Lambent
Datasets
All datasets matching “Lambent”influence-taste-labels
influence-taste-labels: 100k web docs with 17-dimensional gradient-alignment influence labels
100,000 documents from
mlfoundations/dclm-baseline-1.0
(CC-BY-4.0), each labeled with its training influence toward 17
held-out text categories: the cosine between the document's training
gradient and each category's mean held-out gradient, measured at the
end-of-stable-LR checkpoint of a 56M reference model (PleIAs/Monad 8k
tokenizer, 512 hidden × 12 layers, ~0.9B dclm tokens).
These… See the full description on the dataset page: https://huggingface.co/datasets/Lambent/influence-taste-labels.shakespeare_sonnets_diffuseddetails_Lambent__threebird-7Bdisclaimer_behaviorsrp-teacher-synth-wizard-bixtral-sharegptSmall dataset attempting to instruct a model in the usage of system prompts.
Personas were synthesized by Lambent/braidbird-scribe-7B, and the rest was synthesized with WizardLM-2-8x22B.
These are multi-turn, in-character conversations of a specific number of turns.
Text length was not strictly specified beyond setting Wizard's max output length to 1024 tokens.
Having sampled a couple, my estimate is they should mostly fit within 4096 tokens, and certainly within 8192.
qwen3.5-moe-awq-calibration
Qwen3.5 MoE AWQ Calibration Dataset
Calibration dataset for AWQ (Activation-Aware Weight Quantization) of
Qwen/Qwen3.5-35B-A3B and
Qwen/Qwen3.5-35B-A3B-Base.
Designed for MoE expert routing diversity: Qwen3.5-35B-A3B has 256 experts with 8
active per token, so calibration data needs broad domain coverage to exercise as many
routing paths as possible.
Sampling methodology
Source: PleIAs/common_corpus
(open multi-domain corpus with labeled collections)
Filtering:
Token… See the full description on the dataset page: https://huggingface.co/datasets/Lambent/qwen3.5-moe-awq-calibration.