中文版请见：Anki 新算法 FSRS 配置指南
- The Ultra Short Version
- Step 1: Enable FSRS
- Step 2: Configure FSRS settings
- Step 3: Find optimal parameters
- Step 4: (optional) Evaluate the parameters
- Step 5: (optional) Compute minimum recommended retention
- Step 6: (optional) Custom Scheduling
- FAQ
Are you busy and have no time to waste? Here's a summary of the guide.
- Go to deck options and enable FSRS under the "FSRS" section (under the "Advanced" section in Anki versions before 24.04), at the bottom of the deck options page. FSRS can only be enabled globally; you cannot enable it for some presets and disable it for others.
- Ensure that all your learning and re-learning steps are shorter than
1dand can be completed on the same day.23his not recommended even though it's less than one day because you won't be able to finish this step on the same day as your first review. Steps such as10mor30mare good. - Click the "Optimize" button under the "FSRS parameters" field. If you see a message that says "The FSRS parameters currently appear to be optimal", that's fine. In the versions before 24.06, an error message might pop up, saying that you don't have a sufficient number of reviews (400 in Anki 24.04, 1000 in older versions). In that case, use FSRS with the default parameters; it's still better than using the legacy SM-2 algorithm.
- Choose a value of desired retention: the proportion of cards recalled successfully when they are due. This is the most important setting in FSRS. Higher retention leads to shorter intervals and more reviews per day. 80-95% is reasonable, 90% should work fine for most people. Parameters and desired retention are preset-specific, you can make multiple presets with different parameters and desired retention.
Don't forget to click "Save" before closing the deck options window.
FSRS can adapt to almost any habit, except for one habit: pressing "Hard" instead of "Again" when you forget the information. When you press "Hard", FSRS assumes you have recalled the information correctly (though with hesitation and a lot of mental effort). If you press "Hard" when you have failed to recall the information, the intervals will be unreasonably high (for all the ratings). So, if you have this habit, please change it and use "Again" when you forget the information.
Regarding add-on compatibility, as a general rule of thumb, if an add-on affects intervals and scheduling in some way, it shouldn't be used with FSRS.
You are now ready to use FSRS!
To enable FSRS, go to Deck Options, scroll down to the "Advanced" section ("FSRS" in Anki 24.04), and toggle FSRS. This setting is shared by all deck presets. Note that after enabling FSRS, several settings, such as "Graduating interval", "Easy bonus", etc. will disappear. This is because these settings are irrelevant when FSRS is enabled.
If you have previously used FSRS using the custom scheduling method, please delete the FSRS code in the custom scheduling field before enabling the native FSRS. Also, if you are using the FSRS4Anki Helper add-on, check for add-on updates to ensure that the add-on has been updated to the latest version.
The most important setting to configure is the desired retention: the fraction of cards recalled successfully when they are due.
The permissible range for desired retention is 0.70 to 0.97 (0.7 to 0.99 in Anki 23.10.1 or newer). Higher retention leads to more reviews per day.
Be conservative when adjusting this setting - higher values will greatly increase your workload, and lower values can be demoralizing when you forget a lot of material.
The following chart illustrates how the workload changes with retention. The exact shape of the curve depends on the user's parameters and learning habits. To find the optimal retention for you, please use the "Compute minimum recommended retention (experimental)" feature, which is explained in Step 5.
Initially, users were not allowed to set the desired retention outside of the 0.70-0.97 range because it would make learning inefficient. In Anki 23.10.1, the range has been extended to 0.70-0.99 at the request of some users. However, setting the desired retention above 0.97 is still not advised for two main reasons:
- Such a high desired retention will significantly increase your workload (cards per day). The repetitions will be so frequent that you will dread doing your reviews before you even discover the power of spaced repetition.
- With such high retention, each review will contribute minimally to your overall learning. This essentially transforms the spaced repetition system into a massed repetition system, thereby undermining the advantages of the spacing effect.
The Maximum interval setting works the same way as when using the default algorithm. It is the maximum number of days that a card can wait until it is shown again. For more information, see Maximum interval in the Anki manual.
"Historical retention" (called "SM-2 retention" before Anki 24.04) is the average retention in the past.
When some of your review history is missing, FSRS needs to fill in the gaps. By default, it will assume that when you did those old reviews, you remembered 90% of the material. If your old retention was significantly higher or lower than 90%, adjusting this option will allow FSRS to better approximate the missing reviews. Your review history may be incomplete for two reasons:
- Because you've used the 'Ignore reviews before' option to omit (not delete) some of your review history.
- Because you've previously deleted review logs to free up space, used some add-ons that modify the review history, or imported material from a different SRS program.
The latter is quite rare, so unless you've used the 'Ignore reviews before' option, you probably don't need to adjust this setting. Even in Anki 24.04, it is located in the "Advanced" section, not in the "FSRS" section.
When FSRS is enabled, the learning and re-learning steps should be chosen in such a way that all the learning steps can be completed on the same day. In general, any steps longer than 12-14 hours are not recommended because most people will not be able to finish such steps on the same day as their first step. A single reasonable learning step can be 10m, 15m, 20m or 30m.
The reason is that FSRS can determine more optimal intervals but the use of longer (re)learning steps doesn't allow FSRS to schedule the reviews, making the scheduling less optimal. In addition, if longer steps are used, there can be cases where the "Hard" interval exceeds the "Good" interval.
The use of multiple short (re)learning steps, such as "5m 10m 15m 30m", is also discouraged. However, if you notice that your retention for young cards is much lower than desired, adding more intraday learning steps (such as 2h or 4h) may be helpful.
This option controls whether the due dates of cards will be changed when you enable FSRS or change the parameters and/or desired retention. By default, the cards are not rescheduled. This means that future reviews will use the new scheduling, but there will be no immediate change to your workload. This allows a smooth and gradual transition from SM-2 to FSRS.
If rescheduling is enabled, the due dates of cards will be immediately changed. This often results in a large number of cards becoming due, so is not recommended when first switching from SM-2.
This is a new feature added in Anki 24.04. If set, cards which were reviewed before the provided date will be ignored when optimizing and evaluating FSRS parameters. More technically speaking, if a card has no learning steps after the provided date, none of its reviews will be used for optimization and evaluation. This feature doesn't permanently delete review history. Important: if all of your cards have been reviewed before the selected date and you will not be adding any new cards, then the optimizer will always have zero data to work with.
This can be useful if you imported someone else's scheduling data, or if you have been misusing Hard. Hard should be used as a passing grade, not a failing grade. This feature can also be helpful if your learning or rating habits have changed significantly over time. If you are using this feature, it's important to accurately select your Historical Retention.
The FSRS optimizer uses machine learning to learn your memory patterns and find parameters that best fit your review history. So, the optimizer requires several reviews to fine-tune the parameters. Both reviews that were done before and after enabling FSRS count. If you took a long break from Anki and then reviewed cards that are long overdue, don't worry, it won't skew your parameters, as long as you answer honestly. Internally, FSRS treats Again as "fail" and Hard/Good/Easy as "pass".
You can find the optimal parameters for your cards by using the "Optimize" button under the "FSRS parameters" field. The optimal parameters will replace the default parameters automatically.
In Anki 24.06+, there is no minimum number of reviews required for optimization. Based on the number of reviews available, Anki will decide which parameters to optimize. In Anki 24.04, at least 400 reviews are required; in older versions, at least 1000 reviews are required. If you are using one of those versions and don't have enough reviews across all cards that this preset is applied to, please use the default parameters that are already entered into the "FSRS parameters" field. Even with the default parameters, FSRS is better than the default Anki algorithm (SM-2).
The parameters are preset-specific. If you have decks that vary wildly in subjective difficulty (not FSRS difficulty), it is recommended to use separate presets for them because the parameters for easy decks and hard decks will be different. Parameters and desired retention are independent, you do not need to re-optimize parameters if you have changed desired retention.
If you are confident in your knowledge of the algorithm and are willing to do extra work, there is a way to check whether a deck benefits from having it's own preset.
- Go to deck options of the deck you want to create a new preset for. Now copy the parameters from the
FSRS parametersfield. We will use this later for evaluation. - Now create a new preset and save it to this deck.
- Paste the parameters you previously copied in the
FSRS parametersfield of the newly created preset. - Click
Evaluateand write down the RMSE and log loss values. - Now click
Optimizeto obtain a new set of parameters. ClickEvaluateand write down the RMSE and log loss values. - If this new set of parameters results in lower RMSE and log loss values, then save the preset along with the new parameters.
Don't worry, FSRS will still perform well even without the aforementioned steps. This method is optional.
Parameters are calculated from the review history of all decks that use the current preset. If you want to alter which cards are used for optimizing the parameters (such as excluding suspended cards), you can adjust the search before calculating the parameters. The search works the same way as it does in the Browser. For details, see Searching in the Anki Manual. Optimization doesn't happen automatically, you have to manually click "Optimize" or "Optimize all presets".
An option to optimize all presets has been added in Anki 23.12, it's useful if you have a lot of presets. Don't forget to click "Save" after changing settings, otherwise, your changes won't be saved. Also remember that the settings of the preset applied to subdecks take priority over the settings of the preset applied to the parent deck.
If you ever want to reset parameters to their default values, click the anticlockwise open circle arrow to the right and bottom of the field with parameters.
You can use the "Evaluate" button under the "FSRS parameters" field to see metrics that tell how well the parameters in the "FSRS parameters" field fit your review history. Smaller numbers indicate a better fit to your review history.
Log loss doesn't have an intuitive interpretation. RMSE (bins) can be interpreted as the average difference between the predicted probability of recalling a card (R) and the measured (from the review history) probability. For example, RMSE=0.05 means that, on average, FSRS is off by 5% when predicting R.
Note that log loss and RMSE (bins) are not perfectly correlated, so two decks may have similar RMSE values but very different log loss values, and vice-versa.
It is an experimental tool that tries to calculate a value of desired retention that minimizes the ratio of the amount of time spent studying to the total knowledge acquired. Simply put, it tries to find the value of the desired retention that gives you the most efficient study plan. It does so by analyzing how much time you spend on your cards, as well as your habits of pressing Hard/Good/Easy, and creating a simulation.
You can adjust "Days to simulate" to fit your needs. If you are preparing for an exam that is 12 months away, set "Days to simulate" to 365. If you are a language learner, 5 years (1825 days) is a reasonable timeframe.
The recommended value can be useful as a reference when you have no idea of what you want your desired retention to be. However, since the tool is experimental, you should also use your own intuition when choosing desired retention.
In Anki 24.04, this feature has been reworked. You no longer need to input "Deck size" and "Minutes study/day", only "Days to simulate". The result will also depend on the maximum interval setting. Previously, the goal was to find the value of desired retention that would allow you to remember as much as possible within the given time constraints. Now the goal is to find the value of desired retention that minimizes the workload/acquired knowledge ratio. For more details, please read this: https://github.com/open-spaced-repetition/fsrs4anki/wiki/The-Optimal-Retention. It's important to note that you can set your desired retention higher than recommended if you want to do more work (minutes of studying per day) to remember more, but you shouldn't set your desired retention lower than recommended because you would have to do more work to remember less.
In Anki 24.04.1, this feature has been renamed to "Compute minimum recommended retention", to make it more clear that users shouldn't set their desired retention below the recommended value.
"Custom scheduling" allows you to introduce new scheduling rules on top of FSRS. This feature is only for advanced users and developers.
Q1: I am confused about v2, v3, FSRS v4, etc. Can you explain what's the difference?
A1: v2 scheduler (algorithm: SM-2): this is the old Anki scheduler, not supported in Anki 23.10 or newer.
v3 scheduler (algorithm: SM-2 or FSRS): this is a new scheduler for Anki. It handles the order of cards, timezones, and some deck options differently. It does not change the formulas that are used to calculate interval lengths.
SM-2: a simple 30-year-old algorithm developed by Piotr Wozniak, the creator of SuperMemo. Due to its simplicity, it's quite popular and is still used in flashcard apps to this day. It's the default Anki algorithm.
FSRS, or Free Spaced Repetition Scheduler: an open-source algorithm that combines machine learning techniques with universal memory formulas. It has recently been integrated into Anki as an alternative to SM-2. The v3 scheduler must be enabled in order to use FSRS. There are two versions of FSRS: FSRS v4 and FSRS-4.5. They have the same number of parameters, but the shape of the forgetting curve has been changed. All newest versions of Anki use FSRS-4.5, but some older versions, such as Anki 23.10, use FSRS v4.
Q2: Which platforms support FSRS?
A2: As of February 2024, FSRS is supported on all platforms: desktop Anki (Windows, Mac, and Linux), AnkiWeb (browser), AnkiMobile (iOS), and AnkiDroid (Android). Note that the latest version of AnkiDroid doesn't use the year.month.patch naming convention, while Anki and AnkiMobile do.
If you have just updated to AnkiDroid 2.17 and had enabled FSRS on other devices before updating to AnkiDroid 2.17, you may need to force a full-sync from AnkiWeb on AnkiDroid so that FSRS works correctly.
Q3: Does FSRS change the way the card's ease changes?
A3: Anki's built-in ease factor doesn't affect anything once FSRS is enabled. This is also why a lot of settings, such as Starting Ease, are hidden once FSRS is enabled.
Q4: Once I started using FSRS on my existing deck, if I ever wanted to go back to using Anki's built-in algorithm for the same deck, would that still be possible?
A4: Yes, just turn FSRS off. However, the intervals will not change after turning off FSRS.
Q5: I'm sure I have >1000 reviews (>400 in Anki 24.04), yet when I try to optimize parameters for my preset, I get an error telling me that I don't have enough reviews. Is that a bug?
A5: FSRS only takes into account one review per day. If you review a card multiple times per day, only the chronologically first review will be used by the optimizer. Also, if your deck has subdecks, ensure that the preset is applied to the subdecks, not just to the parent deck. Note that in Anki 24.06 and newer, there is no minimum limit for the number of reviews to use "Optimize".
Q6: My first interval is too long! Is this normal?
A6: In short, giving long first intervals is one of the strengths of FSRS. Don't be surprised if your first interval for "Good" is close to a week and your first interval for "Easy" is several weeks long. Read further for a deeper explanation:
For many users, the default algorithm (SM-2) tends to show new cards at unnecessarily short intervals. So, when users switch to FSRS, they tend to feel that the intervals given to new cards are too large. But these larger intervals match the desired retention better. By using these larger intervals, FSRS can prevent many of the unnecessary reviews that happen when using SM-2. So, it is advisable to try using these larger first intervals for a few days and see how it goes. It's worth mentioning that for mature cards, the opposite is true: FSRS is more conservative than SM-2.
If you still want to decrease the intervals, you can increase your desired retention. But note that this will decrease all the intervals, not just the first intervals.
Q7: Suppose I have a parent deck with its own preset, and each subdeck has a different preset. When I click on the parent deck to review a card that came from a subdeck, will the parameters of the preset of the parent deck be applied to the card, or the parameters of the preset of the subdeck that this card came from?
A7: The latter. Simply put, if you have something like ParentDeck::SubDeck, and the card came from the subdeck, the parameters of the preset corresponding to the subdeck will be applied.
Q8: I only use "Again" and "Good", will FSRS work fine?
A8: Yes. According to our research, FSRS is a little more accurate for people who mostly use "Again" and "Good" than for people who use all 4 buttons a lot. However, this conclusion may change as we investigate this further.
Also, unlike SM-2, FSRS doesn't suffer from the problem of "Ease Hell". This problem is solved by mean reversion of difficulty. If you press good continuously, the difficulty will converge to
However, note that you should not change your rating habits. This is because FSRS uses your past rating history to determine optimal intervals for your future reviews.
Q9: How can I grade the card to make FSRS more effective?
A9: The grade should be chosen based only on how easy it was to answer the card, not how long you want to wait until you see it again. For example, if you habitually avoid the easy button because it shows long intervals, you can end up in a negative cycle: you'd be making the "easy" situations even rarer, and "Easy" intervals will become longer and longer. This means you should ignore the intervals shown above the answer buttons and instead focus on how well you recall the information.
It's also very important to not press "Hard" when you forget a card. Press "Again" if you forgot it, and press "Hard" only if you recalled it after a lot of hesitation.
If you still want to see a deck sooner rather than later, for example because you have an exam coming up, you can use the Advance function of the Helper add-on. Advance is the preferable method because it doesn't skew the grading history of the cards.
Q10: How can I confirm that FSRS is working?
A10: Review a new card, remember what intervals you saw above the answer buttons. Undo review. Now set the desired retention either to 0.99 (maximum) or to 0.7 (minimum), and review the card again. You should see different intervals. Alternatively, download the Helper add-on and enable "Display memory state after answer". If the intervals don't change, make sure that you have applied the preset to the right deck. Remember that the settings of the preset applied to subdecks take priority over the settings of the preset applied to the parent deck.
Q11: Is it better to use the same parameters for all my cards or use different presets with different parameters?
A11: The answer to this question depends entirely on how similar your material is. For example, if you are learning Japanese and geography, it is recommended to use two different presets with different parameters. If you have two decks with Japanese vocabulary, you should use the same preset for both of them. Generally speaking, it's usually better to have different presets.
Q12: How often should I re-optimize parameters?
A12: Once per month should be more than enough. A more sophisticated rule is to optimize after every 2^n reviews: after 512, then after 1024, then after 2048, etc. But the "one month" rule is simpler.
Q13: What will happen if I review my cards on a device where FSRS is not supported (or disabled) and then on another device where FSRS is enabled?
A13: Your intervals will become inaccurate, but it won't corrupt your cards and make them unusable. It will just make FSRS bad at what it's supposed to do: maintain your retention at a specified level.
Q14: Does FSRS take into account delays?
A14: Yes, it does. In FSRS, a delay in reviewing (i.e., overdue reviews) affects the next interval as follows:
As the delay increases, retrievability (R) decreases. If the review was successful, the subsequent stability (S) would be higher. However, instead of increasing linearly with the delay like the SM-2/Anki algorithm, the subsequent stability converges to an upper limit, which depends on your FSRS parameters. For details, see The Algorithm.
Q15: Does FSRS take into account the time that I spend reviewing a card?
A15: No, FSRS only needs interval lengths and grades. However, the amount of time you spend on reviews is used when calculating optimal retention using the "Compute minimum recommended retention (experimental)" feature.
Q16: My log loss and RMSE are extremely high, how do I fix this?
A16: There is no way to fix that, the only thing you can do is keep doing reviews. FSRS is more accurate for people who have a lot of data.
Q17: Why is my retention of young cards significantly lower than my retention of mature cards?
A17: When your cards' stability is very low, the best interval should be shorter than 1 day. But Anki doesn't allow cards that are in the "review" phase to have intervals shorter than 1 day, only cards in the "learning" or "relearning" phase can have such short intervals. As a result, FSRS ends up giving you suboptimal intervals that are too long.
For more details, please read this post: https://www.reddit.com/r/Anki/comments/193x8kn/a_specific_case_where_fsrs_couldnt_ensure_the/
Q18: My retention is poor with the default parameters, and the first interval is definitely too long for me. How do I solve it?
A18: The default parameters are generated from 20k collections. They are the median values of 20k sets of parameters. Thus, inevitably, half of the new users will find that their retention is lower than desired retention, while the other half will discover that their retention exceeds their desired retention.
If the true retention is significantly lower than your desired retention, try increasing the desired retention to compensate. You can check your true retention using the FSRS4Anki Helper add-on, just Shift + Left Mouse Click on Stats. Of course, if you have a large number of reviews, it is advisable to use the optimized parameters rather than the default parameters.
Q19: Why my workload increases/decreases significantly after I switch to FSRS?
A19: The workload depends on your retention:
- If your true retention before switching to FSRS was significantly lower than your current desired retention, FSRS will let you review more frequently than before.
- If your previous true retention was significantly higher than your current desired retention, FSRS will let you review less frequently.
Didn't find the answer to your question? You can browse through several other questions asked by users: https://github.com/open-spaced-repetition/fsrs4anki/issues?q=is%3Aissue+label%3Aquestion+
Is your problem still unsolved? Please open a new issue to provide the details: https://github.com/open-spaced-repetition/fsrs4anki/issues/new/choose