datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
sae-activations-llama-3.1-8b-layer19-lmsys-chat-1m
SAE Feature Activations — Llama 3.1 8B Instruct, Layer 19 (LMSYS-Chat-1M)
This dataset contains Sparse Autoencoder (SAE) feature activations extracted from layer 19 of Meta's Llama 3.1 8B Instruct on conversations from LMSYS-Chat-1M.
It also has natural language explainations of features generated by GPT OSS 120B. See subset 4 for details.
The SAE used is Goodfire/Llama-3.1-8B-Instruct-SAE-l19, which decomposes layer-19 residual stream activations into interpretable sparse features.… See the full description on the dataset page: https://huggingface.co/datasets/scaleinvariant/sae-activations-llama-3.1-8b-layer19-lmsys-chat-1m.sovits4.0-768vec-layer12sovits4.0-768vec-layer12底模，
新增large底模，由m4singer+vctk数据集训练，294k为loss14.75的，320k为最终训练步数。
91.2k的d和g模型为loss 16.04的，
100k的d和g模型为最终训练步数，
需要改名为D_0.pth和G_0.pth使用。
新增两组d&g底模
144k是在a10上训练，loss低至14.1的，
216k是在a10上训练的最终训练步数。
classify_layer18_split2classify_layer17_split2llama8b-layer15-sae-probes
Llama8B Sparse Probing Activations
This repository contains activation data accompanying the paper Learning a Generative Meta-Model of LLM Activations.
Project page: https://generative-latent-prior.github.io
Code: https://github.com/g-luo/generative_latent_prior
Quick Start
With this data, you can evaluate GLPs via sparse probing.
The activations are derived from the binary classification datasets from Kantamneni et. al., 2025.
The activations are taken only from… See the full description on the dataset page: https://huggingface.co/datasets/generative-latent-prior/llama8b-layer15-sae-probes.classify_layer19_split1classify_layer19_split3intervener_layer16_1k_dpoclassify_layer17_split1intervener_layer16_2k_dpoAIDA_C2S_cached_activations_layer12llama8b-layer15-llamascope-500
Llama8B LlamaScope SAE Steering
This repository contains SAE steering data accompanying the paper Learning a Generative Meta-Model of LLM Activations.
Project page: https://generative-latent-prior.github.io
Code: https://github.com/g-luo/generative_latent_prior
Quick Start
With this data, you can reproduce the SAE steering experiment (Section 4.1, Figure 5) on Llama-3.1-8B.
The 500 concepts are random Layer 15 LlamaScope L15R-32x features with Neuronpedia… See the full description on the dataset page: https://huggingface.co/datasets/generative-latent-prior/llama8b-layer15-llamascope-500.intervener_layer16_3k_dpollama8b-layer15-meta-neurons
Llama8B Meta-Neurons
This repository contains meta-neuron data accompanying the paper Learning a Generative Meta-Model of LLM Activations.
Project page: https://generative-latent-prior.github.io
Code: https://github.com/g-luo/generative_latent_prior
Quick Start
With this data, you can browse the 98304 meta-neurons of the Llama-3.1-8B GLP (glp-llama8b-d6, Layer 15).
Meta-neurons are the post-SwiGLU activations of the GLP's MLP blocks. For each meta-neuron… See the full description on the dataset page: https://huggingface.co/datasets/generative-latent-prior/llama8b-layer15-meta-neurons.classify_layer18_split1iclr_kernel_steering_self_awareness_general_llama3.2-1B-it_layer10_activationsllama8b-layer15-fineweb-1M
Llama8B 1M Training Activations
This repository contains activation data accompanying the paper Learning a Generative Meta-Model of LLM Activations.
Project page: https://generative-latent-prior.github.io
Code: https://github.com/g-luo/generative_latent_prior
Quick Start
With this data, you can train a GLP on Llama-3.1-8B activations from Layer 15.
The activations are derived from FineWeb.
GLPs are activation diffusion models useful for applications like on-manifold… See the full description on the dataset page: https://huggingface.co/datasets/generative-latent-prior/llama8b-layer15-fineweb-1M.rl-checkpoint-analysis-layer18layer1-resultslayer1-outputllama-3-tqa-seed42-greedy-probe-layer13llama3.2-1B-it_power_seeking_layer10mistral-v0.3-tqa-seed42-greedy-probe-layer13layer1-200-gtwin_lose_pairs_tqa_layer13pythia410m_deduped_sae_hook_resid_pre_layer1Layer1-Lawpersuasion_pairs_new_samples_steered_input_output_positive_layer17persuasion_pairs_new_samples_steered_input_output_positive_layer17_scale10engram-trace-layer1
JackyZZZ111/engram-trace-layer1
This dataset repository contains Engram trace exports.
Files
metadata.json
summary.json
layer_1.parquet
Notes
The Parquet trace files may already use internal ZSTD compression.
That compression stays inside the .parquet file, so there is usually no separate .zst file to upload.