Add 12 model file(s)

t1_pure_3d_s_d5.2_eb512_lr1.5e-3_seed0/checkpoint-2Blb.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:931067f44bbf21a396f033723deab0270419712cff03703b660be89215dc7b60
+size 2102189541

t1_pure_3d_s_d5.2_eb512_lr1.5e-3_seed0/checkpoint-4Blb.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3afeb905565013743e814e8a779cdeb14477652178c51e88b1673c766ad2b70f
+size 2102189541

t1_pure_3d_s_d5.2_eb512_lr1.5e-3_seed0/checkpoint-final.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fbc7d738f197c51fbf2e9d87e4162009f328f9a13ebae8f5eac03c4bd40521d
+size 2102190064

t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0/checkpoint-2Blb.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0d55550475698801a2b0eb03ab8f8bdfcb30596cff1c7011ca793bded1926b2
+size 2102189605

t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0/checkpoint-4Blb.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:965c6886ca572e93e0bf668eebb10f8169bc9e7a502499d379f2ea50a10e24c6
+size 2102189605

t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0/checkpoint-final.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03c8e1b60f0ac7bb83a0328db1687b0355c84d488a34404ff4f7f8b351129261
+size 2102190128

t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0/config.json

@@ -0,0 +1,117 @@
+{
+  "run_name": "t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0",
+  "epoch": 0,
+  "n_params": 179376136,
+  "args": {
+    "run_name": "t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0",
+    "batch_size": 32,
+    "epochs": 1,
+    "total_tokens": -1,
+    "lossbearing_tokens": 5.2,
+    "accum_iter": 16,
+    "effective_batch_size": 512,
+    "lr_batch_scaling": "none",
+    "save_ckpt_freq": 1,
+    "save_ckpt_freq_tokens": 2,
+    "debug_batch_stats": 0,
+    "model": "fm_small_12d_swiglu_nobias",
+    "dtype": "bfloat16",
+    "loss_type": "token",
+    "context_length": 2048,
+    "pos_emb_ablation": "vanilla",
+    "max_blocks": 32,
+    "intra_pos_emb": "default",
+    "video_pos_3d": false,
+    "finetune": "",
+    "opt": "adamw",
+    "opt_eps": 1e-08,
+    "opt_betas": [
+      0.9,
+      0.95
+    ],
+    "compute_grad_norm": true,
+    "clip_grad": 1.0,
+    "skip_grad": null,
+    "momentum": 0.9,
+    "weight_decay": 0.05,
+    "weight_decay_end": 0.05,
+    "blr": 0.001,
+    "min_blr": 5e-05,
+    "scheduler": "wsd",
+    "decay_fraction": 0.1,
+    "warmup_steps": 250,
+    "cooldown_steps": -1,
+    "data_config": "cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml",
+    "data_config_val": "cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml",
+    "cross_eval_val": "",
+    "s3_endpoint": "",
+    "s3_data_endpoint": "",
+    "s3_multipart_chunksize_mb": 512,
+    "s3_multipart_threshold_mb": 512,
+    "s3_max_io_queue": 100,
+    "text_tokenizer_path": "fourm/utils/tokenizer/trained/smollm_tokenizer.json",
+    "eval_freq": 1,
+    "eval_freq_tokens": 1,
+    "dist_eval": true,
+    "eval": false,
+    "val_sampling_mode": "without_replacement",
+    "output_dir": "output/cc/t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0",
+    "device": "cuda",
+    "seed": 0,
+    "resume": "",
+    "auto_resume": true,
+    "start_epoch": 0,
+    "num_workers": 10,
+    "pin_mem": true,
+    "find_unused_params": true,
+    "rlimit": 4096,
+    "print_all": false,
+    "show_user_warnings": false,
+    "s3_save_dir": "",
+    "dist_url": "env://",
+    "log_wandb": true,
+    "wandb_project": "loom",
+    "wandb_entity": null,
+    "wandb_run_name": "t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0",
+    "wandb_group": "cc-solo",
+    "dim": null,
+    "depth": null,
+    "num_heads": null,
+    "mlp_ratio": null,
+    "num_experts": 1,
+    "granularity": 1,
+    "routing": null,
+    "moe_aux_coef": 0.01,
+    "z_loss_coef": null,
+    "moe_compile": false,
+    "log_grad_norms_every": 0,
+    "save_at_tokens": null,
+    "unigram_entropies": null,
+    "cooldown_branch": false,
+    "cooldown_tokens": 0,
+    "config_path": "cfgs/default/4m/experiments/confound/solo_s/3d/t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0.yaml",
+    "rank": 0,
+    "world_size": 1,
+    "gpu": 0,
+    "distributed": true,
+    "dist_backend": "nccl",
+    "num_tasks": 1,
+    "sampling_mode": "without_replacement",
+    "max_micro_steps": 158692,
+    "lossbearing_target": 5200000000.0,
+    "lossbearing_mode": true,
+    "lr": 0.001,
+    "min_lr": 5e-05,
+    "global_micro_step": 158692,
+    "cum_lossbearing": 5200948224.0,
+    "cum_mod_tokens": {
+      "text": 5538816,
+      "tok_3d": 5195409408
+    }
+  },
+  "source_configs": {
+    "cfgs/default/4m/experiments/confound/solo_s/3d/t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0.yaml": "# S solo optimiser control t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0 (confound program, stage 1 \u2014 D6-2 \u00a74 finding 5).\n# Clone of t1/unimodal/3d/run_pure.yaml: D 5.2B, effective batch 512, absolute LR 0.001, seed 0, weight decay 0.05 (base 0.1).\n# Compared with the same batch at the base optimiser (output/hb dose-response / half-batch arms).\n_base_: ../../../../models/dec/4m-s-dec.yaml\nrun_name: t1_pure_3d_s_d5.2_eb512_lr1e-3_wd0.05_seed0\ntotal_tokens: -1\nlossbearing_tokens: 5.2\ndata_config: cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml\ndata_config_val: cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml\nwarmup_steps: 250\neval_freq_tokens: 1\nsave_ckpt_freq_tokens: 2\nwandb_group: cc-solo\neffective_batch_size: 512\nlr_batch_scaling: none\nblr: 0.001\nseed: 0\nweight_decay: 0.05\naccum_iter: 16\nmin_blr: 5.0e-05\n",
+    "cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml": "# T1 pure unimodal training data \u2014 3d\n# Text stripped from all sources (pure modality-only condition)\n\nmodalities:\n- tok_3d\nepoch_size: 50000000\nsampler_type: temperature\ntemperature_tau: 1.0\nsources:\n- name: 3d\n  shards: data/shards/3d_shuffled/train/shard_{000000..000685}.tar\n  weight: 685736\n  modality_tokens:\n    tok_3d: 1875\n  format: npz\nsampling_mode: without_replacement\n",
+    "cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml": "# T1 pure unimodal validation data \u2014 3d\n# Text stripped from all sources (pure modality-only condition)\n\nmodalities:\n- tok_3d\nval_steps: 200\nsampler_type: uniform\nsources:\n- name: 3d_val\n  shards: data/shards/3d_shuffled/val/shard_{000000..000013}.tar\n  weight: 14000\n  modality_tokens:\n    tok_3d: 1875\n  format: npz\n"
+  }
+}

t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0/checkpoint-2Blb.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a2fc8dac8454229f131a51e58b7ed6dd430c9fad73a93bd3b4421fef3305554
+size 2102189605

t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0/checkpoint-4Blb.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88d41646826ccb5cb7f0a5d1b5cf8475cc5f8a25988ca7d9a167354888784093
+size 2102189605

t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0/checkpoint-final.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30f48f37bc1aed80af3389723e9b725926a1a5a39422bcb5016258db3c6ce45f
+size 2102190128

t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0/config.json

@@ -0,0 +1,117 @@
+{
+  "run_name": "t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0",
+  "epoch": 0,
+  "n_params": 179376136,
+  "args": {
+    "run_name": "t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0",
+    "batch_size": 32,
+    "epochs": 1,
+    "total_tokens": -1,
+    "lossbearing_tokens": 5.2,
+    "accum_iter": 16,
+    "effective_batch_size": 512,
+    "lr_batch_scaling": "none",
+    "save_ckpt_freq": 1,
+    "save_ckpt_freq_tokens": 2,
+    "debug_batch_stats": 0,
+    "model": "fm_small_12d_swiglu_nobias",
+    "dtype": "bfloat16",
+    "loss_type": "token",
+    "context_length": 2048,
+    "pos_emb_ablation": "vanilla",
+    "max_blocks": 32,
+    "intra_pos_emb": "default",
+    "video_pos_3d": false,
+    "finetune": "",
+    "opt": "adamw",
+    "opt_eps": 1e-08,
+    "opt_betas": [
+      0.9,
+      0.95
+    ],
+    "compute_grad_norm": true,
+    "clip_grad": 1.0,
+    "skip_grad": null,
+    "momentum": 0.9,
+    "weight_decay": 0.1,
+    "weight_decay_end": 0.1,
+    "blr": 0.001,
+    "min_blr": 5e-05,
+    "scheduler": "wsd",
+    "decay_fraction": 0.1,
+    "warmup_steps": 500,
+    "cooldown_steps": -1,
+    "data_config": "cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml",
+    "data_config_val": "cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml",
+    "cross_eval_val": "",
+    "s3_endpoint": "",
+    "s3_data_endpoint": "",
+    "s3_multipart_chunksize_mb": 512,
+    "s3_multipart_threshold_mb": 512,
+    "s3_max_io_queue": 100,
+    "text_tokenizer_path": "fourm/utils/tokenizer/trained/smollm_tokenizer.json",
+    "eval_freq": 1,
+    "eval_freq_tokens": 1,
+    "dist_eval": true,
+    "eval": false,
+    "val_sampling_mode": "without_replacement",
+    "output_dir": "output/cc/t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0",
+    "device": "cuda",
+    "seed": 0,
+    "resume": "",
+    "auto_resume": true,
+    "start_epoch": 0,
+    "num_workers": 10,
+    "pin_mem": true,
+    "find_unused_params": true,
+    "rlimit": 4096,
+    "print_all": false,
+    "show_user_warnings": false,
+    "s3_save_dir": "",
+    "dist_url": "env://",
+    "log_wandb": true,
+    "wandb_project": "loom",
+    "wandb_entity": null,
+    "wandb_run_name": "t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0",
+    "wandb_group": "cc-solo",
+    "dim": null,
+    "depth": null,
+    "num_heads": null,
+    "mlp_ratio": null,
+    "num_experts": 1,
+    "granularity": 1,
+    "routing": null,
+    "moe_aux_coef": 0.01,
+    "z_loss_coef": null,
+    "moe_compile": false,
+    "log_grad_norms_every": 0,
+    "save_at_tokens": null,
+    "unigram_entropies": null,
+    "cooldown_branch": false,
+    "cooldown_tokens": 0,
+    "config_path": "cfgs/default/4m/experiments/confound/solo_s/3d/t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0.yaml",
+    "rank": 0,
+    "world_size": 1,
+    "gpu": 0,
+    "distributed": true,
+    "dist_backend": "nccl",
+    "num_tasks": 1,
+    "sampling_mode": "without_replacement",
+    "max_micro_steps": 158692,
+    "lossbearing_target": 5200000000.0,
+    "lossbearing_mode": true,
+    "lr": 0.001,
+    "min_lr": 5e-05,
+    "global_micro_step": 158692,
+    "cum_lossbearing": 5200948224.0,
+    "cum_mod_tokens": {
+      "text": 5538816,
+      "tok_3d": 5195409408
+    }
+  },
+  "source_configs": {
+    "cfgs/default/4m/experiments/confound/solo_s/3d/t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0.yaml": "# S solo variant t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0 (confound program D6-1).\n# Clone of t1/unimodal/3d/run_pure.yaml: D 5.2B, effective batch 512, absolute LR 0.001 (lr_batch_scaling none), seed 0, warmup 500 optimiser steps (= the eb1024 run's warmup in tokens).\n# Self-cooled WSD endpoint; compared with the other D6-1 solo variants of the same modality\n# and D (seed spread, LR x batch grid, warmup / z-loss controls).\n_base_: ../../../../models/dec/4m-s-dec.yaml\nrun_name: t1_pure_3d_s_d5.2_eb512_lr1e-3_wu500_seed0\ntotal_tokens: -1\nlossbearing_tokens: 5.2\ndata_config: cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml\ndata_config_val: cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml\nwarmup_steps: 500\neval_freq_tokens: 1\nsave_ckpt_freq_tokens: 2\nwandb_group: cc-solo\neffective_batch_size: 512\nlr_batch_scaling: none\nblr: 0.001\nseed: 0\naccum_iter: 16\nmin_blr: 5.0e-05\n",
+    "cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml": "# T1 pure unimodal training data \u2014 3d\n# Text stripped from all sources (pure modality-only condition)\n\nmodalities:\n- tok_3d\nepoch_size: 50000000\nsampler_type: temperature\ntemperature_tau: 1.0\nsources:\n- name: 3d\n  shards: data/shards/3d_shuffled/train/shard_{000000..000685}.tar\n  weight: 685736\n  modality_tokens:\n    tok_3d: 1875\n  format: npz\nsampling_mode: without_replacement\n",
+    "cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml": "# T1 pure unimodal validation data \u2014 3d\n# Text stripped from all sources (pure modality-only condition)\n\nmodalities:\n- tok_3d\nval_steps: 200\nsampler_type: uniform\nsources:\n- name: 3d_val\n  shards: data/shards/3d_shuffled/val/shard_{000000..000013}.tar\n  weight: 14000\n  modality_tokens:\n    tok_3d: 1875\n  format: npz\n"
+  }
+}

t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0/config.json

@@ -0,0 +1,117 @@
+{
+  "run_name": "t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0",
+  "epoch": 0,
+  "n_params": 179376136,
+  "args": {
+    "run_name": "t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0",
+    "batch_size": 32,
+    "epochs": 1,
+    "total_tokens": -1,
+    "lossbearing_tokens": 5.2,
+    "accum_iter": 16,
+    "effective_batch_size": 512,
+    "lr_batch_scaling": "none",
+    "save_ckpt_freq": 1,
+    "save_ckpt_freq_tokens": 2,
+    "debug_batch_stats": 0,
+    "model": "fm_small_12d_swiglu_nobias",
+    "dtype": "bfloat16",
+    "loss_type": "token",
+    "context_length": 2048,
+    "pos_emb_ablation": "vanilla",
+    "max_blocks": 32,
+    "intra_pos_emb": "default",
+    "video_pos_3d": false,
+    "finetune": "",
+    "opt": "adamw",
+    "opt_eps": 1e-08,
+    "opt_betas": [
+      0.9,
+      0.95
+    ],
+    "compute_grad_norm": true,
+    "clip_grad": 1.0,
+    "skip_grad": null,
+    "momentum": 0.9,
+    "weight_decay": 0.1,
+    "weight_decay_end": 0.1,
+    "blr": 0.002,
+    "min_blr": 0.0001,
+    "scheduler": "wsd",
+    "decay_fraction": 0.1,
+    "warmup_steps": 250,
+    "cooldown_steps": -1,
+    "data_config": "cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml",
+    "data_config_val": "cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml",
+    "cross_eval_val": "",
+    "s3_endpoint": "",
+    "s3_data_endpoint": "",
+    "s3_multipart_chunksize_mb": 512,
+    "s3_multipart_threshold_mb": 512,
+    "s3_max_io_queue": 100,
+    "text_tokenizer_path": "fourm/utils/tokenizer/trained/smollm_tokenizer.json",
+    "eval_freq": 1,
+    "eval_freq_tokens": 1,
+    "dist_eval": true,
+    "eval": false,
+    "val_sampling_mode": "without_replacement",
+    "output_dir": "output/cc/t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0",
+    "device": "cuda",
+    "seed": 0,
+    "resume": "",
+    "auto_resume": true,
+    "start_epoch": 0,
+    "num_workers": 10,
+    "pin_mem": true,
+    "find_unused_params": true,
+    "rlimit": 4096,
+    "print_all": false,
+    "show_user_warnings": false,
+    "s3_save_dir": "",
+    "dist_url": "env://",
+    "log_wandb": true,
+    "wandb_project": "loom",
+    "wandb_entity": null,
+    "wandb_run_name": "t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0",
+    "wandb_group": "cc-solo",
+    "dim": null,
+    "depth": null,
+    "num_heads": null,
+    "mlp_ratio": null,
+    "num_experts": 1,
+    "granularity": 1,
+    "routing": null,
+    "moe_aux_coef": 0.01,
+    "z_loss_coef": null,
+    "moe_compile": false,
+    "log_grad_norms_every": 0,
+    "save_at_tokens": null,
+    "unigram_entropies": null,
+    "cooldown_branch": false,
+    "cooldown_tokens": 0,
+    "config_path": "cfgs/default/4m/experiments/confound/solo_s/3d/t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0.yaml",
+    "rank": 0,
+    "world_size": 1,
+    "gpu": 0,
+    "distributed": true,
+    "dist_backend": "nccl",
+    "num_tasks": 1,
+    "sampling_mode": "without_replacement",
+    "max_micro_steps": 158692,
+    "lossbearing_target": 5200000000.0,
+    "lossbearing_mode": true,
+    "lr": 0.002,
+    "min_lr": 0.0001,
+    "global_micro_step": 158692,
+    "cum_lossbearing": 5200948224.0,
+    "cum_mod_tokens": {
+      "text": 5538816,
+      "tok_3d": 5195409408
+    }
+  },
+  "source_configs": {
+    "cfgs/default/4m/experiments/confound/solo_s/3d/t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0.yaml": "# S solo variant t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0 (confound program D6-1).\n# Clone of t1/unimodal/3d/run_pure.yaml: D 5.2B, effective batch 512, absolute LR 0.002 (lr_batch_scaling none), seed 0.\n# Self-cooled WSD endpoint; compared with the other D6-1 solo variants of the same modality\n# and D (seed spread, LR x batch grid, warmup / z-loss controls).\n_base_: ../../../../models/dec/4m-s-dec.yaml\nrun_name: t1_pure_3d_s_d5.2_eb512_lr2e-3_seed0\ntotal_tokens: -1\nlossbearing_tokens: 5.2\ndata_config: cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml\ndata_config_val: cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml\nwarmup_steps: 250\neval_freq_tokens: 1\nsave_ckpt_freq_tokens: 2\nwandb_group: cc-solo\neffective_batch_size: 512\nlr_batch_scaling: none\nblr: 0.002\nseed: 0\naccum_iter: 16\nmin_blr: 0.0001\n",
+    "cfgs/default/4m/experiments/t1/unimodal/3d/data_train_pure.yaml": "# T1 pure unimodal training data \u2014 3d\n# Text stripped from all sources (pure modality-only condition)\n\nmodalities:\n- tok_3d\nepoch_size: 50000000\nsampler_type: temperature\ntemperature_tau: 1.0\nsources:\n- name: 3d\n  shards: data/shards/3d_shuffled/train/shard_{000000..000685}.tar\n  weight: 685736\n  modality_tokens:\n    tok_3d: 1875\n  format: npz\nsampling_mode: without_replacement\n",
+    "cfgs/default/4m/experiments/t1/unimodal/3d/data_val_pure.yaml": "# T1 pure unimodal validation data \u2014 3d\n# Text stripped from all sources (pure modality-only condition)\n\nmodalities:\n- tok_3d\nval_steps: 200\nsampler_type: uniform\nsources:\n- name: 3d_val\n  shards: data/shards/3d_shuffled/val/shard_{000000..000013}.tar\n  weight: 14000\n  modality_tokens:\n    tok_3d: 1875\n  format: npz\n"
+  }
+}