# Discrete Benchmark

## Classic Control & Box2D Results

SLM Lab v5.2 validates algorithms on [Gymnasium](https://gymnasium.farama.org/) discrete environments using the TorchArc architecture. These benchmarks cover:

* [**Classic Control**](https://gymnasium.farama.org/environments/classic_control/): CartPole, Acrobot, Pendulum—simple physics tasks ideal for algorithm validation
* [**Box2D**](https://gymnasium.farama.org/environments/box2d/): LunarLander—2D physics with more complex dynamics

Results below are from February–March 2026 benchmark runs using Gymnasium v5 environments. TorchArc specs for existing algorithms; CrossQ uses standard MLP specs.

All trained models and metrics are publicly available on [HuggingFace](https://huggingface.co/datasets/SLM-Lab/benchmark).

### Methodology

Results show **Trial-level** performance:

1. **Trial** = 4 Sessions with different random seeds
2. **Session** = One complete training run
3. **Score** = Final 100-checkpoint moving average (`total_reward_ma`)

The trial score is the mean across 4 sessions, providing statistically meaningful results.

### Standardized Settings

| Category        | num\_envs | max\_frame | log\_frequency | ASHA grace\_period |
| --------------- | --------- | ---------- | -------------- | ------------------ |
| Classic Control | 4         | 2e5-3e5    | 500            | 1e4                |
| Box2D           | 8         | 3e5        | 1000           | 5e4                |

The `grace_period` is the minimum frames before ASHA early stopping can terminate underperforming trials.

{% hint style="warning" %}
**v5 vs v4 Difficulty:** Gymnasium environments have stricter termination and reward handling:

* **LunarLander-v3** is notably harder than v2—stricter landing criteria, lower typical scores
* **Pendulum-v1** uses different reward scaling than v0
* Expect **5-15% lower scores** compared to OpenAI Gym benchmarks

See [Gymnasium docs](https://gymnasium.farama.org/) for environment-specific changes.
{% endhint %}

### Running Benchmarks

**Local** - runs on your machine (Classic Control completes in minutes on CPU):

```bash
slm-lab run slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml ppo_cartpole_arc train
slm-lab run slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml ddqn_per_concat_lunar_arc train
```

**Remote** - cloud GPU via [dstack](https://dstack.ai), auto-syncs to HuggingFace:

```bash
source .env && slm-lab run-remote slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml ppo_cartpole_arc train -n cartpole
source .env && slm-lab run-remote slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml ddqn_per_concat_lunar_arc train -n lunar
```

Remote setup: `cp .env.example .env` then set `HF_TOKEN`. See [Remote Training](/slm-lab/using-slm-lab/remote-training.md) for dstack config.

{% hint style="info" %}
**GPU not required for Classic Control.** These environments train fast on CPU. Box2D (LunarLander) benefits from GPU but still runs fine locally.
{% endhint %}

### Download and Replay

```bash
# List all available experiments (requires HF_REPO=SLM-Lab/benchmark in .env)
source .env && slm-lab list

# Download a specific experiment
source .env && slm-lab pull ppo_cartpole_arc

# Replay the trained agent
slm-lab run slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml ppo_cartpole_arc enjoy@data/ppo_cartpole_arc_*/ppo_cartpole_arc_t0_spec.yaml
```

***

## Results

### Classic Control

#### CartPole-v1

[Docs](https://gymnasium.farama.org/environments/classic_control/cart_pole/) | State: Box(4) | Action: Discrete(2) | Target: >400

**Settings**: max\_frame 2e5 | num\_envs 4 | max\_session 4 | log\_frequency 500

| Algorithm | Status | MA     | SPEC\_FILE                                                                                                                                                         | SPEC\_NAME                          | HF Data                                                                                                                                                                         |
| --------- | ------ | ------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| REINFORCE | ✅      | 483.31 | [slm\_lab/spec/benchmark\_arc/reinforce/reinforce\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/reinforce/reinforce_arc.yaml) | reinforce\_cartpole\_arc            | [reinforce\_cartpole\_arc\_2026\_02\_11\_135616](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/reinforce_cartpole_arc_2026_02_11_135616)                     |
| SARSA     | ✅      | 430.95 | [slm\_lab/spec/benchmark\_arc/sarsa/sarsa\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sarsa/sarsa_arc.yaml)                 | sarsa\_boltzmann\_cartpole\_arc     | [sarsa\_boltzmann\_cartpole\_arc\_2026\_02\_11\_135616](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sarsa_boltzmann_cartpole_arc_2026_02_11_135616)        |
| DQN       | ⚠️     | 239.94 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml)        | dqn\_boltzmann\_cartpole\_arc       | [dqn\_boltzmann\_cartpole\_arc\_2026\_02\_11\_135648](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/dqn_boltzmann_cartpole_arc_2026_02_11_135648)            |
| DDQN+PER  | ✅      | 451.51 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml)        | ddqn\_per\_boltzmann\_cartpole\_arc | [ddqn\_per\_boltzmann\_cartpole\_arc\_2026\_02\_11\_140518](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ddqn_per_boltzmann_cartpole_arc_2026_02_11_140518) |
| A2C       | ✅      | 496.68 | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml)        | a2c\_gae\_cartpole\_arc             | [a2c\_gae\_cartpole\_arc\_2026\_02\_11\_142531](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_cartpole_arc_2026_02_11_142531)                        |
| PPO       | ✅      | 498.94 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml)        | ppo\_cartpole\_arc                  | [ppo\_cartpole\_arc\_2026\_02\_11\_144029](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_cartpole_arc_2026_02_11_144029)                                 |
| SAC       | ✅      | 406.09 | [slm\_lab/spec/benchmark\_arc/sac/sac\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_classic_arc.yaml)        | sac\_cartpole\_arc                  | [sac\_cartpole\_arc\_2026\_02\_11\_144155](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_cartpole_arc_2026_02_11_144155)                                 |
| CrossQ    | ⚠️     | 334.59 | [slm\_lab/spec/benchmark/crossq/crossq\_classic.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_classic.yaml)              | crossq\_cartpole                    | [crossq\_cartpole\_2026\_03\_02\_100434](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_cartpole_2026_03_02_100434)                                    |

![CartPole-v1](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/CartPole-v1_multi_trial_graph_mean_returns_ma_vs_frames.png)

#### Acrobot-v1

[Docs](https://gymnasium.farama.org/environments/classic_control/acrobot/) | State: Box(6) | Action: Discrete(3) | Target: >-100

**Settings**: max\_frame 3e5 | num\_envs 4 | max\_session 4 | log\_frequency 500

| Algorithm | Status | MA      | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME                   | HF Data                                                                                                                                                            |
| --------- | ------ | ------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| DQN       | ✅      | -94.17  | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml) | dqn\_boltzmann\_acrobot\_arc | [dqn\_boltzmann\_acrobot\_arc\_2026\_02\_11\_144342](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/dqn_boltzmann_acrobot_arc_2026_02_11_144342) |
| DDQN+PER  | ✅      | -83.92  | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml) | ddqn\_per\_acrobot\_arc      | [ddqn\_per\_acrobot\_arc\_2026\_02\_11\_153725](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ddqn_per_acrobot_arc_2026_02_11_153725)           |
| A2C       | ✅      | -83.99  | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_acrobot\_arc       | [a2c\_gae\_acrobot\_arc\_2026\_02\_11\_153806](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_acrobot_arc_2026_02_11_153806)             |
| PPO       | ✅      | -81.28  | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml) | ppo\_acrobot\_arc            | [ppo\_acrobot\_arc\_2026\_02\_11\_153758](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_acrobot_arc_2026_02_11_153758)                      |
| SAC       | ✅      | -92.60  | [slm\_lab/spec/benchmark\_arc/sac/sac\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_classic_arc.yaml) | sac\_acrobot\_arc            | [sac\_acrobot\_arc\_2026\_02\_11\_162211](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_acrobot_arc_2026_02_11_162211)                      |
| CrossQ    | ✅      | -103.13 | [slm\_lab/spec/benchmark/crossq/crossq\_classic.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_classic.yaml)       | crossq\_acrobot              | [crossq\_acrobot\_2026\_02\_23\_153622](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_acrobot_2026_02_23_153622)                         |

![Acrobot-v1](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/Acrobot-v1_multi_trial_graph_mean_returns_ma_vs_frames.png)

#### Pendulum-v1

[Docs](https://gymnasium.farama.org/environments/classic_control/pendulum/) | State: Box(3) | Action: Box(1) | Target: >-200

**Settings**: max\_frame 3e5 | num\_envs 4 | max\_session 4 | log\_frequency 500

| Algorithm | Status | MA      | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME              | HF Data                                                                                                                                                  |
| --------- | ------ | ------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| A2C       | ❌      | -820.74 | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_pendulum\_arc | [a2c\_gae\_pendulum\_arc\_2026\_02\_11\_162217](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_pendulum_arc_2026_02_11_162217) |
| PPO       | ✅      | -174.87 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml) | ppo\_pendulum\_arc      | [ppo\_pendulum\_arc\_2026\_02\_11\_162156](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_pendulum_arc_2026_02_11_162156)          |
| SAC       | ✅      | -150.97 | [slm\_lab/spec/benchmark\_arc/sac/sac\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_classic_arc.yaml) | sac\_pendulum\_arc      | [sac\_pendulum\_arc\_2026\_02\_11\_162240](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_pendulum_arc_2026_02_11_162240)          |
| CrossQ    | ✅      | -145.66 | [slm\_lab/spec/benchmark/crossq/crossq\_classic.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_classic.yaml)       | crossq\_pendulum        | [crossq\_pendulum\_2026\_02\_28\_130648](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_pendulum_2026_02_28_130648)             |

![Pendulum-v1](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/Pendulum-v1_multi_trial_graph_mean_returns_ma_vs_frames.png)

### Box2D

#### LunarLander-v3 (Discrete)

[Docs](https://gymnasium.farama.org/environments/box2d/lunar_lander/) | State: Box(8) | Action: Discrete(4) | Target: >200

**Settings**: max\_frame 3e5 | num\_envs 8 | max\_session 4 | log\_frequency 1000

| Algorithm | Status | MA     | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME                    | HF Data                                                                                                                                                             |
| --------- | ------ | ------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| DQN       | ⚠️     | 195.21 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml)     | dqn\_concat\_lunar\_arc       | [dqn\_concat\_lunar\_arc\_2026\_02\_11\_201407](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/dqn_concat_lunar_arc_2026_02_11_201407)            |
| DDQN+PER  | ✅      | 265.90 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml)     | ddqn\_per\_concat\_lunar\_arc | [ddqn\_per\_concat\_lunar\_arc\_2026\_02\_13\_105115](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ddqn_per_concat_lunar_arc_2026_02_13_105115) |
| A2C       | ❌      | 27.38  | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_lunar\_arc          | [a2c\_gae\_lunar\_arc\_2026\_02\_11\_224304](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_lunar_arc_2026_02_11_224304)                  |
| PPO       | ⚠️     | 183.30 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_box2d_arc.yaml)     | ppo\_lunar\_arc               | [ppo\_lunar\_arc\_2026\_02\_11\_201303](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_lunar_arc_2026_02_11_201303)                           |
| SAC       | ⚠️     | 106.17 | [slm\_lab/spec/benchmark\_arc/sac/sac\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_box2d_arc.yaml)     | sac\_lunar\_arc               | [sac\_lunar\_arc\_2026\_02\_11\_201417](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_lunar_arc_2026_02_11_201417)                           |
| CrossQ    | ❌      | 139.21 | [slm\_lab/spec/benchmark/crossq/crossq\_box2d.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_box2d.yaml)           | crossq\_lunar                 | [crossq\_lunar\_2026\_02\_28\_130733](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_lunar_2026_02_28_130733)                              |

![LunarLander-v3 Discrete](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/LunarLander-v3_Discrete_multi_trial_graph_mean_returns_ma_vs_frames.png)

#### LunarLander-v3 (Continuous)

[Docs](https://gymnasium.farama.org/environments/box2d/lunar_lander/) | State: Box(8) | Action: Box(2) | Target: >200

**Settings**: max\_frame 3e5 | num\_envs 8 | max\_session 4 | log\_frequency 1000

| Algorithm | Status | MA     | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME                       | HF Data                                                                                                                                                                   |
| --------- | ------ | ------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| A2C       | ❌      | -76.81 | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_lunar\_continuous\_arc | [a2c\_gae\_lunar\_continuous\_arc\_2026\_02\_11\_224301](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_lunar_continuous_arc_2026_02_11_224301) |
| PPO       | ⚠️     | 132.58 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_box2d_arc.yaml)     | ppo\_lunar\_continuous\_arc      | [ppo\_lunar\_continuous\_arc\_2026\_02\_11\_224229](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_lunar_continuous_arc_2026_02_11_224229)          |
| SAC       | ⚠️     | 125.00 | [slm\_lab/spec/benchmark\_arc/sac/sac\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_box2d_arc.yaml)     | sac\_lunar\_continuous\_arc      | [sac\_lunar\_continuous\_arc\_2026\_02\_12\_222203](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_lunar_continuous_arc_2026_02_12_222203)          |
| CrossQ    | ✅      | 268.91 | [slm\_lab/spec/benchmark/crossq/crossq\_box2d.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_box2d.yaml)           | crossq\_lunar\_continuous        | [crossq\_lunar\_continuous\_2026\_03\_01\_140517](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_lunar_continuous_2026_03_01_140517)             |

![LunarLander-v3 Continuous](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/LunarLander-v3_Continuous_multi_trial_graph_mean_returns_ma_vs_frames.png)

**Legend:** ✅ Solved | ⚠️ Close (>80%) | ❌ Failed

***

## Historical Results (v4)

<details>

<summary>OpenAI Gym Results (v4) - click to expand</summary>

{% hint style="info" %}
These results from SLM Lab v4 used OpenAI Gym environments (now deprecated). Environment versions differ from current Gymnasium versions. Unity environments are no longer included in the core package.
{% endhint %}

* [Upload PR #427](https://github.com/kengz/SLM-Lab/pull/427)
* [Google Drive data](https://drive.google.com/file/d/1lb9Hn22Uzb67ndotRULiUspk3TnmKp6Z/view?usp=sharing)

| Env. \ Alg. | DQN   | DDQN+PER | A2C (GAE) | A2C (n-step) | PPO       | SAC     |
| ----------- | ----- | -------- | --------- | ------------ | --------- | ------- |
| Breakout    | 80.88 | 182      | 377       | 398          | **443**   | 3.51\*  |
| Pong        | 18.48 | 20.5     | 19.31     | 19.56        | **20.58** | 19.87\* |
| Qbert       | 5494  | 11426    | 12405     | **13590**    | 13460     | 923\*   |
| Seaquest    | 1185  | **4405** | 1070      | 1684         | 1715      | 171\*   |
| LunarLander | 192   | 233      | 25.21     | 68.23        | 214       | **276** |

> Episode score at the end of training. Reported scores are the average over the last 100 checkpoints, averaged over 4 Sessions. Results marked with `*` used async SAC.

</details>

For the full Atari benchmark, see [Atari Benchmark](/slm-lab/benchmark-results/atari-benchmark.md).


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://slm-lab.gitbook.io/slm-lab/benchmark-results/discrete-benchmark.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.