Datasets:
prompt_id stringlengths 29 61 | prompt stringlengths 9 37.2k | source_dataset stringclasses 5
values | model_name stringclasses 1
value | generated_text stringlengths 361 262k | prompt_mode stringclasses 1
value | max_new_tokens int64 4.1k 4.1k | temperature float64 0.7 0.7 | top_p float64 0.9 0.9 | generation_seed int64 1.23k 1.23k | prompt_generation_seed int64 1.23k 1.38k | chunk_summary listlengths 256 4.1k | dataset_run stringclasses 5
values | shard stringclasses 112
values | source_file stringlengths 108 118 | degenerating bool 2
classes |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
zwhe99_DeepMath-103K-train-8479 | "Evaluate the integral for $N > 0$:\n$$ \\frac{\\exp\\left(-\\frac N2\\right)}{2\\pi N} \\iint \\exp(...TRUNCATED) | zwhe99/DeepMath-103K | swiss-ai/Apertus-8B-Instruct-2509 | "To evaluate the integral, we can use the following steps:\n\n1. **Change of Variables:**\n Let $x(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,313 | [{"degenerating":false,"repetition":0.3686274509803922,"token_index":0},{"degenerating":false,"repet(...TRUNCATED) | deepmath103k_apertus8b_instruct_MNT4096_sharded | shard_56 | "outputs/generations/instruct/deepmath103k_apertus8b_instruct_MNT4096_sharded/shard_56/data/generati(...TRUNCATED) | true |
AI-MO_NuminaMath-1.5-train-1496 | "Given that $\\sin α$, $\\sin 2α$, $\\sin 4α$ form a geometric sequence, find the value of $\\cos(...TRUNCATED) | AI-MO/NuminaMath-1.5 | swiss-ai/Apertus-8B-Instruct-2509 | "Given that $\\sin α$, $\\sin 2α$, $\\sin 4α$ form a geometric sequence, we need to find the valu(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,380 | [{"degenerating":false,"repetition":0.5058823529411764,"token_index":0},{"degenerating":false,"repet(...TRUNCATED) | numinamath15_apertus8b_instruct_MNT4096_sharded | shard_09 | "outputs/generations/instruct/numinamath15_apertus8b_instruct_MNT4096_sharded/shard_09/data/generati(...TRUNCATED) | false |
zwhe99_DeepMath-103K-train-4002 | "Let $A \\in M_n(\\mathbb{C})$ and $\\lambda$ be an eigenvalue of $A$. Determine whether there exist(...TRUNCATED) | zwhe99/DeepMath-103K | swiss-ai/Apertus-8B-Instruct-2509 | "Yes, such a sequence exists. To see this, note that any matrix $A$ can be diagonalized, meaning tha(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,336 | [{"degenerating":false,"repetition":0.43529411764705883,"token_index":0},{"degenerating":false,"repe(...TRUNCATED) | deepmath103k_apertus8b_instruct_MNT4096_sharded | shard_26 | "outputs/generations/instruct/deepmath103k_apertus8b_instruct_MNT4096_sharded/shard_26/data/generati(...TRUNCATED) | false |
AI-MO_NuminaMath-1.5-train-15526 | "What is the radius of the circle inscribed in triangle $XYZ$ if $XY = 7, XZ = 8, YZ = 9$? Express y(...TRUNCATED) | AI-MO/NuminaMath-1.5 | swiss-ai/Apertus-8B-Instruct-2509 | "The radius of the circle inscribed in triangle $XYZ$ is $\\frac{27}{24}$ or $\\frac{9}{8}$.\n\nExpl(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,310 | [{"degenerating":false,"repetition":0.30980392156862746,"token_index":0},{"degenerating":false,"repe(...TRUNCATED) | numinamath15_apertus8b_instruct_MNT4096_sharded | shard_103 | "outputs/generations/instruct/numinamath15_apertus8b_instruct_MNT4096_sharded/shard_103/data/generat(...TRUNCATED) | true |
AI-MO_NuminaMath-1.5-train-4373 | "If $a$ and $b$ are integers, and the distinct roots of the equation $ax^2+bx-2008=0$ are both prime(...TRUNCATED) | AI-MO/NuminaMath-1.5 | swiss-ai/Apertus-8B-Instruct-2509 | "Let the roots of the equation be $p$ and $q$, both prime numbers. Then, by Vieta's formulas, we hav(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,257 | [{"degenerating":true,"repetition":0.8431372549019608,"token_index":0},{"degenerating":true,"repetit(...TRUNCATED) | numinamath15_apertus8b_instruct_MNT4096_sharded | shard_29 | "outputs/generations/instruct/numinamath15_apertus8b_instruct_MNT4096_sharded/shard_29/data/generati(...TRUNCATED) | true |
nvidia_Llama-Nemotron-Post-Training-Dataset-chat-3663 | "reusable function to show modal. no bug if called multiple times. modal just stack one after the o(...TRUNCATED) | nvidia/Llama-Nemotron-Post-Training-Dataset | swiss-ai/Apertus-8B-Instruct-2509 | "Sure, here's a reusable function that uses Bootstrap 5 to show a modal. This function will stack th(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,297 | [{"degenerating":false,"repetition":0.1098039215686275,"token_index":0},{"degenerating":false,"repet(...TRUNCATED) | llama_nemotron_chat_apertus8b_instruct_MNT4096_sharded | shard_24 | "outputs/generations/instruct/llama_nemotron_chat_apertus8b_instruct_MNT4096_sharded/shard_24/data/g(...TRUNCATED) | false |
zwhe99_DeepMath-103K-train-15103 | Evaluate the integral \( \int_{|z|=2} \frac{|z| e^z}{z^2} \, dz \). | zwhe99/DeepMath-103K | swiss-ai/Apertus-8B-Instruct-2509 | "To evaluate the integral \\( \\int_{|z|=2} \\frac{|z| e^z}{z^2} \\, dz \\), we can use the followin(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,337 | [{"degenerating":false,"repetition":0.43137254901960786,"token_index":0},{"degenerating":false,"repe(...TRUNCATED) | deepmath103k_apertus8b_instruct_MNT4096_sharded | shard_100 | "outputs/generations/instruct/deepmath103k_apertus8b_instruct_MNT4096_sharded/shard_100/data/generat(...TRUNCATED) | true |
zwhe99_DeepMath-103K-train-13692 | "Given the quartic equation \\(x^4 + kx^2 + 90x - 2009 = 0\\) with roots \\(x_1, x_2, x_3, x_4\\), w(...TRUNCATED) | zwhe99/DeepMath-103K | swiss-ai/Apertus-8B-Instruct-2509 | "To find the value of \\(k\\) using Vieta's formulas, we need to consider the relationships between (...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,276 | [{"degenerating":false,"repetition":0.5490196078431373,"token_index":0},{"degenerating":false,"repet(...TRUNCATED) | deepmath103k_apertus8b_instruct_MNT4096_sharded | shard_91 | "outputs/generations/instruct/deepmath103k_apertus8b_instruct_MNT4096_sharded/shard_91/data/generati(...TRUNCATED) | false |
FreedomIntelligence_medical-o1-verifiable-problem-train-10559 | In patients with chronic renal failure, what is the major cause of death? | FreedomIntelligence/medical-o1-verifiable-problem | swiss-ai/Apertus-8B-Instruct-2509 | "In patients with chronic renal failure, the major cause of death is usually related to cardiovascul(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,293 | [{"degenerating":false,"repetition":0.1725490196078432,"token_index":0},{"degenerating":false,"repet(...TRUNCATED) | medical_o1_verifiable_apertus8b_instruct_MNT4096_sharded | shard_70 | "outputs/generations/instruct/medical_o1_verifiable_apertus8b_instruct_MNT4096_sharded/shard_70/data(...TRUNCATED) | false |
nvidia_Llama-Nemotron-Post-Training-Dataset-chat-500 | "I want you to act as a screenwriter and brainstorm some ideas for a plot. Caroline Hertig, a beauti(...TRUNCATED) | nvidia/Llama-Nemotron-Post-Training-Dataset | swiss-ai/Apertus-8B-Instruct-2509 | "1. The hospital is full: The hospital where Caroline was supposed to be taken after her accident ha(...TRUNCATED) | chat | 4,096 | 0.7 | 0.9 | 1,234 | 1,284 | [{"degenerating":false,"repetition":0.1333333333333333,"token_index":0},{"degenerating":false,"repet(...TRUNCATED) | llama_nemotron_chat_apertus8b_instruct_MNT4096_sharded | shard_03 | "outputs/generations/instruct/llama_nemotron_chat_apertus8b_instruct_MNT4096_sharded/shard_03/data/g(...TRUNCATED) | false |
End of preview. Expand in Data Studio
degeneration-probe-instruct-token-level-balanced
Downsampled (1:3 positive:negative) variant of luca-sartori/degeneration-probe-instruct-token-level.
An example is considered positive if it contains at least one token with repetition >= 0.8 in the chunk_summary field. The downsampling keeps all positive examples and adds a random subset of negative examples in a 1:3 ratio.
Examples whose chunk_summary contained no scored tokens (every repetition value null) have been dropped, since they cannot contribute to training or evaluation (every label maps to the ignore sentinel -100).
Splits
| Split | Examples | Positives | Negatives |
|---|---|---|---|
| train | 2,379 | 915 | 1,464 |
| validation | 299 | 113 | 186 |
| test | 295 | 113 | 182 |
The dropped rows were exclusively negatives generated by swiss-ai/Apertus-8B-Instruct-2509, so the 1:3 pos:neg ratio is no longer preserved after the cleanup (now roughly 1:1.6).
Reproducibility
- Source dataset:
luca-sartori/degeneration-probe-instruct-token-level - Positive threshold:
repetition >= 0.8(any token inchunk_summary) - Ratio (before cleanup): 3 negatives per positive, per split
- Negative sampling:
numpy.random.default_rng(seed=42).choice(..., replace=False) - Cleanup: rows where every entry in
chunk_summaryhasrepetition is Noneare dropped - Schema: identical to the source dataset (16 columns, including
chunk_summary)
Intended use
Use this dataset for faster experimentation when training token-level degeneration probes on the original heavily-imbalanced data (≈1:72 at example level, ≈1:9 at token level) is too slow.
Note that validation and test splits are also downsampled, so metrics on this dataset are not directly comparable to metrics on the full source dataset.
- Downloads last month
- 40