Datasets:
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -66,11 +66,11 @@ dataset_info:
|
|
| 66 |
|
| 67 |
# Terminator-Qwen3-8B Training Data
|
| 68 |
|
| 69 |
-
[Project page](https://terminator-llm.github.io/) | [Paper](https://arxiv.org/abs/2603.12529) | [Code](https://github.com/acnagle/terminator
|
| 70 |
|
| 71 |
The tokenized, labeled data used to train [Terminator-Qwen3-8B](https://huggingface.co/acnagle/Terminator-Qwen3-8B). [Terminator](https://terminator-llm.github.io/) is a small network that predicts when a reasoning model has reached its final answer so the remaining reasoning can be skipped.
|
| 72 |
|
| 73 |
-
Each row is one chain of thought (CoT) from Qwen3-8B on a math, coding or science problem, with per-token labels marking where the CoT first reaches its final answer. This dataset holds training data only. To evaluate on the test benchmarks, generate their CoTs with the [code](https://github.com/acnagle/terminator
|
| 74 |
|
| 75 |
## Contents
|
| 76 |
|
|
@@ -95,7 +95,7 @@ tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")
|
|
| 95 |
text = tokenizer.decode(row["input_ids"]) # the prompt and the CoT
|
| 96 |
```
|
| 97 |
|
| 98 |
-
To train with the [code](https://github.com/acnagle/terminator
|
| 99 |
|
| 100 |
```bash
|
| 101 |
hf download acnagle/Terminator-Qwen3-8B-Training-Data train.jsonl val.jsonl \
|
|
@@ -123,7 +123,7 @@ Each problem has 3 CoTs, sampled from Qwen3-8B with temperature 0.6, top-p 0.95,
|
|
| 123 |
|
| 124 |
`prepare_training_data.py` tokenized those CoTs and built the labels with its default settings, holding out 50 problems per source for validation using seed 1337. The rows are grouped by source, so shuffle them before training.
|
| 125 |
|
| 126 |
-
Rebuilding this data from scratch requires generating CoTs and rerunning the pipeline in the [code](https://github.com/acnagle/terminator
|
| 127 |
|
| 128 |
## Sources and licenses
|
| 129 |
|
|
|
|
| 66 |
|
| 67 |
# Terminator-Qwen3-8B Training Data
|
| 68 |
|
| 69 |
+
[Project page](https://terminator-llm.github.io/) | [Paper](https://arxiv.org/abs/2603.12529) | [Code](https://github.com/acnagle/terminator) | [Trained weights](https://huggingface.co/acnagle/Terminator-Qwen3-8B)
|
| 70 |
|
| 71 |
The tokenized, labeled data used to train [Terminator-Qwen3-8B](https://huggingface.co/acnagle/Terminator-Qwen3-8B). [Terminator](https://terminator-llm.github.io/) is a small network that predicts when a reasoning model has reached its final answer so the remaining reasoning can be skipped.
|
| 72 |
|
| 73 |
+
Each row is one chain of thought (CoT) from Qwen3-8B on a math, coding or science problem, with per-token labels marking where the CoT first reaches its final answer. This dataset holds training data only. To evaluate on the test benchmarks, generate their CoTs with the [code](https://github.com/acnagle/terminator).
|
| 74 |
|
| 75 |
## Contents
|
| 76 |
|
|
|
|
| 95 |
text = tokenizer.decode(row["input_ids"]) # the prompt and the CoT
|
| 96 |
```
|
| 97 |
|
| 98 |
+
To train with the [code](https://github.com/acnagle/terminator), download the files to the path its scripts expect, then run the training script from the repo root:
|
| 99 |
|
| 100 |
```bash
|
| 101 |
hf download acnagle/Terminator-Qwen3-8B-Training-Data train.jsonl val.jsonl \
|
|
|
|
| 123 |
|
| 124 |
`prepare_training_data.py` tokenized those CoTs and built the labels with its default settings, holding out 50 problems per source for validation using seed 1337. The rows are grouped by source, so shuffle them before training.
|
| 125 |
|
| 126 |
+
Rebuilding this data from scratch requires generating CoTs and rerunning the pipeline in the [code](https://github.com/acnagle/terminator).
|
| 127 |
|
| 128 |
## Sources and licenses
|
| 129 |
|