Datasets:
ability stringclasses 1
value | backend stringclasses 2
values | data_source stringclasses 1
value | dataset_index int64 0 99 | elapsed_seconds float64 43.9 15.5k | final_response stringclasses 1
value | index int64 0 3.2k | max_model_len int64 262k 262k | max_tokens int64 16.4k 16.4k | messages listlengths 5 43 | model stringclasses 1
value | problem_id stringclasses 100
values | problem_name stringclasses 100
values | reference_code stringclasses 100
values | rollout int64 0 15 | stop_reason stringclasses 1
value | temperature float64 1 1 | top_p float64 0.95 0.95 | turns listlengths 1 17 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 373.069712 | KERNELGYM_FINAL | 16 | 262,144 | 16,384 | [
{
"content": "You write custom Triton kernels to replace the pytorch operators in the given architecture to get speedups. \n\n You have complete freedom to choose the set of operators you want to replace. You may make the decision to replace some operators with custom Triton kernels and leave others unchange... | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | import torch
import torch.nn as nn
class Model(nn.Module):
"""
A model that performs a transposed 3D convolution, clamps the output to a minimum value,
and then divides the result by a constant.
"""
def __init__(self, in_channels, out_channels, kernel_size, stride, padding, min_value, divisor):
... | 0 | model_final | 1 | 0.95 | [
{
"elapsed_seconds": 266.8030159473,
"evaluation": {
"compiled": true,
"completed_at": null,
"correctness": true,
"decoy_kernel": false,
"error_code": null,
"error_message": null,
"kernel_runtime": 6.01,
"metadata": {
"backend": "cuda",
"correc... |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 1,916.658511 | KERNELGYM_FINAL | 17 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 1 | model_final | 1 | 0.95 | [{"elapsed_seconds":1038.2887003422,"evaluation":{"compiled":false,"completed_at":null,"correctness"(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 1,067.943249 | KERNELGYM_FINAL | 18 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 2 | model_final | 1 | 0.95 | [{"elapsed_seconds":908.8781988621,"evaluation":{"compiled":true,"completed_at":null,"correctness":t(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 1,105.041174 | KERNELGYM_FINAL | 19 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 3 | model_final | 1 | 0.95 | [{"elapsed_seconds":1084.9613864422,"evaluation":{"compiled":true,"completed_at":null,"correctness":(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 846.717401 | KERNELGYM_FINAL | 20 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 4 | model_final | 1 | 0.95 | [{"elapsed_seconds":621.465259552,"evaluation":{"compiled":true,"completed_at":null,"correctness":tr(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 832.637697 | KERNELGYM_FINAL | 21 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 5 | model_final | 1 | 0.95 | [{"elapsed_seconds":580.1658945084,"evaluation":{"compiled":true,"completed_at":null,"correctness":t(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 954.86949 | KERNELGYM_FINAL | 22 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 6 | model_final | 1 | 0.95 | [{"elapsed_seconds":842.7015314102,"evaluation":{"compiled":true,"completed_at":null,"correctness":t(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 868.527015 | KERNELGYM_FINAL | 23 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 7 | model_final | 1 | 0.95 | [{"elapsed_seconds":693.4257361889,"evaluation":{"compiled":true,"completed_at":null,"correctness":t(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 567.670671 | KERNELGYM_FINAL | 24 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 8 | model_final | 1 | 0.95 | [{"elapsed_seconds":348.9681723118,"evaluation":{"compiled":true,"completed_at":null,"correctness":t(...TRUNCATED) |
kernel_optimization | cuda | kernelbench_level2_validation | 0 | 1,620.673129 | KERNELGYM_FINAL | 25 | 262,144 | 16,384 | [{"content":"You write custom Triton kernels to replace the pytorch operators in the given architect(...TRUNCATED) | zai-org/GLM-5.2-FP8 | 100 | 100_ConvTranspose3d_Clamp_Min_Divide | "import torch\nimport torch.nn as nn\n\nclass Model(nn.Module):\n \"\"\"\n A model that perfor(...TRUNCATED) | 9 | model_final | 1 | 0.95 | [{"elapsed_seconds":1061.1975018978,"evaluation":{"compiled":true,"completed_at":null,"correctness":(...TRUNCATED) |
End of preview. Expand in Data Studio
GLM-5.2 KernelGym Rollouts
This dataset contains 3,200 feedback-driven GPU-kernel optimization trajectories
generated by zai-org/GLM-5.2-FP8: 100 validation tasks, two backends (inline
CUDA and Triton), and 16 rollouts per task.
Each trajectory retains the prompt/feedback message history, model responses and reasoning, extracted kernel code, KernelGym compilation and correctness results, profiling metadata, token usage, and stopping decision. Every published record ended with the model-controlled final decision and passed correctness, custom kernel execution, and nonnegative runtime validation.
Collection configuration
- Source tasks:
hkust-nlp/drkernel-validation-data, validation split - KernelGym: https://github.com/hkust-nlp/KernelGYM at
3a84417f8c0efaadb215ef638b37d12e71ed20f3 - Model:
zai-org/GLM-5.2-FP8 - Sampling: temperature 1.0, top-p 0.95, up to 16,384 generated tokens
- Context length: 262,144 tokens
- Maximum feedback turns: 128
Quality summary
| Backend | Records | Tasks | Mean turns | Median speedup | Faster than baseline | Unique final code |
|---|---|---|---|---|---|---|
| CUDA | 1,600 | 100 | 1.889 | 1.072x | 75.4% | 100.0% |
| Triton | 1,600 | 100 | 1.976 | 1.427x | 88.1% | 100.0% |
Speedups below 1.0 are retained because the dataset records valid optimization trajectories, not only improvements.
Schema
The compressed JSONL shards preserve the original nested rollout records. Core
fields include dataset_index, problem_id, problem_name, reference_code,
backend, rollout, messages, turns, stop_reason, and final_response.
Each turn contains the model response and reasoning, extracted kernel code,
evaluation results, usage, and elapsed time.
Provenance and licensing
The records include source task text and reference implementations from
hkust-nlp/drkernel-validation-data. Review that dataset, KernelGym, and
GLM-5.2 terms before redistribution or commercial use.
- Downloads last month
- 65