Estimate it

One rule that a lot of DMs forget is that not every single combat encounter—or more specifically, every single moment within combat—needs to be handled in the Combat-Initiative Abstraction. A group of level 5 characters facing off against a single CR 1/4 goblin doesn't need to open with Initiative, and similarly, if a group of characters are up against any encounter whose difficulty is trivial to surmount (or they've whittled an encounter down to this point), you shouldn't feel compelled to mechanically iterate through each moment of that combat. Just use a single d20 roll (maybe a contested ability check?) to determine how many hit points are used up.

Surrender!

It depends on what kinds of creatures the players are up against, but most NPC creatures possess a modicum of intelligence and sense of self-preservation. If they see that the fight has clearly gone south for them, it's reasonable for them to surrender or run away, eliding what would otherwise be the grueling remainder of the fight.

Like in the example above: instead of narrating that the level five characters mercilessly brutalize the goblin, maybe the goblin just immediately surrenders without putting up a fight—or hopefully the players didn't intend ill will towards the goblin in the first place!

There's direct precedent in some adventure modules for this as well. For example, in the Mina Perdida de Phandelver adventure, the very first encounter in the module suggests this course of actions for the goblins the PCs encounter:

Four goblins are hiding in the woods, two on each side of the road. They wait until someone approaches the bodies and then attack.

This will likely be the first of many combat encounters in the adventure. Here are the steps you should follow to run it effectively:

• [...]

• When three goblins are defeated, the last goblin attempts to flee, heading for the goblin trail.

—Goblin Ambush, Lost Mines of Phandelver, D&D 5th Edition Starter Set

Having said that...

I do need to slightly challenge the premise of this question. It's one thing if you're only expecting players to lose a few more hit points to a lucky hit by their enemies, but if you're expecting players to still use up spell slots, that implies pretty strongly that the outcome of the fight is very much não set in stone. Even at high levels, a spellcaster would probably prefer to use their cantrips if they don't expect to necessidade their spell slots.

So while the above advice is, in my opinion, a pretty good rule to follow, it should be tempered with the advice that if the players can be expected to expend expensive resources as part of their effort to end the conflict, it's probably too early to call it.

Offer bartering resources to solve encounters.

I once had an RP-table that wanted to try how a "real" dungeon feels like so I prepared a complicated dungeon with multiple complex traps, puzzles and several creature-encounters.

The whole dungeon was an attrition-based set-up of multiple dungeon layers. The group had a lot of fun solving puzzles and traps, but then there were the creature-encounters.

After three creature-encounters, it was absolutely clear that the group rather wanted to do their usual low creature-encounter play-style. So we decided to take a pause, it was time for food anyway, and we took that time to talk about how we could salvage the problem.

Should we delete the creature-encounters and destroy the careful attrition-balance? This was not an option as we still wanted that "real" dungeon experience.

So instead, we came up with a solution of trading resources for encounters. The group could decide once the initiative was rolled (after set-ups, sneaking around and surprise rounds) to trade spell slots, hp and ability use to "auto-solve" the creature-encounters.

We did this by bartering after the decision was made to solve an encounter by spending resources - I told them the creatures involved and their CRs. This turned out to be a more enjoyable and less grindy experience. I doubt that it works at tables that aren't in the mood to get silly and creative.

The trades worked like this: I present the creatures and the group meta-games solutions and offers me resources that they deem appropriate, I accept or demand more (never less). The mood was less serious and more playful an example:

For the enemy with invisibility, we offer you a pound of flour, and we all should probably make a Dex-save and take half of the dust explosion, that dmg is your call.

For the narrative, I improvised narrating the last actions of the encounter based on the resources spent.

So if your table is more into creative solutions and bartering, then a trade can save time and make the time spend on creature-encounters more enjoyable for your Rp-heavy group.

A few caveats, I can easily see this grow tedious in a rules lawyer heavy-group and only recommend this solution if it suits your table.

This bartering system also heavily relies on estimations and the willingness of your players to offer their resources, they may become more stingy if they feel like they have already solved the encounter. Especially after they realise the rest of the encounter which robs the mystery of hidden enemies etc. so be open about communicating your intentions clearly - that includes that you expect your group to actually offer resources even if the rest of the encounter seems trivial.

"After a few more blows and cantrips, the defeated enemies lie dead at your feet. What do you want to do now?"

Once the narrative question of the scene has been answered the scene is over so transition to the next one. Scenes in a role playing game really only have three purposes: exposition, character development or conflict resolution. Guess which one a combat encounter primarily is?

Once the conflict has been resolved (i.e. the PCs will obviously win), the encounter is over. When that happens it's no longer a combat - it's mopping-up. Unless that involves exposition, character development or conflict resolution there’s no point in playing it out.

As for the issue of losing additional resources - if the battle is obviously won, players will not spend limited resources like spell slots. Similarly, any hit points that might be lost in the mopping-up will be trivial. Otherwise, the battle isn’t obviously won, is it?

A few ideas I often implement that I picked up from various GMs and websites. Some of these involve retreating, but I'll list them anyways to be complete.

Most of these kick in (initiate, or are triggered) when either the enemy leader dies, or 1/2 of the enemy units (round up) are defeated. When one of those two happen:

1. All enemy units make a save (whatever you feel is fitting). If they fail they check, they surrender, flee, or die/vanish (whatever you feel is fitting for the NPC). Some NPCs (based on type or nature) won't make this check, but >2/3 will.

2. When there are less than 1/2 of the enemies left, enemies reduced to X% of their HP are instantly defeated. (Normally I go with 20%)

3. Damage-dealing critical hits instantly defeat "normal" enemies, no matter the remaining HP.

A last option I often use is grant an additional enemy killed per turn if the player gives a sufficiently good/entertaining (mostly: did the table enjoy it? They normally do) tale of how they killed the enemy.

If the rules don't provide you the pacing you want, change the rules. Either of the game, or the encounter.

Your problem is that fights start out exciting and dynamic, and peter out to pedestrian.

There are games where this doesn't happen. In 13th age, there is an escalation die that steadily increases the stakes of combat (as its most passive effect, everyone gets a +X to all attacks on turn X of the combat). In Wushu, early parts of a scene involve clearing out the threats; only once the threats are suppressed can you narrate your awesome victory.

Both games have mechanics that push the "drama spike" to the end of combat.

D&D 5e, by default, has combat-as-attrition when there is more than one foe.

Threat is maximum when all the foes are standing. Once a few foes have been dropped, the threat plummets. If (in a simple model) the PCs kill foes at a constant rate per round, and the foes threaten in proportion to their numbers, then when half of the foes are dead you have already experienced 3/4 of the encounter's threat. But the encounter is only half over.

*
* *
* * *
* * * *
* * * *|*
* * * *|* *
* * * *|* * *
* * * *|* * * *

8 foes, 1 dead/round, each does * damage per round. The | is the half way point of the combat. You'll take 10 * damage after the half way point, and took 30 * damage before hand.

One approach is to work out roughly where that point is, and make a two-phase combat.

*        |
* *     *|
* * *   *|*
* * * * *|* *
* * * * *|* * *
* * * * *|* * * *
* * * * *|* * * * *

here is a 10 round combat where reinforcements arrive at round 5. Left of the line is 28 threat, right is 15; 35% of the threat is after the half way point, instead of 25%.

Another approach is to make foes flee or break when half are defeated:

*
* *
* *|*
* *|* *
* *|* *
* *|* *
* *|* *
* *|* *

here, 15 total threat before the half way point, 11 after. 42% of the threat is after the half way point.

Note that the "total threat" of these cases is menos per unit monster, but as the DM you control how many monsters there are and how they behave.

In a typical situation, N monsters are (N+1)(N)/2 times more total threat than 1 monster.

If N monsters "break" at 50% and are no longer threatening (they flee, surrender, etc), they generate 3/4 of that (total) threat. But the encounter excitement curve is much flatter.

For each encounter, introduce an element to keep the encounter flatter.

In a "boss fight", use threat dice; at the start of turn 2, place a 1d4 on the table with a 1; everyone takes a -1 penalty to saving throws and gets a +1 bonus to attack rolls. Each round increase it by 1.

That (like in 13th age) causes the stakes to raise each round.

Against creatures with morale, increase their numbers by 25% and state that they will break and when 50% have dropped.

Against creatures without morale (undead hordes?), make waves or a monster-flow standard.

As a concrete example, instead of 8 zombies, have 4 zombies to start, but then a flow of additional zombies arriving (3 per round for 3 rounds might be enough to make the "omg we are screwed" moment move from round 1 to round 3).

Waves, morale and escalation dice (and, more generally, escalation mechanics) are all time tested ideas, and they all solve your problem.

From one of your comments:

My players often take out one enemy at a time instead of bringing all enemies' health down gradually. As a result, the last few rounds are spent fighting with just one or two enemies, but they're close to full health. The enemies are too few to do much damage, but healthy enough to take several rounds to take down.

It sounds like you're making this problem for yourself: you've set the stage so every enemy has lots of health but is not much threat.

Slash their health and increase their numbers:

• Trash mobs should go down in a hit or two: the majority of the enemy combatants are there to prevent the party from immediately ganging up on the leader: their strength is numbers, not individual hp.
• Leaders should be beefy. If a leader is still up they represent a threat in their own right. Maybe they have a lieutenant who has some health, but most of their minions should not.

When your players realize they don't need to gang up on every enemy, they'll spread the love in order to reduce the threat as quickly as possible. This is doubly true if the enemy is not in one grand melee, but spread out.

I've been running a 4-person path for 6 experienced players and, in general, doubling the number of enemies (rather than their strength) plus a bump for the leader has kept the challenges well fitting.

Now the end of the fight is either with the leader, who still has the ability to put someone in the ground even if he's unlikely to win, or, if they killed the leader while some of his minions still stand, only takes a round or two to mop up the trash.

tl dr dr: give the health to those who can use it, not the minions.

Have the brave or panicked monsters take risks

As I mentioned in my answer to Como posso acelerar o combate?, one alternative to having enemies break morale and flee is to have them start fighting more aggressively—perhaps they are either brave, and start acting more heroically, or panicked, and start making mistakes.

Either way, this recklessness will cause the monsters to take more damage themselves, and end the fight more quickly. For example:

• Pushing past characters to try and enter a flanking position, provoking opportunity attacks
• Leaping from tall places to deal extra damage, but taking damage themselves in the process
• Throwing away their shield to wield their weapon two-handed, in order to deal more damage
• Switching from ranged to melee weapons
• Switching from safe ranged spells to dangerous close-up spells
• Using area effect spells even though their allies are within the spell's area of effect
• Triggering dangerous effects like lighting fire to a rope bridge
• Attempting to sacrifice themselves to drag PCs off edges or into pits

This does increase the damage PCs take per-round, but since it kills monsters more quickly, the monsters get fewer rounds, so it evens out. Fewer rounds ends combat more quickly, and as monsters are eliminated, combat rounds go faster.

Players generally perceive it to be fair, since it's within the normal ability of a monster to change its tactics to suit the situation. The perceived increased risk to the players increases drama and tension at what is normally the boring part of a fight. The DM isn't perceived to be going easy on the players, as they might if the enemies simply surrendered or stopped making attacks.

"Overwhelmed" status grants advantage

The DM can give Advantage to the players whenever it seems warranted. It's within your rights to declare that a grossly overpowered or outnumbered enemy group is "overwhelmed", a non-standard status effect giving attacks against them Advantage and all saving throws Disadvantage. This eliminates some miss chances, so combat goes faster. It can also be used against the party when you want to signal that the enemy they are facing is intentionally way too powerful and they should flee.

Consider the narrative wrap-up method

Personally, although you mentioned a dislike of this approach, my preferred method is to simply wrap up the combat narratively. Describe how the combat ends with the enemies clearly unable to mount any practical resistance, cowering, scattered, or overwhelmed.

The PCs save a few resources (hit points, spells), but in practice I find this doesn't unbalance the game much. The benefit of saving significant amounts of time in each combat is more important, in my opinion, especially with large groups, slow combat, or impatient players.

I've seen this used in D&D 4e, where the players were often billed a healing surge (roughly equivalent to one hit die in D&D 5e) for the privilege of ending a combat soon, as a kind of abstraction of the hit points and resources you would have spent by running the fight to the finish. Unfortunately, players would sometimes prefer to risk the continued fight than pay the cost. Hence I suggest making the wrap-up automatic.

Although this approach would require more up-front effort and tedious recordkeeping, it might fit your bill:

Run a combat encounter to the point where victory is basically inevitable, but keep track of the average damage done to the players, the number of rounds combat lasted so far, and resources (like spell slots) used. When you have your data, play your combat to the 50% point as usual. Then you can offer to conclude the combat for 1/4 to 1/8 of the resources consumed prior to that point. The 1/4 to 1/8 consumption is to account for the proportional decline in the strength of the enemy forces, though the exact modifier you can use is up to interpretation. Err on the side of less resources consumed to make the proposition more palatable.

Or you could do an approach similar to what Akixkisu suggested, but instead of having the players offer resources to be consumed, the GM offers a general "ballpark estimate" deal. "You are clearly on the path to victory, but you can guarantee success by each taking 5hp of damage, using one damage-causing item, or sacrificing one of your lowest remaining spell slots."

For purposes of full disclosure, I have not tried the first approach because it's too much for me to keep track of while trying to keep combat running quickly. I have used the latter approach, both as player and GM, to end the occasional combat encounter. It's still kind of clunky to employ, but it can save time in the long run. My sessions don't usually get to that level of granularity, though, so your mileage may vary.

I haven't had any problems with players objecting to stuff like this, so I don't have a good answer for what to do if half the party wants to keep fighting and the other half wants to auto-win. As long as the players feel like there is some value in ending the combat early and have the option to choose what penalty they incur, it seems to go smoothly.

Discuss it with your players

You have options: you've already noted that you can have enemies flee or fight to the death, and several others have already posted excellent alternatives here. Let your players know that you have a lot to cover during the session, so you want to speed things up. Let them also know that you want them to lose resources in the process. Ask them in advance how they'd prefer you to handle it, and describe some of the options. You don't need to necessarily do what they say, but you can get an idea of how detailed or abstracted they would like it to be.

Note that this will work better with more experienced players.

Outras opções

These are inspired by the excellent answer by @Akixkisu.

Planned Trade Packages

When planning each encounters, also plan a custom trade or two to offer the players when they reach a point that you determine. You can be as detailed as you like when you design the trade offers, even calculating it to the gold piece based on XP gains and resources used up per encounter of a given CR if you like. You can have one or more triggers plus one or more trade offers.

Exemplo:

IF there are only 3 of the 10 minions remaining, OR the semi-boss plus one minion (desencadeia), ENTÃO tell the party that you'll fast-forward through the combat if they accept the following: each player rolls a raw 1d20 and if they roll too low, they take 2d6 damage and spellcasters lose a 3rd level spell slot.

You could say that they choose one or two Outcomes (see below) per remaining enemy.

Simplified Bartering

(Note: this is intended to streamline the bartering system suggested by @Akixkisu.)

You determine during the session (or in advance) how many "units" the party must pay to fast-forward the combat. You also create a table in advance that quantifies how many "units" various things are worth and share it with the party. They then get to decide which of their resources they'll "pay" in order to resolve the encounter.

The Bartering Table might look something like the following:

• Damage: 1 "unit" per 5 damage received by a character
• Ammo: 1 "unit" per 2 ammo
• Spells: 1 "unit" per spell level
• Scrolls: 2 "units" per spell level
• Potions: 3 "units" per spell level
• Class Ability Uses: [you'll have to determine this based on your party's abilities!]
• Magical Item Uses: [you'll have to determine this based on your party's items!]
• Conditions (paralyzed, poisoned, exhausted, etc.): [GM's discretion]
• Capture the enemies alive: costs N extra units per enemy! :-)

Example: If there are 2 of 6 CR3 skeletons left in combat, you might ask the party to pay — say — 10 units.

Randomized Attrition

You create a table of outcomes in advance that are indexed by a dice roll or a card draw. Each player rolls a dice or draws a card and gets the outcome associated with that result. This is like a Deck of Many Things, but customized to your party and the kinds of encounters they're likely to face.

Somebody should make a series of open-source tables for these, each appropriate for different party levels, and publish them online. ;-)

Variações

• Escala: You could design a table of grande Outcomes intended for the whole party (one or two outcomes per encounter that affects everyone), OR a table of menor Outcomes intended for individual characters (each character rolls for their Outcome).
• Aplicabilidade: If an Outcome doesn't apply to a character, you could let them ignore that Outcome or you could force them to get a different Outcome instead.
• Sinergias: if they draw multiple Outcomes, then you can include things that modify other Outcomes. Examples: 1) double damage for tanks, 2) double the quantity of another Outcome of their choice, 3) draw two more Outcomes, 4) they choose a different Outcome to ignore, 5) they choose somebody else's Outcome to ignore, etc.
• Quantidade: the party rolls a single (e.g.) 1d6 and receives that many small Outcomes.
  • You could roll to see who gets the Outcomes they pick: 1-->the barbarian, 2-->rogue, 3-->warlock, 4-->cleric, etc.
  • You could let them distribute the Outcomes amongst themselves however they decide. This gives them the ability to make sacrifices for each other as they would in real combat.

Exemplos

Many of the following outcomes can be repeated in your table several times, saving you time. Clearly, these should be adjusted for the challenge rating and the nature of the encounter.

You could write up generic Outcome tables (meant to be usable throughout your campaign), adventure-specific Outcome tables (for a particular dungeon or region with common types of threats), or encounter-specific Outcome tables (

Whole-Party Outcomes:

• The enemies run for their lives! (If the party decides to pursue, make a "speed check", and if the party catches up, draw another Outcome.)
• You slay them all, without breaking a sweat!
• They surrender. Party chooses how to respond.
• You notice that one of the "slain" enemies is actually still alive -- mal!
• They trip and fall down the nearby cliff, presumably to their deaths
• The party takes 10d6 damage (distributed however they choose)
• Each party member rolls 1d20. Whoever rolled lowest takes 6d6 damage, and the next-lowest roller takes 3d6 damage.
• Each character takes 3d8 damage.
• Each character takes 1d4 damage.
• A piece of equipment per character, chosen randomly from each character's inventory, becomes damaged and unusable. (You decide what kinds of stuff it affects.)
• The party loses 1d4 potions (their choice)
• The party loses spell slots with a total level equal to 3d6 (players' choice)
• Casters each lose 1d4 spell slots (3rd level or higher, their choice), ranged fighters lose 1d6 ammo, and melee fighters take 2d10 damage.

Individual Outcomes:

• Describe how you defeat one of the remaining enemies (great for role players!)
• Take 2d4 damage
• Take 5d6 damage.
• Your mount receives 3d6 damage
• Your familiar is knocked unconscious
• Lose 2d6 ammo
• Lose 1d4 spell slots or ability uses above 1st level
• Lose 1d4 spell slots or ability uses above 4th level
• Lose 1d2 potions and 1d2 scrolls (their choice)
• You become poisoned or diseased (DM's choice of poison/disease)
• You lose the use of an appendage or have disadvantage when using it for the next 1d4 days (e.g. sprained ankle, broken arm)
• Your weapon malfunctions/breaks and needs repair
• You are knocked hard, and are stunned for 10 x (1d6 minutes)
• You get one more level of Exhaustion
• Lose 1d2 of a given stat (temporary; intended for higher levels)
• You become Mad (DM decides how, based on the encounter; intended for higher levels)