Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Metric List","local":"metric-list","sections":[{"title":"Automatic metrics for multiple-choice tasks","local":"automatic-metrics-for-multiple-choice-tasks","sections":[],"depth":2},{"title":"Automatic metrics for perplexity and language modeling","local":"automatic-metrics-for-perplexity-and-language-modeling","sections":[],"depth":2},{"title":"Automatic metrics for generative tasks","local":"automatic-metrics-for-generative-tasks","sections":[],"depth":2},{"title":"LLM-as-Judge","local":"llm-as-judge","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/lighteval/pr_744/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/entry/start.a615223c.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/chunks/scheduler.7da89386.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/chunks/singletons.8c5be8fd.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/chunks/paths.86a4d49d.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/entry/app.b0033d27.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/chunks/index.20910acc.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/nodes/0.c40ee5c5.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/nodes/9.ba2b2988.js"> | |
| <link rel="modulepreload" href="/docs/lighteval/pr_744/en/_app/immutable/chunks/index.c9cd5e8b.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Metric List","local":"metric-list","sections":[{"title":"Automatic metrics for multiple-choice tasks","local":"automatic-metrics-for-multiple-choice-tasks","sections":[],"depth":2},{"title":"Automatic metrics for perplexity and language modeling","local":"automatic-metrics-for-perplexity-and-language-modeling","sections":[],"depth":2},{"title":"Automatic metrics for generative tasks","local":"automatic-metrics-for-generative-tasks","sections":[],"depth":2},{"title":"LLM-as-Judge","local":"llm-as-judge","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="metric-list" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#metric-list"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Metric List</span></h1> <h2 class="relative group"><a id="automatic-metrics-for-multiple-choice-tasks" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#automatic-metrics-for-multiple-choice-tasks"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Automatic metrics for multiple-choice tasks</span></h2> <p data-svelte-h="svelte-108o49i">These metrics use log-likelihood of the different possible targets.</p> <ul data-svelte-h="svelte-17kxjf7"><li><code>loglikelihood_acc</code>: Fraction of instances where the choice with the best logprob was correct - also exists in a faster version for tasks where the possible choices include only one token (<code>loglikelihood_acc_single_token</code>).</li> <li><code>loglikelihood_acc_norm</code>: Fraction of instances where the choice with the best logprob, normalized by sequence length, was correct - also exists in a faster version for tasks where the possible choices include only one token (<code>loglikelihood_acc_norm_single_token</code>).</li> <li><code>loglikelihood_acc_norm_nospace</code>: Fraction of instances where the choice with the best logprob, normalized by sequence length, was correct, with the first space ignored.</li> <li><code>loglikelihood_f1</code>: Corpus level F1 score of the multichoice selection - also exists in a faster version for tasks where the possible choices include only one token (<code>loglikelihood_f1_single_token</code>).</li> <li><code>mcc</code>: Matthew’s correlation coefficient (a measure of agreement between statistical distributions).</li> <li><code>recall_at_1</code>: Fraction of instances where the choice with the best logprob was correct - also exists in a faster version for tasks where the possible choices include only one token per choice (<code>recall_at_1_single_token</code>).</li> <li><code>recall_at_2</code>: Fraction of instances where the choice with the 2nd best logprob or better was correct - also exists in a faster version for tasks where the possible choices include only one token per choice (<code>recall_at_2_single_token</code>).</li> <li><code>mrr</code>: Mean reciprocal rank, a measure of the quality of a ranking of choices ordered by correctness/relevance - also exists in a faster version for tasks where the possible choices include only one token (<code>mrr_single_token</code>).</li> <li><code>target_perplexity</code>: Perplexity of the different choices available.</li> <li><code>acc_golds_likelihood</code>: A bit different, it actually checks if the average logprob of a single target is above or below 0.5.</li> <li><code>multi_f1_numeric</code>: Loglikelihood F1 score for multiple gold targets.</li></ul> <p data-svelte-h="svelte-1u8onu4">All these metrics also exist in a “single token” version (<code>loglikelihood_acc_single_token</code>, <code>loglikelihood_acc_norm_single_token</code>, <code>loglikelihood_f1_single_token</code>, <code>mcc_single_token</code>, <code>recall@2_single_token</code> and <code>mrr_single_token</code>). When the multichoice option compares only one token (ex: “A” vs “B” vs “C” vs “D”, or “yes” vs “no”), using these metrics in the single token version will divide the time spent by the number of choices. Single token evals also include:</p> <ul data-svelte-h="svelte-hkq1ua"><li><code>multi_f1_numeric</code>: Computes the f1 score of all possible choices and averages it.</li></ul> <h2 class="relative group"><a id="automatic-metrics-for-perplexity-and-language-modeling" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#automatic-metrics-for-perplexity-and-language-modeling"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Automatic metrics for perplexity and language modeling</span></h2> <p data-svelte-h="svelte-3tccvl">These metrics use log-likelihood of prompt.</p> <ul data-svelte-h="svelte-zznuqn"><li><code>word_perplexity</code>: Perplexity (log probability of the input) weighted by the number of words of the sequence.</li> <li><code>byte_perplexity</code>: Perplexity (log probability of the input) weighted by the number of bytes of the sequence.</li> <li><code>bits_per_byte</code>: Average number of bits per byte according to model probabilities.</li> <li><code>log_prob</code>: Predicted output’s average log probability (input’s log prob for language modeling).</li></ul> <h2 class="relative group"><a id="automatic-metrics-for-generative-tasks" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#automatic-metrics-for-generative-tasks"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Automatic metrics for generative tasks</span></h2> <p data-svelte-h="svelte-14ncypm">These metrics need the model to generate an output. They are therefore slower.</p> <ul data-svelte-h="svelte-hbi7zb"><li>Base:<ul><li><code>perfect_exact_match</code>: Fraction of instances where the prediction matches the gold exactly.</li> <li><code>exact_match</code>: Fraction of instances where the prediction matches the gold with the exception of the border whitespaces (= after a <code>strip</code> has been applied to both).</li> <li><code>quasi_exact_match</code>: Fraction of instances where the normalized prediction matches the normalized gold (normalization done on whitespace, articles, capitalization, …). Other variations exist, with other normalizers, such as <code>quasi_exact_match_triviaqa</code>, which only normalizes the predictions after applying a strip to all sentences.</li> <li><code>prefix_exact_match</code>: Fraction of instances where the beginning of the prediction matches the gold at the exception of the border whitespaces (= after a <code>strip</code> has been applied to both).</li> <li><code>prefix_quasi_exact_match</code>: Fraction of instances where the normalized beginning of the prediction matches the normalized gold (normalization done on whitespace, articles, capitalization, …).</li> <li><code>exact_match_indicator</code>: Exact match with some preceding context (before an indicator) removed.</li> <li><code>f1_score_quasi</code>: Average F1 score in terms of word overlap between the model output and gold, with both being normalized first.</li> <li><code>f1_score</code>: Average F1 score in terms of word overlap between the model output and gold without normalisation.</li> <li><code>f1_score_macro</code>: Corpus level macro F1 score.</li> <li><code>f1_score_macro</code>: Corpus level micro F1 score.</li> <li><code>maj_at_5</code> and <code>maj_at_8</code>: Model majority vote. Takes n (5 or 8) generations from the model and assumes the most frequent is the actual prediction.</li></ul></li> <li>Summarization:<ul><li><code>rouge</code>: Average ROUGE score <a href="https://aclanthology.org/W04-1013/" rel="nofollow">(Lin, 2004)</a>.</li> <li><code>rouge1</code>: Average ROUGE score <a href="https://aclanthology.org/W04-1013/" rel="nofollow">(Lin, 2004)</a> based on 1-gram overlap.</li> <li><code>rouge2</code>: Average ROUGE score <a href="https://aclanthology.org/W04-1013/" rel="nofollow">(Lin, 2004)</a> based on 2-gram overlap.</li> <li><code>rougeL</code>: Average ROUGE score <a href="https://aclanthology.org/W04-1013/" rel="nofollow">(Lin, 2004)</a> based on longest common subsequence overlap.</li> <li><code>rougeLsum</code>: Average ROUGE score <a href="https://aclanthology.org/W04-1013/" rel="nofollow">(Lin, 2004)</a> based on longest common subsequence overlap.</li> <li><code>rouge_t5</code> (BigBench): Corpus level ROUGE score for all available ROUGE metrics.</li> <li><code>faithfulness</code>: Faithfulness scores based on the SummaC method of <a href="https://aclanthology.org/2022.tacl-1.10/" rel="nofollow">Laban et al. (2022)</a>.</li> <li><code>extractiveness</code>: Reports, based on <a href="https://aclanthology.org/N18-1065/" rel="nofollow">(Grusky et al., 2018)</a>:<ul><li><code>summarization_coverage</code>: Extent to which the model-generated summaries are extractive fragments from the source document,</li> <li><code>summarization_density</code>: Extent to which the model-generated summaries are extractive summaries based on the source document,</li> <li><code>summarization_compression</code>: Extent to which the model-generated summaries are compressed relative to the source document.</li></ul></li> <li><code>bert_score</code>: Reports the average BERTScore precision, recall, and f1 score <a href="https://openreview.net/pdf?id=SkeHuCVFDr" rel="nofollow">(Zhang et al., 2020)</a> between model generation and gold summary.</li></ul></li> <li>Translation:<ul><li><code>bleu</code>: Corpus level BLEU score <a href="https://aclanthology.org/P02-1040/" rel="nofollow">(Papineni et al., 2002)</a> - uses the sacrebleu implementation.</li> <li><code>bleu_1</code>: Average sample BLEU score <a href="https://aclanthology.org/P02-1040/" rel="nofollow">(Papineni et al., 2002)</a> based on 1-gram overlap - uses the nltk implementation.</li> <li><code>bleu_4</code>: Average sample BLEU score <a href="https://aclanthology.org/P02-1040/" rel="nofollow">(Papineni et al., 2002)</a> based on 4-gram overlap - uses the nltk implementation.</li> <li><code>chrf</code>: Character n-gram matches f-score.</li> <li><code>ter</code>: Translation edit/error rate.</li></ul></li> <li>Copyright:<ul><li><code>copyright</code>: Reports:<ul><li><code>longest_common_prefix_length</code>: Average length of longest common prefix between model generation and reference,</li> <li><code>edit_distance</code>: Average Levenshtein edit distance between model generation and reference,</li> <li><code>edit_similarity</code>: Average Levenshtein edit similarity (normalized by the length of longer sequence) between model generation and reference.</li></ul></li></ul></li> <li>Math:<ul><li><code>quasi_exact_match_math</code>: Fraction of instances where the normalized prediction matches the normalized gold (normalization done for math, where latex symbols, units, etc are removed).</li> <li><code>maj_at_4_math</code>: Majority choice evaluation, using the math normalisation for the predictions and gold.</li> <li><code>quasi_exact_match_gsm8k</code>: Fraction of instances where the normalized prediction matches the normalized gold (normalization done for gsm8k, where latex symbols, units, etc are removed).</li> <li><code>maj_at_8_gsm8k</code>: Majority choice evaluation, using the gsm8k normalisation for the predictions and gold.</li></ul></li></ul> <h2 class="relative group"><a id="llm-as-judge" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#llm-as-judge"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>LLM-as-Judge</span></h2> <ul data-svelte-h="svelte-kd4sgy"><li><code>llm_judge_gpt3p5</code>: Can be used for any generative task, the model will be scored by a GPT3.5 model using the OpenAI API.</li> <li><code>llm_judge_llama_3_405b</code>: Can be used for any generative task, the model will be scored by a Llama 3.405B model using the HuggingFace API.</li> <li><code>llm_judge_multi_turn_gpt3p5</code>: Can be used for any generative task, the model will be scored by a GPT3.5 model using the OpenAI API. It is used for multiturn tasks like mt-bench.</li> <li><code>llm_judge_multi_turn_llama_3_405b</code>: Can be used for any generative task, the model will be scored by a Llama 3.405B model using the HuggingFace API. It is used for multiturn tasks like mt-bench.</li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/lighteval/blob/main/docs/source/metric-list.mdx" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p> | |
| <script> | |
| { | |
| __sveltekit_1kl62qe = { | |
| assets: "/docs/lighteval/pr_744/en", | |
| base: "/docs/lighteval/pr_744/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/lighteval/pr_744/en/_app/immutable/entry/start.a615223c.js"), | |
| import("/docs/lighteval/pr_744/en/_app/immutable/entry/app.b0033d27.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 9], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 19.7 kB
- Xet hash:
- 917b974c4ddea64d36bacc954a7f8806a7330cb2c0b897cf41483fa7b07d94fc
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.