adapter
Datasets
All datasets matching “adapter”VL-Adapter-datasets
VL-Adapter Datasets
Processed CLIP-ResNet101 grid features and annotations for
VL-Adapter: Parameter-Efficient Transfer Learning for Vision-and-Language Tasks
(Sung, Cho, Bansal — CVPR 2022), code at ylsung/VL_adapter.
This repository replaces the original Google Drive download, which is no longer
available. It holds the same data in a Hub-native layout, plus a script that
rebuilds the exact datasets/ directory tree the training code expects.
Quick start — rebuild… See the full description on the dataset page: https://huggingface.co/datasets/ylsung/VL-Adapter-datasets.vision-adapter-embeddings
Vision Adapter MoonViT Embeddings
Precomputed visual embeddings used to train lightweight vision→LLM projectors
without re-running a vision tower: each row is the frozen MoonViT-V2 output
for one training image, stored as raw bfloat16 bytes.
Shards: 103 Parquet files (data/emb_0000.parquet … data/emb_0102.parquet),
1360 rows each, ~139k rows total, ~0.9 TB.
Schema per row:
column
type
meaning
key
string
embedding id, embeddings/<sha1[:20]>.pt; matches emb in… See the full description on the dataset page: https://huggingface.co/datasets/keypa/vision-adapter-embeddings.adaptercast-lora-trajectories
AdapterCast LoRA Trajectory Corpus
Dense LoRA fine-tuning trajectories (adapter weight snapshots + gauge-invariant
spectral features) for research on fine-tuning dynamics and trajectory forecasting.
Companion dataset to the AdapterCast paper (theory: GL(r) gauge structure of LoRA;
measurements: balancedness decay law, Adam pump, regime map; machine: the
NeuralGraphLoRA forecaster).
Layout (per run)
trajectory.dat.zst — zstd fp16 memmap, shape (n_states, n_params)… See the full description on the dataset page: https://huggingface.co/datasets/EntropyCoder/adaptercast-lora-trajectories.Harbor-Adapter
Harbor Adapters — Agent Trajectories
Complete agent execution trajectories from a large-scale agentic evaluation —
every (agent × model × task) rollout with its full step log, tool calls,
model output, and verifier scoring.
Each trajectory is a single (agent, model) run on one task: the agent's steps and
tool calls, the model's outputs, the environment's responses, and the verifier's
reward. Start from the manifest to filter down, then pull only the trajectories
you need.… See the full description on the dataset page: https://huggingface.co/datasets/kendx/Harbor-Adapter.theo_qwen2.5-7b-it_whitebox-adapter-gpu-verification
Status: NOT the paper's results. GPU verification of white-box reads on adapter organisms (2026-08-28). The paper's canonical results are Misalignment-Empirics/theo_impulsive-qwen_2_5-7b-14b-32b-big_eval_results, run runs/impulsive-qwen_2_5-7b-14b-32b-20260925/.
theo_qwen2.5-7b-it_whitebox-adapter-gpu-verification
adapter-based-multimodal-fusion
Falcon-Audio Training Dataset
Training-ready Parquet shards for Falcon-Audio. Rows contain Gemma-tokenized inputs/labels and fp16 Whisper encoder features encoded as raw bytes.