CoolFace
Agents
Live
Leaderboard
Models
Community
Search
Create
Alerts
Menu
Model
public
andrew-healey
/
context-compression
source
Hugging Face
updated 2y ago
View on Hugging Face
0
likes
Like
Save
Clone
overview
files
community
commits
settings
root
dir
12_head_baseline_lr_10e-4_head_dim_22_fixed/
dir
12_head_baseline_lr_11e-4/
dir
12_head_baseline_lr_12e-4_head_dim_22_fixed/
dir
12_head_baseline_lr_14e-4_head_dim_22_fixed/
dir
12_head_baseline_lr_16e-4/
dir
12_head_baseline_lr_16e-4_head_dim_22/
dir
12_head_baseline_lr_16e-4_head_dim_22_fixed/
dir
12_head_baseline_lr_20e-4_head_dim_22/
dir
12_head_baseline_lr_20e-4_head_dim_22_fixed/
dir
12_head_baseline_lr_25e-4_head_dim_22_fixed/
dir
12_head_baseline_lr_30e-4_head_dim_22/
dir
12_head_baseline_lr_30e-4_head_dim_22_fixed/
dir
12_head_baseline_lr_35e-4_head_dim_22/
dir
12_head_baseline_lr_40e-4_head_dim_22/
dir
12_head_baseline_lr_45e-4_head_dim_22/
dir
12_head_baseline_lr_50e-4_head_dim_22/
dir
12_head_baseline_lr_6e-4/
dir
12_head_baseline_lr_8e-4_head_dim_22_fixed/
dir
12_head_one_mask_per_head_4_latent_masks/
dir
12_mini_head_one_mask_per_head_1_latent_mask/
dir
12_mini_head_one_mask_per_head_1_latent_mask_halved_lr/
dir
12_mini_head_one_mask_per_head_2_latent_masks/
dir
12_mini_head_two_masks_4_heads/
dir
4x_smaller_bs_half_lr_half_seq_len/
dir
allowing_more_selection_patterns/
dir
att_conv_playground/
dir
attention_kindselective_n_heads2_seed1338/
dir
attention_kindselective_n_heads2_seed1339/
dir
attention_kindselective_n_heads2_seed1340/
dir
attention_kindselective_n_heads2_seed1341/
dir
attention_kindselective_n_heads4_seed1338/
dir
attention_kindselective_n_heads4_seed1339/
dir
attention_kindselective_n_heads4_seed1340/
dir
attention_kindselective_n_heads4_seed1341/
dir
attention_kindselective_n_heads4_seed1342/
dir
attention_kindselective_n_heads4_seed1343/
dir
attention_kindselective_n_heads4_seed1344/
dir
attention_kindselective_n_heads4_seed1345/
dir
attention_kindselective_n_heads8_seed1338/
dir
attention_kindselective_n_heads8_seed1339/
dir
attention_kindselective_n_heads8_seed1340/
dir
attention_kindselective_n_heads8_seed1341/
dir
attention_kindself_n_heads2_seed1338/
dir
attention_kindself_n_heads2_seed1339/
dir
attention_kindself_n_heads2_seed1340/
dir
attention_kindself_n_heads2_seed1341/
dir
attention_kindself_n_heads4_seed1338/
dir
attention_kindself_n_heads4_seed1339/
dir
attention_kindself_n_heads4_seed1340/
dir
attention_kindself_n_heads4_seed1341/
dir
baseline/
dir
baseline_lr_30e-4_seed_1339/
dir
dense_attention_repro_old_codepath/
dir
fix_128_heads_divergence/
dir
half_total_bs_sqrt_lr/
dir
latent_masks_degenerate/
dir
latent_masks_learnable/
dir
logs/
dir
lr0.75e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective/
dir
lr1.5e-3_total_batch_size20480_seq_len128/
dir
lr1.5e-3_total_batch_size20480_seq_len512/
dir
lr1.5e-3_total_batch_size40960_seq_len128/
dir
lr1.5e-3_total_batch_size40960_seq_len512/
dir
lr1.5e-3_total_batch_size5120_seq_len128/
dir
lr1.5e-3_total_batch_size5120_seq_len512/
dir
lr1.5e-4_total_batch_size10240_n_heads2_seed1338/
dir
lr1.5e-4_total_batch_size10240_n_heads2_seed1339/
dir
lr1.5e-4_total_batch_size10240_n_heads2_seed1340/
dir
lr1.5e-4_total_batch_size10240_n_heads4_seed1338/
dir
lr1.5e-4_total_batch_size10240_n_heads4_seed1340/
dir
lr1.5e-4_total_batch_size5120_n_heads2_seed1338/
dir
lr1.5e-4_total_batch_size5120_n_heads2_seed1339/
dir
lr1.5e-4_total_batch_size5120_n_heads2_seed1340/
dir
lr1.5e-4_total_batch_size5120_n_heads4_seed1338/
dir
lr1.5e-4_total_batch_size5120_n_heads4_seed1339/
dir
lr1.5e-4_total_batch_size5120_n_heads4_seed1340/
dir
lr1.5e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective/
dir
lr1.5e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps1/
dir
lr1.75e-4_total_batch_size10240_n_heads2_seed1339/
dir
lr1.75e-4_total_batch_size10240_n_heads2_seed1340/
dir
lr1.75e-4_total_batch_size10240_n_heads4_seed1338/
dir
lr1.75e-4_total_batch_size10240_n_heads4_seed1339/
dir
lr1.75e-4_total_batch_size10240_n_heads4_seed1340/
dir
lr1.75e-4_total_batch_size5120_n_heads2_seed1338/
dir
lr1.75e-4_total_batch_size5120_n_heads2_seed1339/
dir
lr1.75e-4_total_batch_size5120_n_heads2_seed1340/
dir
lr1.75e-4_total_batch_size5120_n_heads4_seed1338/
dir
lr1.75e-4_total_batch_size5120_n_heads4_seed1339/
dir
lr1.75e-4_total_batch_size5120_n_heads4_seed1340/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps1_seed1338/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps1_seed1339/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps200_seed1338/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps200_seed1339/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps1/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps1_seed1338/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps1_seed1339/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps200_seed1338/
dir
lr1.75e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps200_seed1339/
dir
lr10e-4_total_batch_size61440_baseline_seed1338/
dir
lr10e-4_total_batch_size61440_baseline_seed1339/
dir
lr11e-4_total_batch_size122880_baseline_seed1339/
dir
lr11e-4_total_batch_size122880_baseline_seed1340/
dir
lr12e-4_total_batch_size61440_baseline_seed1338/
dir
lr12e-4_total_batch_size61440_baseline_seed1339/
dir
lr13e-4_total_batch_size122880_baseline_seed1339/
dir
lr16e-4_total_batch_size61440_baseline_seed1338/
dir
lr16e-4_total_batch_size61440_baseline_seed1339/
dir
lr1e-4_total_batch_size10240_seq_len1024/
dir
lr1e-4_total_batch_size10240_seq_len128/
dir
lr1e-4_total_batch_size10240_seq_len256/
dir
lr1e-4_total_batch_size10240_seq_len512/
dir
lr1e-4_total_batch_size10240_seq_len64/
dir
lr1e-4_total_batch_size20480_seq_len1024/
dir
lr1e-4_total_batch_size20480_seq_len128/
dir
lr1e-4_total_batch_size20480_seq_len256/
dir
lr1e-4_total_batch_size20480_seq_len512/
dir
lr1e-4_total_batch_size40960_seq_len1024/
dir
lr1e-4_total_batch_size40960_seq_len128/
dir
lr1e-4_total_batch_size40960_seq_len256/
dir
lr1e-4_total_batch_size40960_seq_len512/
dir
lr1e-4_total_batch_size5120_seq_len128/
dir
lr1e-4_total_batch_size5120_seq_len256/
dir
lr1e-4_total_batch_size5120_seq_len256_decay_lrfalse/
dir
lr1e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective/
dir
lr1e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps200/
dir
lr1e-4_total_batch_size5120_seq_len512/
dir
lr1e-4_total_batch_size5120_seq_len64/
dir
lr2e-4_total_batch_size10240_n_heads2_seed1338/
dir
lr2e-4_total_batch_size10240_n_heads2_seed1339/
dir
lr2e-4_total_batch_size10240_n_heads2_seed1340/
dir
lr2e-4_total_batch_size10240_n_heads4_seed1338/
dir
lr2e-4_total_batch_size10240_n_heads4_seed1339/
dir
lr2e-4_total_batch_size10240_n_heads4_seed1340/
dir
lr2e-4_total_batch_size10240_seq_len1024/
dir
lr2e-4_total_batch_size10240_seq_len128/
dir
lr2e-4_total_batch_size10240_seq_len256/
dir
lr2e-4_total_batch_size10240_seq_len512/
dir
lr2e-4_total_batch_size5120_n_heads2_seed1338/
dir
lr2e-4_total_batch_size5120_n_heads2_seed1339/
dir
lr2e-4_total_batch_size5120_n_heads2_seed1340/
dir
lr2e-4_total_batch_size5120_n_heads4_seed1338/
dir
lr2e-4_total_batch_size5120_n_heads4_seed1339/
dir
lr2e-4_total_batch_size5120_n_heads4_seed1340/
dir
lr2e-4_total_batch_size5120_seq_len128/
dir
lr2e-4_total_batch_size5120_seq_len256/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps1_seed1338/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps1_seed1339/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps200_seed1338/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrfalse_attention_kindselective_warmup_steps200_seed1339/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps1_seed1338/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps1_seed1339/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps200_seed1338/
dir
lr2e-4_total_batch_size5120_seq_len256_decay_lrtrue_attention_kindselective_warmup_steps200_seed1339/
dir
lr2e-4_total_batch_size5120_seq_len512/
dir
lr2e-4_total_batch_size61440_baseline_seed1338/
dir
lr2e-4_total_batch_size61440_baseline_seed1339/
dir
lr2e-4_total_batch_size61440_baseline_seed1340/
dir
lr2e-4_total_batch_size61440_baseline_seed1341/
dir
lr3e-3_total_batch_size20480_seq_len128/
dir
lr3e-3_total_batch_size20480_seq_len512/
dir
lr3e-3_total_batch_size40960_seq_len128/
dir
lr3e-3_total_batch_size40960_seq_len512/
dir
lr3e-3_total_batch_size5120_seq_len128/
dir
lr3e-3_total_batch_size5120_seq_len512/
dir
lr3e-4_total_batch_size20480_seq_len128/
dir
lr4e-4_total_batch_size10240_seq_len128/
dir
lr4e-4_total_batch_size20480_seq_len128/
dir
lr4e-4_total_batch_size20480_seq_len512/
dir
lr4e-4_total_batch_size40960_seq_len128/
dir
lr4e-4_total_batch_size40960_seq_len512/
dir
lr4e-4_total_batch_size5120_seq_len128/
dir
lr4e-4_total_batch_size5120_seq_len512/
dir
lr5.5e-5_total_batch_size61440_baseline_seed1338/
dir
lr5.5e-5_total_batch_size61440_baseline_seed1339/
dir
lr5.5e-5_total_batch_size61440_baseline_seed1340/
dir
lr5.5e-5_total_batch_size61440_baseline_seed1341/
dir
lr5.5e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1339/
dir
lr5.5e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1340/
dir
lr5.5e-5_total_batch_size61440_two_masks_4_heads_seed1338/
dir
lr5.5e-5_total_batch_size61440_two_masks_4_heads_seed1339/
dir
lr5.5e-5_total_batch_size61440_two_masks_4_heads_seed1340/
dir
lr5.5e-5_total_batch_size61440_two_masks_4_heads_seed1341/
dir
lr5e-4_total_batch_size20480_seq_len128/
dir
lr6.5e-5_total_batch_size61440_baseline_seed1338/
dir
lr6.5e-5_total_batch_size61440_baseline_seed1339/
dir
lr6.5e-5_total_batch_size61440_baseline_seed1341/
dir
lr6.5e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1338/
dir
lr6.5e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1340/
dir
lr6.5e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1341/
dir
lr6.5e-5_total_batch_size61440_two_masks_4_heads_seed1338/
dir
lr6.5e-5_total_batch_size61440_two_masks_4_heads_seed1339/
dir
lr6.5e-5_total_batch_size61440_two_masks_4_heads_seed1340/
dir
lr6.5e-5_total_batch_size61440_two_masks_4_heads_seed1341/
dir
lr6e-4_total_batch_size10240_seq_len128/
dir
lr6e-4_total_batch_size20480_seq_len128/
dir
lr6e-4_total_batch_size5120_seq_len128/
dir
lr6e-4_total_batch_size61440_baseline_seed1338/
dir
lr6e-4_total_batch_size61440_baseline_seed1339/
dir
lr6e-4_total_batch_size61440_baseline_seed1340/
dir
lr6e-4_total_batch_size61440_baseline_seed1341/
dir
lr6e-5_total_batch_size61440_baseline_seed1338/
dir
lr6e-5_total_batch_size61440_baseline_seed1339/
dir
lr6e-5_total_batch_size61440_baseline_seed1340/
dir
lr6e-5_total_batch_size61440_baseline_seed1341/
dir
lr6e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1338/
dir
lr6e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1339/
dir
lr6e-5_total_batch_size61440_one_mask_per_head_2_latent_vectors_seed1341/
dir
lr6e-5_total_batch_size61440_two_masks_4_heads_seed1338/
dir
lr6e-5_total_batch_size61440_two_masks_4_heads_seed1339/
dir
lr6e-5_total_batch_size61440_two_masks_4_heads_seed1340/
dir
lr6e-5_total_batch_size61440_two_masks_4_heads_seed1341/
dir
lr8e-4_total_batch_size20480_seq_len128/
dir
lr8e-4_total_batch_size20480_seq_len512/
dir
lr8e-4_total_batch_size40960_seq_len128/
dir
lr8e-4_total_batch_size40960_seq_len512/
dir
lr8e-4_total_batch_size5120_seq_len128/
dir
lr8e-4_total_batch_size5120_seq_len512/
dir
lr8e-4_total_batch_size61440_baseline_seed1338/
dir
lr8e-4_total_batch_size61440_baseline_seed1339/
dir
lr8e-4_total_batch_size61440_baseline_seed1340/
dir
lr8e-4_total_batch_size61440_baseline_seed1341/
dir
lr9e-4_total_batch_size61440_baseline_seed1338/
dir
lr9e-4_total_batch_size61440_baseline_seed1339/
dir
memory_loss_run_0/
dir
mha_baseline_lr_tuning/
dir
mha_const_hd_scale_nh/
dir
mha_const_width_scale_hd/
dir
mha_conv_coord_check_4/
dir
mha_coord_check/
dir
mha_numerics/
dir
n_heads12_lr10e-4_total_batch_size61440_seed1338/
dir
n_heads12_lr10e-4_total_batch_size61440_seed1339/
dir
n_heads12_lr14e-4_total_batch_size61440_seed1338/
dir
n_heads12_lr14e-4_total_batch_size61440_seed1339/
dir
new_mha_const_hd_scale_nh/
dir
one_mask_per_head/
dir
one_mask_per_head_1_latent_mask/
dir
one_mask_per_head_1_latent_vector/
dir
one_mask_per_head_2_latent_masks/
dir
one_mask_per_head_2_latent_vectors/
dir
one_mask_per_head_4_latent_masks/
dir
one_mask_per_head_4_latent_vectors/
dir
protection_none_torch_compile/
dir
same_total_bs_half_seq_len/
dir
scaling_mha_conv_granularity_2/
dir
scaling_mha_conv_granularity_3/
dir
scaling_mha_granularity_2/
dir
scaling_mha_nh/
dir
scaling_mha_nh_2/
dir
scaling_mha_nh_3/
dir
scaling_sdpa_granularity_2/
dir
scaling_sdpa_nh/
dir
scaling_sdpa_nh_2/
dir
sdpa_16_maybe_fixed/
dir
sdpa_16_spooky/
dir
selective_run_0/
dir
selective_run_0_continued/
dir
self_to_selective_run_0_restarted/
dir
self_to_selective_run_1_restarted_with_ko_zero/
dir
self_to_selective_run_1_restarted_with_memory_penalty/
dir
self_to_selective_run_1_restarted_with_o_zero/
dir
test_logs/
dir
trying_new_latent_masks/
dir
two_latent_masks/
dir
two_latent_masks_compile_comparison/
dir
two_masks_12_heads/
dir
two_masks_4_heads/
dir
two_masks_shared_1_head/
dir
unselective_run_0/
dir
unselective_run_0_continued/
dir
unselective_run_0_restarted/
dir
unselective_run_0_restarted_with_head/
dir
unselective_run_1_restarted/
dir
unselective_run_1_restarted_with_ko_zero/
dir
unselective_run_1_restarted_with_o_rescaled/
dir
unselective_run_1_restarted_with_o_zero/
file
.gitattributes
2 KB
download
andrew-healey/context-compression · main · files are served by the source, never re-hosted here