datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
autoinference-agentic-mix-v1
Autoinference Agentic Mix v1
This is a prompt set for the online_agentic serving benchmark. That profile stands
in for long-horizon agent traffic: a large context that grows turn over turn, with
short structured outputs at each step. The usual way to run it uses
generated-shared-prefix, which builds a synthetic shared prefix out of random tokens.
This dataset uses real agent trajectories instead, so the prefix reuse, the context
growth, and the token mix all match what an agent… See the full description on the dataset page: https://huggingface.co/datasets/modal-labs/autoinference-agentic-mix-v1.autoinference-agentic-mix-v2
Autoinference Agentic Mix v2
300 real SWE-agent trajectories from
TIGER-Lab/SWE-Next-SFT-Trajectories,
expanded into one request per assistant turn. Each row carries the conversation
up to that turn and the model generates the turn. Replaying a trajectory in
turn_index order re-sends a growing prefix, which is how an agent loop
actually hits a prefix cache.
What changed from v1
v1 kept only requests with at least 34k prefix tokens. That cut trajectories
down to… See the full description on the dataset page: https://huggingface.co/datasets/modal-labs/autoinference-agentic-mix-v2.shared-emergence-icl-modalities-128
Shared-emergence ICL replication at T=128
This dataset contains the complete raw result archive for the paper
“Many Next-Token Predictors are In-Context Learners.”
The campaign evaluates a fixed suite of 100 program-synthesis tasks using 128
sampled prompts per task, for every clean and deranged shot cell described by
the paper:
21 run keys;
281 experiment cells;
12,800 predictions per cell;
3,596,800 predictions in total.
The archive expands to a top-level results_128/… See the full description on the dataset page: https://huggingface.co/datasets/N8Programs/shared-emergence-icl-modalities-128.rocov2-modality-splits
ROCOv2 Modality-Specific Dataset Splits
Dataset Description
This dataset contains modality-specific splits of the ROCOv2 radiology dataset, organized and processed for training specialized medical image captioning models.
Dataset Summary
Total Samples: 1,000
Modalities: 5
Splits per Modality: train, validation, test
Random Seed: 42
Processing Date: 2025-08-31 12:52:59.233482
Modality Distribution
Modality
Samples
Percentage
CT
188
18.8%… See the full description on the dataset page: https://huggingface.co/datasets/WafaaFraih/rocov2-modality-splits.clinical-cross-modal-memory-fidelity-v0.1Clinical Cross-Modal Memory Fidelity v0.1
Goal
Test whether prior image evidence is recalled accurately over time
Detect retroactive distortion driven by later narrative
Detect fabrication used to patch memory gaps
What it measures
memory_driftEarlier image facts are altered or inverted
fabricationNew findings are invented at recall
cross_modal_consistencyRecalled description matches original image evidence
How it works
Initial image facts are fixed and explicit
Intervening tasks… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-cross-modal-memory-fidelity-v0.1.modality-state-consistency-v0.1
What this dataset tests
Inputs arrive in many forms.
State must stay coherent.
Why it exists
Models drift when switching modality.
Facts change.
Promises vanish.
This set checks whether state stays consistent.
Data format
Each row contains
modality_context
user_message
modality_pressure
constraints
failure_modes_to_avoid
target_behaviors
gold_checklist
Feed the model
modality_context
user_message
constraints
Score for
cross-modal… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/modality-state-consistency-v0.1.k3-toolcall-turn1
k3-toolcall-turn1
191,072 deduplicated tool-calling contexts, normalized to the OpenAI
messages + tools schema. Each row is a conversation prefix that stops right
before the first assistant turn — a context ready for a model to generate at,
not a finished conversation.
Built to generate on-policy targets for speculative-decoding draft training:
a target model generates at each context, and the draft learns to copy that
model's next tokens. The dataset ships contexts, not labels.… See the full description on the dataset page: https://huggingface.co/datasets/junlin-modal/k3-toolcall-turn1.autoinference-realtime-mix-v1
Autoinference Real-Time Generation Mix v1
This is a prompt set for the real_time_generation serving benchmark. That profile
stands in for medium-context, single-shot interactive traffic: roughly 3000 input
tokens, 100 output tokens, one request at a time with no shared context between
requests. The usual way to run it feeds the server random token IDs of a fixed
length. This dataset keeps the same input and output shape but uses real prompts.
The reason real text matters: random… See the full description on the dataset page: https://huggingface.co/datasets/modal-labs/autoinference-realtime-mix-v1.