Very nice. It’s mind-blowing to me that it used to take me months to build and deploy a forecasting model five years ago. Now in minutes I can use something like this, implemented with help from an LLM, and get apparently quite good results (and fine-tune on my data, it seems).
Speed looks nice too. I want an open weight model with this performance now.
Meh....time series data is fundamentally different from textual data. I see why a time series "foundation model" seems appealing, especially to the marketing folks at Mistral. Real-world usability is very limited, which is probably the reason why the blog post features an imaginary customer rather than an actual reference.
Time based data is extremely important, and there are tons of techniques out there to train super fast super small models, but they are a pain in the ass to tune. I could see a foundation model that is kind of good at a lot of things, that could be fine tuned or used when you want something more straight forward to use to analyze a bunch of data.
Honestly this might be very good for feature selection to refine a much more specialized model.
Any 'vibe check' on these; I don't have benchmarks but I tried out Le Chonk for general coding work that they claim it works really well at and I was very disappointed, it seemed to make all kinds of obvious errors of the kind that I haven't seen from opus, sol, deepseek flash etc in over a year
I've been using Mistral Large 4 and some of my other friends have as well.
Something is going on with the performance (as in speed of the service, not quality). Some days it's quite fast, some days it's so slow I interrupt it because I thought it was hanging.
I think they're shifting hosting capacity around, or otherwise messing with the model. Note that it's still a "preview" version, meaning they're still training it further for the final release and they are likely still figuring out the optimal hosting configuration as well.
It wouldn't surprise me to see further changes in both quality of the model and speed before the final release.
It was for me too, and it's only become slow very recently. I can't say with certainty, but I think it was around the time Mistral 4 launched that it seemed to slow down.
Odd, I am in France too. Progressive rollout maybe? I find it weird to announce a model with fanfare but not make it immediately available to customers, let alone paid ones.
I updated my vibe config to point to Mistral 4 and it charges my subscription capacity, not my API allowance. So it is included in subscriptions, you just need to do a little bit of legwork.
My field is primarily audio/vision ML so I don't have any particular domain expertise, but I don't get how you can lump all "time-series" together in one big bucket. My naive understanding is that the way to improve the accuracy is to gain as much joint information as possible. Like:
> Known-future covariates. Some drivers of sensor behaviour are known in advance, such as production schedules, planned shutdowns, weather forecasts. Timestral can condition on them to improve forecast quality, so a scheduled load increase is expected rather than flagged as a surprise.
To me this should be the entire basis of the system, not a condition to improve forecast quality? Like if you're modelling electricity consumption, if all you have is a window of N previous samples, you're going to get completely beaten by a comparatively simple system that knows the time of day, that there's holidays coming up, the weather forecast, etc.
I would also think that these time-series forecasting models would gain from a sophisticated understanding at least the labels of the covariates? A 5M model has no idea what a "bank holiday" is, it just sees an arbitrary class go from 0 to 1, and won't have any useful inductive biases as to what it means or how to link them to other pieces of information.
E.g., wouldn't you predict €/MWh of electricity far more accurately with access to price of natural gas (itself influenced by many geopolitical factors), breakdown in composition of production of each energy source, predictions about roll-out (requiring a sophisticated LLM capable of understanding news and their implications), etc. What's the use in trying to predict it from 2 variables and a narrow window looking back at just 8192 or 32768 samples?
Or perhaps I misunderstand the use-case and this is more for extremely short term movements?
You're right but, if the time series looks like music then you can apply your music domain knowledge. You don't have to only apply knowledge that's general across every domain, part of the magic of large neural nets is that you can do recognition and prediction at the same time thereby using (statistically) things known from training about how to identify a sequence and then also how to continue it.
Sure adding outside info helps even more, but with enough data it should become unnecessary (or so goes the theory).
But why would you throw away that data at all? These are tiny models compared to domain-specific audio/music/image models, and any capacity spent on recognition is capacity that could have been used for prediction.
Is there a reason you don't throw in something like cross-attention on an embedding of the sequence description/data labels? I've trained numerous generative models for audio and images and it's always the case that the unconditional model performs worse, trains slower, etc. than the conditional one (with the best performance from a hybrid using classifier-free guidance or similar scheme).
Again I don't have experience with the domain of time-series forecasting directly, but I would think there are generalities between autoregressive generative models and time-series forecasting, given that both involve predicting the next sample(s) given N regularly sampled past inputs.
Do you actually find that for some use cases foundation models work better than AutoARIMA or etc? I do a lot of ecom projections and it’s never been the case for me, most of my cases are best with seasonal naive.
Would love to know actual time series foundation model use cases!
Exactly this. It's a zero-shot model that can understand what a time series is with very little a-priori information. I would assume they've trained across a wide diversity of real world signals.
For some kinds of forecasting it doesn't need to be exactly accurate - eg say you are forecasting traffic for how to scale a load balancer, it can be a bit off it will just cost you a little and you can adjust after - a lot of feedback systems dont need full accuracy. You wouldnt use this to forecast whether to buy shares. It basically replaces various traditional curve fitting tools. If you need a really good forecast yes you are going to build a domain model. At least this is my understanding...
Agree on the share-buying point. I keep a side project that tracks about twenty published bitcoin price models side by side and grades each one on its own forecast record. Two lessons carried over to any time-series benchmark for me. First, freeze every forecast as it stood the day it was made; refitting on later data flatters everything. Second, put the naive baseline (tomorrow looks like today) in the same table. The best-known model in my set, stock-to-flow, sits more than 100% off over the past year, and it looked brilliant in backtests. A 0.03 gain means little to me until I see it next to persistence.
> say you are forecasting traffic for how to scale a load balancer, it can be a bit off
Problem is you really don't even need ML for the curve fitting tools. Just ARIMA or Holt Winters is enough and likely faster. Heck, Kubernetes HPA doesn't even implement full PID, just P.
It seems like the main challenge is going to be finding domains where accuracy is important enough to spend inference token money on, but not so important that you'd hire a team to build a custom model. Or maybe that team borrows Timestral weights to fine tune to the domain, but they probably already have a different model customized to the problem?
I interviewed for a company which used time series modelling to predict engine failure on ships. Though they didn't call it that, they called it just "AI" and weren't familiar with the time series literature at all. (Neither was I, but I did at least know that the time series literature was a thing.)
They just used a bunch of sensors attached to the engine.
That seems like a scenario where there wouldn't be external sources of information capable of improving your predictions. It also seemed to me like a scenario where good old fashioned statistical methods would outperform LSTMs, which was what they used.
(I didn't get the job - I was offered it, but they rescinded their offer at the last moment, supposedly because their investors had insisted they should hire a web developer instead.)
I appreciate this analysis, but I wonder, setting aside unpredictable current events for the moment, aren’t seasonal and scheduled events also covariates? eg forecasting baseball scores year round?
I'm doing time series forecasting for a biological system. I haven't used these foundation models but my experience is that knowledge of how things are related is baked into what variables you put into the model (which can be more than raw data, you can make any derived product you think would be useful)
but it's up to the model to figure out how they are related (if I knew exactly how they were related I wouldn't be training a forecasting model).
I'm sure someone who actually knows this stuff has a better easy to view it but my impression was that these time series foundation models aren't trained for a specific dataset but more they are trained on a procedure for how to look at datasets and make predictions. It doesn't seem to lump things into a bucket much more than any other model type (which also are just consistent procedures to look at datasets). It's not like you have radically different model types across different fields.
I can see another use for time-series generative models; to create the right shape of noisy data.
To improve the robustness of machine learning models, artificial noise is often introduced into actual training data. Something like Timestral could simplify the process of generating it.
I mean their latest model is gated behind API pricing that's more than enough to prevent any pupil from using it. I did not try ML4 (I'll wait till the end of the preview at least) so I can't tell anything about its performance outside of benchmarks, but on paper it's basically at the level of Muse Spark Open and I've not seen nearly as many people spitting on that model.
A lot of misconception around the utility of foundational time series models on HN.
Sure, you could train a XGboost model on a time series with holiday or event covariates, but what if you have 10k different series to forecast for?
You could train 10k different models, one for each series, but then they have very little history to learn from. Alternatively, you could train a single generic model to cover all 10k, but then you have to deal with scale mismatch between series in addition to the feature dev overhead between series. A TSFM saves a ton of work in modeling each of these series and could potentially generalize better to produce more accurate forecasts.
What's the input like? Does it incorporate language understanding, so that each time series has a natural language explanation of the series and all the data fields? If it takes in really descriptive text like that I can see the opportunity for generalization. If it's just a time series of non-descript numbers I see much much less opportunity for cross-series learning.
All the businesses with product catalogs for example, every product can have several time series associated with it.
If a business has 10k employees, you have, at minimum, 10k time series to perform employee analyst operations over. There are several hundred companies in the US with 10k+ employees.
There are many more than 10k postal zipcodes in the US to perform forecasts over.
The difference is the same, whether you have 10k series representing the same variables or 10k series representing different variables decomposes into the same forecasting problem that you can either over-generalize across all time-series, producing worse accuracy at the tails, or train a model for each individual forecast, producing worse accuracy in general due to limited data input.
What kind of infrastructure do you use for that scale? Are these computationally light ARIMA models?
We're running 150k models per day and while I'm sure I could scale this to running them every 30 minutes, it would mostly be me throwing money at the problem to parallelize the compute.
But why is the question, I run about 400k models every 30 minutes at work. I have tried all other approaches, but xgboost model per timeseries has never been beaten before. Even if I wasn't targetting accuracy, Xgboost loading a pretrained model and predicting with it and still faster.
This is the pseudo-intellectual version of your boomer relative ranting about the weatherman on TV, “how does this guy still have a job when he’s always wrong!”
Or the poll-bashing that happened after the 2016 election when an outcome with a 29% chance per one model was observed.
Any interesting forecasting problem is underconstrained. Monte Carlo methods are the classic way to mitigate that one-universe problem.
Nobody can predict the future with certainty. All models are wrong, but some are useful.
Monte Carlo methods are used to estimate the risk associated with a prediction (and presumably an action taken against that prediction). Does that answer your question? I can elaborate if you like.
It estimates risk based on your current understanding of the universe.
1. No guarantee that the universe tomorrow will behave as the distributions you used in your Monte Carlo today.
2. If the universe stats serving you 10 unlikely events in a row, is your model model wrong, or just 10 unlucky draws? If your model was correct and you start "correcting it" based on these 10 draws, you now made it wrong.
Like the other person said, all models are wrong but some are useful (a quote they did not make up btw). It sounds like you understand how MC works. Do you feel like you've answered your own question regarding how MC helps? I'm trying to be helpful since you asked above.
To meaningfully test a forecasting model, use a data-generating process with a high signal-to-noise ratio where a classical method (empirical conditional distributions, i.e., counting) can capture most of the signal. Then check whether the model captures that signal too. If you only test on high-noise processes, any model looks comparable to SOTA because there's little signal to differentiate them. Also testing on a range of signals with different signal-to-noise ratio would allow us to know how much of the signal the model captures in different scenarios.
My question: What kind of dynamical system observability does Timestral provide? Is it a blackbox y = F(x, params) model for offline use? Can it provide the actual dynamical system equations? Or are you stuck forever querying the Timestral server (which is bad, think about if you're grid system operator doing day ahead, month ahead forecasts, and Timestral goes down)
Typically, in linear, non-linear, and NN models, we fit some noisy data to A) An underlying model with unknown parameters or B) A model free search which attempts to construct combination of "basis" functions or C) A purely black box model which we only care about Input/Output error.
A) is useful because often we know an underlying model we want to match, but don't know the parameter values. If we can determine the parameters, we get insights about the real system. E.g, how are these two variables coupled and how strongly -> We can make business decisions about what happens if we adjust some parameters.
So what exactly does this timeseries foundation model do and provide?
Maybe, the observability is that you have a forecast, and you adjust the covariates and see what happens. But one would want much more insights than a "executive summary" of the forecast.
The important parts for me as a signals processing R&D engineer.
> Classical models such as ARIMA and linear regression remain well-established tools for forecasting time series. These models are usually fitted to a single series and learn only from that series' history.
> Deep learning models such as PatchTST add the capacity to learn nonlinear patterns across many related series, but they are still typically trained from scratch on each dataset, so what they learn doesn't carry over to new domains.
> Time series foundation models bring the pre-training approach to time series. Trained on a large and diverse corpus, our model learns patterns that transfer across series, forecasting zero-shot on data it has never seen, even in domains where data is scarce.
Edit: Looks like it's a pure black box which you can run on premise.
> Output: Probabilistic forecasts, 21 quantiles
> Inputs: Multivariate targets, past-only and known-future covariates
perhaps this is wrong, but to me at some abstract level LLMs feel like a more general class of time-series models. like a general purpose sequence model
edot | 10 hours ago
Speed looks nice too. I want an open weight model with this performance now.
OKRainbowKid | 10 hours ago
nelaggy | 10 hours ago
dudu24 | 8 hours ago
beernet | 10 hours ago
saithound | 10 hours ago
AtHeartEngineer | 10 hours ago
Honestly this might be very good for feature selection to refine a much more specialized model.
ritzaco | 10 hours ago
maelito | 9 hours ago
Edit : working with GLM5.3 now, it's also slow. Maybe Mistral's under heavy load.
Disclaimer : Not Cofounder of The Forecasting Company here
sspiff | 9 hours ago
Something is going on with the performance (as in speed of the service, not quality). Some days it's quite fast, some days it's so slow I interrupt it because I thought it was hanging.
I think they're shifting hosting capacity around, or otherwise messing with the model. Note that it's still a "preview" version, meaning they're still training it further for the final release and they are likely still figuring out the optimal hosting configuration as well.
It wouldn't surprise me to see further changes in both quality of the model and speed before the final release.
isidor_nikolic | 3 hours ago
<mistral vibe pm here>
pims | 8 hours ago
maelito | 6 hours ago
nozzlegear | 6 hours ago
I agree with you though. It's slow as hell, and so is their GLM offering.
maelito | 5 hours ago
nozzlegear | 5 hours ago
maelito | 4 hours ago
pims | 5 hours ago
isidor_nikolic | 3 hours ago
isidor_nikolic | 3 hours ago
<mistral vibe pm here>
data-ottawa | 3 hours ago
joefourier | 9 hours ago
> Known-future covariates. Some drivers of sensor behaviour are known in advance, such as production schedules, planned shutdowns, weather forecasts. Timestral can condition on them to improve forecast quality, so a scheduled load increase is expected rather than flagged as a surprise.
To me this should be the entire basis of the system, not a condition to improve forecast quality? Like if you're modelling electricity consumption, if all you have is a window of N previous samples, you're going to get completely beaten by a comparatively simple system that knows the time of day, that there's holidays coming up, the weather forecast, etc.
I would also think that these time-series forecasting models would gain from a sophisticated understanding at least the labels of the covariates? A 5M model has no idea what a "bank holiday" is, it just sees an arbitrary class go from 0 to 1, and won't have any useful inductive biases as to what it means or how to link them to other pieces of information.
E.g., wouldn't you predict €/MWh of electricity far more accurately with access to price of natural gas (itself influenced by many geopolitical factors), breakdown in composition of production of each energy source, predictions about roll-out (requiring a sophisticated LLM capable of understanding news and their implications), etc. What's the use in trying to predict it from 2 variables and a narrow window looking back at just 8192 or 32768 samples?
Or perhaps I misunderstand the use-case and this is more for extremely short term movements?
Y_Y | 9 hours ago
You're right but, if the time series looks like music then you can apply your music domain knowledge. You don't have to only apply knowledge that's general across every domain, part of the magic of large neural nets is that you can do recognition and prediction at the same time thereby using (statistically) things known from training about how to identify a sequence and then also how to continue it.
Sure adding outside info helps even more, but with enough data it should become unnecessary (or so goes the theory).
joefourier | 8 hours ago
Is there a reason you don't throw in something like cross-attention on an embedding of the sequence description/data labels? I've trained numerous generative models for audio and images and it's always the case that the unconditional model performs worse, trains slower, etc. than the conditional one (with the best performance from a hybrid using classifier-free guidance or similar scheme).
Again I don't have experience with the domain of time-series forecasting directly, but I would think there are generalities between autoregressive generative models and time-series forecasting, given that both involve predicting the next sample(s) given N regularly sampled past inputs.
ekabod | 4 hours ago
Won't you need to attribute a weight to the data coming from the model and the weight for your data?
jc4p | 4 hours ago
Would love to know actual time series foundation model use cases!
stanfordkid | an hour ago
justincormack | 8 hours ago
btchashpoint | 7 hours ago
jldugger | an hour ago
Problem is you really don't even need ML for the curve fitting tools. Just ARIMA or Holt Winters is enough and likely faster. Heck, Kubernetes HPA doesn't even implement full PID, just P.
It seems like the main challenge is going to be finding domains where accuracy is important enough to spend inference token money on, but not so important that you'd hire a team to build a custom model. Or maybe that team borrows Timestral weights to fine tune to the domain, but they probably already have a different model customized to the problem?
vintermann | 8 hours ago
That seems like a scenario where there wouldn't be external sources of information capable of improving your predictions. It also seemed to me like a scenario where good old fashioned statistical methods would outperform LSTMs, which was what they used.
(I didn't get the job - I was offered it, but they rescinded their offer at the last moment, supposedly because their investors had insisted they should hire a web developer instead.)
cwmoore | 7 hours ago
tdb7893 | 6 hours ago
I'm sure someone who actually knows this stuff has a better easy to view it but my impression was that these time series foundation models aren't trained for a specific dataset but more they are trained on a procedure for how to look at datasets and make predictions. It doesn't seem to lump things into a bucket much more than any other model type (which also are just consistent procedures to look at datasets). It's not like you have radically different model types across different fields.
sheepscreek | 5 hours ago
To improve the robustness of machine learning models, artificial noise is often introduced into actual training data. Something like Timestral could simplify the process of generating it.
amai | 9 hours ago
https://github.com/Nixtla/nixtla
colesantiago | 8 hours ago
No need to pay for proprietary TimeGPT.
pquerna | 8 hours ago
Calling it "Modified" MIT is just misleading.
kooi | an hour ago
Sounds like a great problem to have.
glimshe | 8 hours ago
With such strong human capital, both in quantity and quality, Europe should in theory be competitive with US and China in AI.
earthnail | 8 hours ago
glimshe | 5 hours ago
"Mistral is so bad, it's the only AI my school doesn't bother to block"
dadoum | 4 hours ago
aniviacat | 6 hours ago
logancbrown | 8 hours ago
Sure, you could train a XGboost model on a time series with holiday or event covariates, but what if you have 10k different series to forecast for? You could train 10k different models, one for each series, but then they have very little history to learn from. Alternatively, you could train a single generic model to cover all 10k, but then you have to deal with scale mismatch between series in addition to the feature dev overhead between series. A TSFM saves a ton of work in modeling each of these series and could potentially generalize better to produce more accurate forecasts.
articlepan | 8 hours ago
dominotw | 5 hours ago
who has these though? i cannot imagine somoone making sense of 10k timeseries.
probablypower | 5 hours ago
speedgoose | 4 hours ago
logancbrown | an hour ago
logancbrown | 4 hours ago
dominotw | 4 hours ago
logancbrown | an hour ago
niltecedu | 3 hours ago
magimas | 43 minutes ago
We're running 150k models per day and while I'm sure I could scale this to running them every 30 minutes, it would mostly be me throwing money at the problem to parallelize the compute.
niltecedu | 3 hours ago
logancbrown | an hour ago
ottomengis | 8 hours ago
rubenfiszel | 7 hours ago
whatever1 | 6 hours ago
Which is why at the end of the day why nobody trusts or blames the forecasting models, specially in all or nothing types of decisions.
sh34r | 5 hours ago
Or the poll-bashing that happened after the 2016 election when an outcome with a 29% chance per one model was observed.
Any interesting forecasting problem is underconstrained. Monte Carlo methods are the classic way to mitigate that one-universe problem.
Nobody can predict the future with certainty. All models are wrong, but some are useful.
whatever1 | 5 hours ago
I fully agree that all models are wrong.
Their main usage is just replaying the past or encoding the modelers assumptions for the future.
But there is no way to know if a model is good or bad.
There would be if we had infinite universes.
tom_mc | 2 hours ago
whatever1 | 2 hours ago
1. No guarantee that the universe tomorrow will behave as the distributions you used in your Monte Carlo today.
2. If the universe stats serving you 10 unlikely events in a row, is your model model wrong, or just 10 unlucky draws? If your model was correct and you start "correcting it" based on these 10 draws, you now made it wrong.
tom_mc | 12 minutes ago
alex7o | 6 hours ago
davedx | 6 hours ago
dingaling911 | 6 hours ago
swalsh | 5 hours ago
LelouBil | 4 hours ago
dominotw | 5 hours ago
lol. so what?
browningstreet | 3 hours ago
Trusteando | 5 hours ago
AndrewKemendo | 5 hours ago
Spacetimeformer was imo a huge leap forward that got overshadowed by all the language model progress
Excited to see more effort put into time series forecasting
kooi | 4 hours ago
Typically, in linear, non-linear, and NN models, we fit some noisy data to A) An underlying model with unknown parameters or B) A model free search which attempts to construct combination of "basis" functions or C) A purely black box model which we only care about Input/Output error.
A) is useful because often we know an underlying model we want to match, but don't know the parameter values. If we can determine the parameters, we get insights about the real system. E.g, how are these two variables coupled and how strongly -> We can make business decisions about what happens if we adjust some parameters.
So what exactly does this timeseries foundation model do and provide?
Maybe, the observability is that you have a forecast, and you adjust the covariates and see what happens. But one would want much more insights than a "executive summary" of the forecast.
The important parts for me as a signals processing R&D engineer.
> Classical models such as ARIMA and linear regression remain well-established tools for forecasting time series. These models are usually fitted to a single series and learn only from that series' history.
> Deep learning models such as PatchTST add the capacity to learn nonlinear patterns across many related series, but they are still typically trained from scratch on each dataset, so what they learn doesn't carry over to new domains.
> Time series foundation models bring the pre-training approach to time series. Trained on a large and diverse corpus, our model learns patterns that transfer across series, forecasting zero-shot on data it has never seen, even in domains where data is scarce.
Edit: Looks like it's a pure black box which you can run on premise.
> Output: Probabilistic forecasts, 21 quantiles
> Inputs: Multivariate targets, past-only and known-future covariates
traceroute66 | 4 hours ago
"We have published a GitHub repository so you can run Timestral on your own data."
[1] https://github.com/mistralai/timestral
bojackh | 4 hours ago
MisterMunchkin | 57 minutes ago
vdfs | 40 minutes ago