Datasets:
File size: 2,123 Bytes
5bb3330 09bc027 5bb3330 5903856 5bb3330 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 | ---
license: other
task_categories:
- text-generation
language:
- en
tags:
- code
- code-optimization
- cpp
- sft
- grpo
configs:
- config_name: sft
data_files:
- split: train
path: sft/train.jsonl
- config_name: grpo
data_files:
- split: train
path: grpo/train.jsonl
- config_name: evaluation
data_files:
- split: validation
path: eval/validation.jsonl
---
# GLM-4.7-Flash PIE C++ Post-Training Data
The exact prepared dataset used for the GLM-4.7-Flash C++ performance
post-training runs.
## Splits
| File | Rows | Purpose |
| --- | ---: | --- |
| `sft/train.jsonl` | 7,864 | Supervised fine-tuning |
| `grpo/train.jsonl` | 7,887 | GRPO prompt and reward evaluation |
| `eval/validation.jsonl` | 1,259 | Full held-out evaluation |
| `eval/validation_mini126.jsonl` | 126 | Fast evaluation |
| `eval/validation_mini4.jsonl` | 4 | Smoke evaluation |
| `tasks.tar.gz` | 9,146 task JSONs | Reward scoring and evaluation harness inputs |
Each row carries a stable task ID, problem ID, split, prompt, label, and task
metadata. SFT rows additionally contain the user and assistant messages used by
the trainer. The task archive contains every `tasks/...` path referenced by the
training and evaluation rows. The canonical repository downloader verifies and
extracts it into `data/tasks`.
## Preparation
The source tasks were filtered by compiling and executing the oracle answer in
the C++ sandbox. Training rows were retained when the response format was valid,
all tests passed, and the reward result was `correct`.
The source preparation run scored 8,350 training tasks, retained 7,887, and
rejected 463. The complete preparation receipt is in `manifest.json`.
## Provenance
This is a prepared derivative of the
[PIE C++ performance dataset](https://github.com/madaan/pie-perf), which is
based on IBM Project CodeNet. Use and redistribution are subject to the
applicable upstream dataset terms.
Training code:
[TokenBender/browser-is-all-you-need](https://github.com/tokenbender/browser-is-all-you-need/tree/client/glm47-h100-posttraining)
|