and evaluate your application, so you can ship quickly and with confidence. Use of [LangChain's open source frameworks](https://python.langchain.com) is not necessary - LangSmith works on its own! +**LangSmith** is a platform for building production-grade LLM applications. +It allows you to closely monitor and evaluate your application, so you can ship quickly and with confidence. +LangChain's open source frameworks [langchain](https://python.langchain.com) and [langgraph](https://langchain-ai.github.io/langgraph/) work seemlessly with LangSmith but are not necessary - LangSmith works on its own! ## 1. Install LangSmith @@ -95,47 +96,61 @@ Evaluation requires a system to test, data to serve as test cases, and optionall dict: + return {"response": "Welcome " + inputs["postfix"]} + + # Define your evaluator(s) + def exact_match(outputs: dict, reference_outputs: dict) -> bool: + return outputs["response"] == reference_outputs["response"] + + # Run the evaluation + experiment_results = client.evaluate( + dummy_app, # Your AI system goes here + data=dataset, # The data to predict and grade over + evaluators=[exact_match], # The evaluators to score the results + experiment_prefix="sample-experiment", # The name of the experiment + metadata={"version": "1.0.0", "revision_id": "beta"}, # Metadata about the experiment + max_concurrency=4, # Add concurrency. + ) + + # Analyze the results via the UI or programmatically + # If you have 'pandas' installed you can view the results as a + # pandas DataFrame by uncommenting below: + + # experiment_results.to_pandas() + `, - }, typescript` - import { Client, Run, Example } from "langsmith"; - import { EvaluationResult, evaluate } from "langsmith/evaluation"; - +import { Client } from "langsmith"; +import { EvaluationResult, evaluate } from "langsmith/evaluation"; + const client = new Client(); - + // Define dataset: these are your test cases const datasetName = "Sample Dataset"; const dataset = await client.createDataset(datasetName, { @@ -147,37 +162,37 @@ experiment_results = evaluate( { postfix: "to Evaluations in LangSmith" }, ], outputs: [ - { output: "Welcome to LangSmith" }, - { output: "Welcome to Evaluations in LangSmith" }, + { response: "Welcome to LangSmith" }, + { response: "Welcome to Evaluations in LangSmith" }, ], datasetId: dataset.id, }); - - // Define your evaluator - const exactMatch = async ( - run: Run, - example: Example - ): Promise => { + + // Define your evaluator(s) + const exactMatch = async ({ outputs, referenceOutputs }: { + outputs?: Record; + referenceOutputs?: Record; + }): Promise => { return { key: "exact_match", - score: run.outputs?.output === example?.outputs?.output, + score: outputs?.response === referenceOutputs?.response, }; }; - - await evaluate( - (input: { postfix: string }) => ({ output: \`Welcome $\{input.postfix\}\` }), + + // Run the evaluation + const experimentResults = await evaluate( + (inputs: { postfix: string }) => ({ response: \`Welcome $\{inputs.postfix\}\` }), { data: datasetName, evaluators: [exactMatch], - metadata: { - version: "1.0.0", - revision_id: "beta", - }, + metadata: { version: "1.0.0", revision_id: "beta" }, + maxConcurrency: 4, } ); `, - ]} - groupId="client-language" + +]} +groupId="client-language" /> - Click the link printed out by your evaluation run to access the LangSmith experiments UI, diff --git a/docs/reference/index.md b/docs/reference/index.md index 9c1ffb59..6f4e073c 100644 --- a/docs/reference/index.md +++ b/docs/reference/index.md @@ -14,7 +14,7 @@ Technical reference that covers components, APIs, and other aspects of LangSmith ### SDK -- [Python SDK Reference](https://langsmith-sdk.readthedocs.io/en/latest/) +- [Python SDK Reference](https://docs.smith.langchain.com/reference/python) - [LangChain off-the-shelf evaluators (Python only)](./reference/sdk_reference/langchain_evaluators) ### Common data types diff --git a/docs/reference/sdk_reference/_category_.json b/docs/reference/sdk_reference/_category_.json deleted file mode 100644 index b9f47770..00000000 --- a/docs/reference/sdk_reference/_category_.json +++ /dev/null @@ -1,5 +0,0 @@ -{ - "label": "SDK reference", - "collapsed": false, - "collapsible": true -} diff --git a/docs/reference/sdk_reference/index.mdx b/docs/reference/sdk_reference/index.mdx deleted file mode 100644 index bbb9e33a..00000000 --- a/docs/reference/sdk_reference/index.mdx +++ /dev/null @@ -1,3 +0,0 @@ -# SDK Reference - -- [Python SDK Reference](https://langsmith-sdk.readthedocs.io/en/latest/) diff --git a/docusaurus.config.js b/docusaurus.config.js index 33c16111..615e9960 100644 --- a/docusaurus.config.js +++ b/docusaurus.config.js @@ -134,9 +134,19 @@ const config = { position: "right", }, { - href: "https://api.smith.langchain.com/redoc", - label: "Go to API Docs", + type: "dropdown", + label: "API Reference", position: "left", + items: [ + { + label: "REST", + href: "https://api.smith.langchain.com/redoc", + }, + { + label: "Python", + to: "https://langsmith-docs-git-bagatur-rfcbuiltinsdkref-langchain.vercel.app/reference/python", + }, + ], }, ], }, diff --git a/package.json b/package.json index 54d64b71..2eb6fdf6 100644 --- a/package.json +++ b/package.json @@ -7,8 +7,8 @@ "docusaurus": "docusaurus", "test": "jest", "start": "rm -rf ./docs/api && docusaurus start", - "build": "rm -rf ./build && docusaurus build", - "vercel-build": "git submodule update --init --recursive && python3 -m ensurepip --default-pip && python3 -m pip install -r subdirectories/scripts/requirements.txt && python3 subdirectories/scripts/build_cookbook.py && ls docs && npm run build", + "build": "make vercel-build", + "vercel-build": "ls .; cat Makefile; make vercel-build", "swizzle": "docusaurus swizzle", "deploy": "docusaurus deploy", "clear": "docusaurus clear", diff --git a/sidebars.js b/sidebars.js index eb048e28..6b99e9d1 100644 --- a/sidebars.js +++ b/sidebars.js @@ -162,6 +162,11 @@ module.exports = { link: { type: "doc", id: "prompt_engineering/concepts/index" }, }, "langgraph_cloud", + { + type: "html", + value: + '
', + }, { type: "category", label: "Administration", @@ -205,7 +210,6 @@ module.exports = { ], link: { type: "doc", id: "administration/tutorials/index" }, }, - "administration/pricing", ], link: { type: "doc", id: "administration/concepts/index" }, }, @@ -222,6 +226,12 @@ module.exports = { ], link: { type: "doc", id: "self_hosting/index" }, }, + "administration/pricing", + { + type: "html", + value: + '
', + }, { type: "category", label: "Reference", diff --git a/src/components/QuickStart.js b/src/components/QuickStart.js index c4f87ac0..8a3a6e60 100644 --- a/src/components/QuickStart.js +++ b/src/components/QuickStart.js @@ -103,10 +103,9 @@ patch_run(parent_run_id, {"answer": chat_completion.choices[0].message.content}) export const PythonSDKTracingCode = () => `import openai -from langsmith.wrappers import wrap_openai -from langsmith import traceable\n +from langsmith import wrappers, traceable\n # Auto-trace LLM calls in-context -client = wrap_openai(openai.Client())\n +client = wrappers.wrap_openai(openai.Client())\n @traceable # Auto-trace this function def pipeline(user_input: str): result = client.chat.completions.create( diff --git a/vercel.json b/vercel.json index 75bd5c29..e5808ad4 100644 --- a/vercel.json +++ b/vercel.json @@ -201,6 +201,14 @@ { "source": "/evaluation/how_to_guides/human_feedback/:path*", "destination": "/evaluation/how_to_guides/:path*" + }, + { + "source": "/reference/python(/?)", + "destination": "/reference/python/reference" + }, + { + "source": "/reference/sdk_reference(/?)", + "destination": "/reference/" } ], "builds": [