Add files using upload-large-folder tool

LTA_openwebtext_dualt/logs/lta_lm1b_dirichlet_len1024_Cv_to_2v_nosep_gbs512_b32_4gpu_20k_save1k_gumbelwatch_20260525_watcher.sh

@@ -0,0 +1,87 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+cd /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt
+export PYTHONPATH="$(pwd)${PYTHONPATH:+:$PYTHONPATH}"
+export TOKENIZERS_PARALLELISM=false
+export PYTHONUNBUFFERED=1
+
+: "${RUN_DIR:?RUN_DIR is required}"
+: "${OUT_BASE:?OUT_BASE is required}"
+: "${LOG_DIR:?LOG_DIR is required}"
+: "${TOKENIZER_PATH:?TOKENIZER_PATH is required}"
+: "${SCORER:?SCORER is required}"
+
+RUN_STEM="$(basename "${RUN_DIR}")"
+TEMP_TAG="${ENDPOINT_TEMP//./p}"
+PROCESSED_FILE="${LOG_DIR}/processed_${RUN_STEM}_steps${STEPS}_c${CMIN}_${CMAX}_gumbel_t${TEMP_TAG}_n${N_SAMPLES}.txt"
+
+mkdir -p "${OUT_BASE}" "${LOG_DIR}"
+touch "${PROCESSED_FILE}"
+
+echo "[watch-gumbel] run_dir=${RUN_DIR}"
+echo "[watch-gumbel] out_base=${OUT_BASE}"
+echo "[watch-gumbel] interval=${STEP_INTERVAL} max_len=${MAX_LEN} steps=${STEPS} c=${CMIN}->${CMAX} temp=${ENDPOINT_TEMP} top_p=${ENDPOINT_TOP_P} tau=${GUMBEL_TAU_START}->${GUMBEL_TAU_END} n=${N_SAMPLES}"
+
+while true; do
+  shopt -s nullglob
+  ckpts=("${RUN_DIR}"/step_*.pt)
+  shopt -u nullglob
+
+  if (( ${#ckpts[@]} == 0 )); then
+    echo "[watch-gumbel] $(date +%F_%T) no ckpt yet"
+    sleep "${SLEEP_SECONDS}"
+    continue
+  fi
+
+  printf "%s\n" "${ckpts[@]}" | sort | while read -r ckpt; do
+    base="$(basename "${ckpt}")"
+    step="${base#step_}"
+    step="${step%.pt}"
+    step_num=$((10#${step}))
+    if (( step_num % STEP_INTERVAL != 0 )); then
+      continue
+    fi
+    if grep -Fxq "${ckpt}" "${PROCESSED_FILE}"; then
+      continue
+    fi
+
+    out_dir="${OUT_BASE}/step_${step}"
+    log_file="${LOG_DIR}/infer_${RUN_STEM}_step_${step}.log"
+    mkdir -p "${out_dir}"
+
+    echo "[watch-gumbel] $(date +%F_%T) infer ${ckpt} -> ${out_dir}" | tee -a "${log_file}"
+    CUDA_VISIBLE_DEVICES="${WATCH_CUDA_VISIBLE_DEVICES}" python scripts/eval_lm1b_c1024_fullycoupled_sde_genppl.py \
+      --checkpoint "${ckpt}" \
+      --tokenizer_path "${TOKENIZER_PATH}" \
+      --scorer "${SCORER}" \
+      --out_dir "${out_dir}" \
+      --n_samples "${N_SAMPLES}" \
+      --max_len "${MAX_LEN}" \
+      --steps "${STEPS}" \
+      --batch_size "${DECODE_BATCH}" \
+      --score_batch "${SCORE_BATCH}" \
+      --score_max_length "${SCORE_MAX_LENGTH}" \
+      --concentration_min "${CMIN}" \
+      --concentration_max "${CMAX}" \
+      --endpoint_temp "${ENDPOINT_TEMP}" \
+      --endpoint_projection gumbel_softmax \
+      --endpoint_top_p "${ENDPOINT_TOP_P}" \
+      --gumbel_tau_start "${GUMBEL_TAU_START}" \
+      --gumbel_tau_end "${GUMBEL_TAU_END}" \
+      --model_t_mode support_t \
+      --mean_mode endpoint_only \
+      --semantic_power 1.0 \
+      --noise_init dirichlet \
+      --noise_dirichlet_concentration "${CMIN}" \
+      --sde_resample dirichlet \
+      --final_from blend_0.5 \
+      --seed 20260524 \
+      2>&1 | tee -a "${log_file}"
+
+    echo "${ckpt}" >> "${PROCESSED_FILE}"
+    echo "[watch-gumbel] $(date +%F_%T) done step_${step}" | tee -a "${log_file}"
+  done
+
+  sleep "${SLEEP_SECONDS}"
+done

LTA_openwebtext_dualt/logs/lta_owt_bert_absrope_time4_dirichlet_len1024_Cv_to_2v_mask1_sameT_gbs512_b32_8gpu_1m_save10k_20260525_watcher.pid

@@ -0,0 +1 @@
+10250

LTA_openwebtext_dualt/logs/lta_owt_gpt2cached_len1024_p05_b64_ckpt_bench4gpu_20260513_152918.log

@@ -0,0 +1,523 @@
+
+*****************************************
+Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
+*****************************************
+NCCL version 2.25.1+cuda12.8
+{
+  "device": "cuda:0",
+  "rank": 0,
+  "world_size": 4,
+  "samples": "owt_cached_chunks:8734897",
+  "vocab_size": 50257,
+  "tokenizer_vocab_size": 50257,
+  "save_dir": "runs/lta_owt_gpt2cached_len1024_p05_b64_ckpt_bench4gpu_20260513_152918",
+  "batch_size": 64,
+  "grad_accum": 2,
+  "effective_batch_size": 512,
+  "global_batch_size": 512,
+  "lr_schedule": "cosine",
+  "optimizer": "adamw",
+  "warmup_steps": 3,
+  "min_lr": 6e-05,
+  "weight_decay": 0.1,
+  "adamw_param_groups": "nanogpt",
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.95,
+  "adam_eps": 1e-08,
+  "muon_momentum": 0.95,
+  "muon_ns_steps": 5,
+  "muon_update_scale": 1.0,
+  "ema_decay": 0.0,
+  "ema_start_step": 0,
+  "model_type": "ddit",
+  "dual_t": true,
+  "corrupt_t_mode": "same",
+  "corrupt_min_t": 0.0,
+  "corrupt_max_t": 1.0,
+  "prefix_block_prob": 0.0,
+  "prefix_block_len": 128,
+  "dirichlet_endpoint_mode": "categorical_dual_t",
+  "dirichlet_semantic_t_mode": "same",
+  "dirichlet_semantic_t_value": 0.0,
+  "categorical_wrong_from_full_vocab": true,
+  "categorical_wrong_from_batch_valid_tokens": false,
+  "mask_mixture_original_prob": 0.0,
+  "mask_mixture_lowk_prob": 0.0,
+  "mask_mixture_lowcorrupt_prob": 0.0,
+  "mask_mixture_block_prob": 0.0,
+  "mask_mixture_all_prob": 0.0,
+  "mask_mixture_lowk_clean_tokens": "1,2,4,8,16,32,64",
+  "mask_mixture_lowcorrupt_tokens": "1,2,4,8,16,32,64",
+  "mask_mixture_block_tokens": "64,128",
+  "simplex_bridge_sampler": "dirichlet",
+  "logistic_normal_sigma_min": 0.18,
+  "logistic_normal_sigma_max": 2.2,
+  "logistic_normal_tau_min": 0.65,
+  "logistic_normal_tau_max": 1.15,
+  "torch_compile": false,
+  "compile_mode": "max-autotune",
+  "state_format": "prob",
+  "target_loss": "hard_ce",
+  "meanflow_weight": 0.0,
+  "rollout_train_prob": 0.5,
+  "rollout_train_steps": 1,
+  "rollout_train_infer_steps": 64,
+  "rollout_train_temp": 1.45,
+  "rollout_train_max_gamma": 1.0,
+  "rollout_train_corrupt_only": true,
+  "rollout_train_samplewise": false,
+  "rollout_train_compute_always": false,
+  "bridge_noise_init": "logistic_normal",
+  "noise_sigma": -1.0,
+  "allow_tf32": true,
+  "activation_checkpointing": true,
+  "activation_checkpoint_interval": 1,
+  "ddp_static_graph": false,
+  "ddp_gradient_as_bucket_view": true,
+  "blocking_data_transfer": false,
+  "dataloader_prefetch_factor": 4,
+  "full_train_stats": false,
+  "record_pad_truncate": false,
+  "record_add_eos": false,
+  "record_add_special_tokens": false,
+  "record_pad_token": "pad",
+  "record_shuffle_buffer": 10000,
+  "wrap": true,
+  "wrap_mode": "stream",
+  "wrap_record_buffer_size": 200,
+  "owt_cached_chunks": true,
+  "owt_chunk_cache_dir": "/e2e-data/evad-tech-vla/wanghan58/data/small_benchmarks/langflow_2604_11748/openwebtext_lta_cached_chunks/gpt2_len1024_train_minus_100k",
+  "owt_chunk_cache_rebuild": false,
+  "owt_chunk_cache_write_batch": 4096,
+  "owt_exact_repeat_per_chunk": 0,
+  "online_chunk_shuffle": false,
+  "online_chunk_shuffle_buffer": 10000,
+  "openwebtext_split": "train_minus_100k",
+  "detokenizer": "auto",
+  "resolved_detokenizer": null,
+  "num_workers": 8,
+  "latest_every": 100000,
+  "resume_path": ""
+}
+[rank0]: Traceback (most recent call last):
+[rank0]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1540, in <module>
+[rank0]:     main()
+[rank0]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1459, in main
+[rank0]:     loss.backward()
+[rank0]:   File "/usr/local/lib/python3.12/dist-packages/torch/_tensor.py", line 648, in backward
+[rank0]:     torch.autograd.backward(
+[rank0]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/__init__.py", line 347, in backward
+[rank0]:     _engine_run_backward(
+[rank0]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/graph.py", line 823, in _engine_run_backward
+[rank0]:     return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[rank0]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+[rank0]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 1128, in unpack_hook
+[rank0]:     frame.check_recomputed_tensors_match(gid)
+[rank0]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 902, in check_recomputed_tensors_match
+[rank0]:     raise CheckpointError(
+[rank0]: torch.utils.checkpoint.CheckpointError: torch.utils.checkpoint: Recomputed values for the following tensors have different metadata than during the forward pass.
+[rank0]: tensor at position 0:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 128]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 1:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 3:
+[rank0]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 4:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 5:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 6:
+[rank0]: saved metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 7:
+[rank0]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 8:
+[rank0]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 9:
+[rank0]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 11:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 12:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 13:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 15:
+[rank0]: saved metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 16:
+[rank0]: saved metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 17:
+[rank0]: saved metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 18:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 19:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 20:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 21:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 22:
+[rank0]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 23:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 24:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 25:
+[rank0]: saved metadata: {'shape': torch.Size([768, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 26:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1024, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 27:
+[rank0]: saved metadata: {'shape': torch.Size([3072, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=0)}
+[rank0]: tensor at position 28:
+[rank0]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+[rank0]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=0)}
+
+[rank1]: Traceback (most recent call last):
+[rank1]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1540, in <module>
+[rank1]:     main()
+[rank1]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1459, in main
+[rank1]:     loss.backward()
+[rank1]:   File "/usr/local/lib/python3.12/dist-packages/torch/_tensor.py", line 648, in backward
+[rank1]:     torch.autograd.backward(
+[rank1]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/__init__.py", line 347, in backward
+[rank1]:     _engine_run_backward(
+[rank1]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/graph.py", line 823, in _engine_run_backward
+[rank1]:     return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[rank1]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+[rank1]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 1128, in unpack_hook
+[rank1]:     frame.check_recomputed_tensors_match(gid)
+[rank1]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 902, in check_recomputed_tensors_match
+[rank1]:     raise CheckpointError(
+[rank1]: torch.utils.checkpoint.CheckpointError: torch.utils.checkpoint: Recomputed values for the following tensors have different metadata than during the forward pass.
+[rank1]: tensor at position 0:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 128]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 1:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 3:
+[rank1]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 4:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 5:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 6:
+[rank1]: saved metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 7:
+[rank1]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 8:
+[rank1]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 9:
+[rank1]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 11:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 12:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 13:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 15:
+[rank1]: saved metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 16:
+[rank1]: saved metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 17:
+[rank1]: saved metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 18:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 19:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 20:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 21:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 22:
+[rank1]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 23:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 24:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 25:
+[rank1]: saved metadata: {'shape': torch.Size([768, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 26:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1024, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 27:
+[rank1]: saved metadata: {'shape': torch.Size([3072, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=1)}
+[rank1]: tensor at position 28:
+[rank1]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+[rank1]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=1)}
+
+[rank2]: Traceback (most recent call last):
+[rank2]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1540, in <module>
+[rank2]:     main()
+[rank2]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1459, in main
+[rank2]:     loss.backward()
+[rank2]:   File "/usr/local/lib/python3.12/dist-packages/torch/_tensor.py", line 648, in backward
+[rank2]:     torch.autograd.backward(
+[rank2]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/__init__.py", line 347, in backward
+[rank2]:     _engine_run_backward(
+[rank2]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/graph.py", line 823, in _engine_run_backward
+[rank2]:     return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[rank2]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+[rank2]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 1128, in unpack_hook
+[rank2]:     frame.check_recomputed_tensors_match(gid)
+[rank2]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 902, in check_recomputed_tensors_match
+[rank2]:     raise CheckpointError(
+[rank2]: torch.utils.checkpoint.CheckpointError: torch.utils.checkpoint: Recomputed values for the following tensors have different metadata than during the forward pass.
+[rank2]: tensor at position 0:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 128]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 1:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 3:
+[rank2]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 4:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 5:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 6:
+[rank2]: saved metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 7:
+[rank2]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 8:
+[rank2]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 9:
+[rank2]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 11:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 12:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 13:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 15:
+[rank2]: saved metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 16:
+[rank2]: saved metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 17:
+[rank2]: saved metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 18:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 19:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 20:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 21:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 22:
+[rank2]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 23:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 24:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 25:
+[rank2]: saved metadata: {'shape': torch.Size([768, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 26:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1024, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 27:
+[rank2]: saved metadata: {'shape': torch.Size([3072, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=2)}
+[rank2]: tensor at position 28:
+[rank2]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+[rank2]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=2)}
+
+[rank3]: Traceback (most recent call last):
+[rank3]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1540, in <module>
+[rank3]:     main()
+[rank3]:   File "/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/train.py", line 1459, in main
+[rank3]:     loss.backward()
+[rank3]:   File "/usr/local/lib/python3.12/dist-packages/torch/_tensor.py", line 648, in backward
+[rank3]:     torch.autograd.backward(
+[rank3]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/__init__.py", line 347, in backward
+[rank3]:     _engine_run_backward(
+[rank3]:   File "/usr/local/lib/python3.12/dist-packages/torch/autograd/graph.py", line 823, in _engine_run_backward
+[rank3]:     return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[rank3]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+[rank3]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 1128, in unpack_hook
+[rank3]:     frame.check_recomputed_tensors_match(gid)
+[rank3]:   File "/usr/local/lib/python3.12/dist-packages/torch/utils/checkpoint.py", line 902, in check_recomputed_tensors_match
+[rank3]:     raise CheckpointError(
+[rank3]: torch.utils.checkpoint.CheckpointError: torch.utils.checkpoint: Recomputed values for the following tensors have different metadata than during the forward pass.
+[rank3]: tensor at position 0:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 128]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 1:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 3:
+[rank3]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 4:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 5:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 6:
+[rank3]: saved metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 7:
+[rank3]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 8:
+[rank3]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([768, 2304]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 9:
+[rank3]: saved metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 11:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 12:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 13:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([1, 1, 1024, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 15:
+[rank3]: saved metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 16:
+[rank3]: saved metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 17:
+[rank3]: saved metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 12, 64]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 18:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 12, 1024]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 19:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([2]), 'dtype': torch.int64, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 20:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([768, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 21:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([65536, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 22:
+[rank3]: saved metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 23:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 24:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 25:
+[rank3]: saved metadata: {'shape': torch.Size([768, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 26:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1024, 3072]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 1]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 27:
+[rank3]: saved metadata: {'shape': torch.Size([3072, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([1, 1, 768]), 'dtype': torch.float32, 'device': device(type='cuda', index=3)}
+[rank3]: tensor at position 28:
+[rank3]: saved metadata: {'shape': torch.Size([64, 1, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+[rank3]: recomputed metadata: {'shape': torch.Size([64, 1024, 768]), 'dtype': torch.bfloat16, 'device': device(type='cuda', index=3)}
+
+[rank0]:[W513 15:30:07.195396563 ProcessGroupNCCL.cpp:1487] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
+W0513 15:30:08.147000 637993 torch/distributed/elastic/multiprocessing/api.py:898] Sending process 638060 closing signal SIGTERM
+W0513 15:30:08.148000 637993 torch/distributed/elastic/multiprocessing/api.py:898] Sending process 638062 closing signal SIGTERM
+W0513 15:30:08.148000 637993 torch/distributed/elastic/multiprocessing/api.py:898] Sending process 638063 closing signal SIGTERM
+E0513 15:30:08.377000 637993 torch/distributed/elastic/multiprocessing/api.py:870] failed (exitcode: 1) local_rank: 1 (pid: 638061) of binary: /usr/bin/python
+Traceback (most recent call last):
+  File "<frozen runpy>", line 198, in _run_module_as_main
+  File "<frozen runpy>", line 88, in _run_code
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 922, in <module>
+    main()
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
+    return f(*args, **kwargs)
+           ^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 918, in main
+    run(args)
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 909, in run
+    elastic_launch(
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
+    return launch_agent(self._config, self._entrypoint, list(args))
+           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
+    raise ChildFailedError(
+torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
+============================================================
+train.py FAILED
+------------------------------------------------------------
+Failures:
+  <NO_OTHER_FAILURES>
+------------------------------------------------------------
+Root Cause (first observed failure):
+[0]:
+  time      : 2026-05-13_15:30:08
+  host      : localhost
+  rank      : 1 (local_rank: 1)
+  exitcode  : 1 (pid: 638061)
+  error_file: <N/A>
+  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
+============================================================

LTA_openwebtext_dualt/logs/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_watch/infer_lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525_step_0029000.log

@@ -0,0 +1,199 @@
+[watch-dualline] 2026-05-26_10:00:59 infer runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0029000.pt -> docs/lta_samples/metrics_20260525/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_dirres_c1_1024_n128/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0029000
+[decode] max_len=128 generated=1/128
+[decode] max_len=128 generated=2/128
+[decode] max_len=128 generated=3/128
+[decode] max_len=128 generated=4/128
+[decode] max_len=128 generated=5/128
+[decode] max_len=128 generated=6/128
+[decode] max_len=128 generated=7/128
+[decode] max_len=128 generated=8/128
+[decode] max_len=128 generated=9/128
+[decode] max_len=128 generated=10/128
+[decode] max_len=128 generated=11/128
+[decode] max_len=128 generated=12/128
+[decode] max_len=128 generated=13/128
+[decode] max_len=128 generated=14/128
+[decode] max_len=128 generated=15/128
+[decode] max_len=128 generated=16/128
+[decode] max_len=128 generated=17/128
+[decode] max_len=128 generated=18/128
+[decode] max_len=128 generated=19/128
+[decode] max_len=128 generated=20/128
+[decode] max_len=128 generated=21/128
+[decode] max_len=128 generated=22/128
+[decode] max_len=128 generated=23/128
+[decode] max_len=128 generated=24/128
+[decode] max_len=128 generated=25/128
+[decode] max_len=128 generated=26/128
+[decode] max_len=128 generated=27/128
+[decode] max_len=128 generated=28/128
+[decode] max_len=128 generated=29/128
+[decode] max_len=128 generated=30/128
+[decode] max_len=128 generated=31/128
+[decode] max_len=128 generated=32/128
+[decode] max_len=128 generated=33/128
+[decode] max_len=128 generated=34/128
+[decode] max_len=128 generated=35/128
+[decode] max_len=128 generated=36/128
+[decode] max_len=128 generated=37/128
+[decode] max_len=128 generated=38/128
+[decode] max_len=128 generated=39/128
+[decode] max_len=128 generated=40/128
+[decode] max_len=128 generated=41/128
+[decode] max_len=128 generated=42/128
+[decode] max_len=128 generated=43/128
+[decode] max_len=128 generated=44/128
+[decode] max_len=128 generated=45/128
+[decode] max_len=128 generated=46/128
+[decode] max_len=128 generated=47/128
+[decode] max_len=128 generated=48/128
+[decode] max_len=128 generated=49/128
+[decode] max_len=128 generated=50/128
+[decode] max_len=128 generated=51/128
+[decode] max_len=128 generated=52/128
+[decode] max_len=128 generated=53/128
+[decode] max_len=128 generated=54/128
+[decode] max_len=128 generated=55/128
+[decode] max_len=128 generated=56/128
+[decode] max_len=128 generated=57/128
+[decode] max_len=128 generated=58/128
+[decode] max_len=128 generated=59/128
+[decode] max_len=128 generated=60/128
+[decode] max_len=128 generated=61/128
+[decode] max_len=128 generated=62/128
+[decode] max_len=128 generated=63/128
+[decode] max_len=128 generated=64/128
+[decode] max_len=128 generated=65/128
+[decode] max_len=128 generated=66/128
+[decode] max_len=128 generated=67/128
+[decode] max_len=128 generated=68/128
+[decode] max_len=128 generated=69/128
+[decode] max_len=128 generated=70/128
+[decode] max_len=128 generated=71/128
+[decode] max_len=128 generated=72/128
+[decode] max_len=128 generated=73/128
+[decode] max_len=128 generated=74/128
+[decode] max_len=128 generated=75/128
+[decode] max_len=128 generated=76/128
+[decode] max_len=128 generated=77/128
+[decode] max_len=128 generated=78/128
+[decode] max_len=128 generated=79/128
+[decode] max_len=128 generated=80/128
+[decode] max_len=128 generated=81/128
+[decode] max_len=128 generated=82/128
+[decode] max_len=128 generated=83/128
+[decode] max_len=128 generated=84/128
+[decode] max_len=128 generated=85/128
+[decode] max_len=128 generated=86/128
+[decode] max_len=128 generated=87/128
+[decode] max_len=128 generated=88/128
+[decode] max_len=128 generated=89/128
+[decode] max_len=128 generated=90/128
+[decode] max_len=128 generated=91/128
+[decode] max_len=128 generated=92/128
+[decode] max_len=128 generated=93/128
+[decode] max_len=128 generated=94/128
+[decode] max_len=128 generated=95/128
+[decode] max_len=128 generated=96/128
+[decode] max_len=128 generated=97/128
+[decode] max_len=128 generated=98/128
+[decode] max_len=128 generated=99/128
+[decode] max_len=128 generated=100/128
+[decode] max_len=128 generated=101/128
+[decode] max_len=128 generated=102/128
+[decode] max_len=128 generated=103/128
+[decode] max_len=128 generated=104/128
+[decode] max_len=128 generated=105/128
+[decode] max_len=128 generated=106/128
+[decode] max_len=128 generated=107/128
+[decode] max_len=128 generated=108/128
+[decode] max_len=128 generated=109/128
+[decode] max_len=128 generated=110/128
+[decode] max_len=128 generated=111/128
+[decode] max_len=128 generated=112/128
+[decode] max_len=128 generated=113/128
+[decode] max_len=128 generated=114/128
+[decode] max_len=128 generated=115/128
+[decode] max_len=128 generated=116/128
+[decode] max_len=128 generated=117/128
+[decode] max_len=128 generated=118/128
+[decode] max_len=128 generated=119/128
+[decode] max_len=128 generated=120/128
+[decode] max_len=128 generated=121/128
+[decode] max_len=128 generated=122/128
+[decode] max_len=128 generated=123/128
+[decode] max_len=128 generated=124/128
+[decode] max_len=128 generated=125/128
+[decode] max_len=128 generated=126/128
+[decode] max_len=128 generated=127/128
+[decode] max_len=128 generated=128/128
+[
+  {
+    "checkpoint": "runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0029000.pt",
+    "ckpt_step": 29000,
+    "max_len": 128,
+    "decode_rule": "dual_line_resample",
+    "support_power": 1.0,
+    "semantic_power": 1.5,
+    "steps": 128,
+    "c_min": 1.0,
+    "c_max": 1024.0,
+    "anchor_mode": "state",
+    "model_t_mode": "flow",
+    "time_schedule": "uniform",
+    "time_logit_mean": -1.5,
+    "time_logit_std": 0.8,
+    "time_power": 2.0,
+    "input_noise_scale": 0.0,
+    "input_noise_until": 1.0,
+    "input_noise_dirichlet_concentration": 1.0,
+    "endpoint_softening": "none",
+    "endpoint_soft_power": 2.0,
+    "endpoint_soft_min_conf": 0.0,
+    "endpoint_soft_max_conf": 1.0,
+    "soft_target_decode_mode": "off",
+    "soft_target_power": 1.0,
+    "soft_target_min_conf": 0.0,
+    "soft_target_max_conf": 1.0,
+    "soft_target_debias_start": 0.7,
+    "final_from": "blend",
+    "final_decode": "argmax",
+    "final_sample_temp": 1.0,
+    "final_top_k": 0,
+    "final_top_p": 1.0,
+    "commit_mode": "off",
+    "commit_conf_threshold": 0.0,
+    "commit_margin_threshold": 0.0,
+    "commit_start": 0.0,
+    "commit_min_ratio": 0.0,
+    "commit_max_ratio": 1.0,
+    "commit_power": 2.0,
+    "commit_freq_max_frac": 0.08,
+    "early_temp": 2.8,
+    "late_temp": 1.45,
+    "temp_end": 0.55,
+    "temp_power": 1.5,
+    "pos_extend": "repeat",
+    "fixed_first_token_id": null,
+    "fixed_first_token_text": "",
+    "fixed_first_initial_argmax": false,
+    "use_ema": false,
+    "n_samples": 128,
+    "sample_entropy": 1.7398678565049452,
+    "unique_tokens": 701,
+    "token_count": 16384,
+    "distinct_1": 0.04278564453125,
+    "distinct_2": 0.15483513779527558,
+    "top_token_mass": 0.3388671875,
+    "texts_preview": [
+      "[CLS], \", \" sbs \", \" sbs, sbss. \" he was still upset, he added,, \" sbs, \", \" sbs, that the sheriff never changed the situation. \" sbs, according to the sheriff ' s office at the time, \" sbs \", \" sbs, \", \" sbs, and sbs, \" \" sbs, \", \" d. c. \", \" sbs \", \" sbs \", \" sbs, \", \" sbs \", \" sbs \", \" sbs, \", \" sbs, i. d., \", \" sbs, \" sbs, \", \" [SEP]",
+      "[CLS]ja kojaja koja koja ko kon koja koja ko naja kojajau kojaja koja kozja ko koyja na koja koja koja narja zaz zaya najajajau zak zazja najaya za najaja zaya najajakka koya kokuna kojajak koja kovjaja kovjaja koka koya koja ko koka kot koya koya koan koya ko kojajak kojaja koja kojajaya koka [SEP]",
+      "[CLS] ko kojaka koya koja naja na koka koja na koya koja ko ko koya koka koya koya na koyaya kojaja koyaka na koka ni kona koz nika koka nila na na nita nina naka zak naka na nina na naku naba na na nala na nina naka naja naja na naka na naja naka jaz na kola koka ko koka na koka koya na na nau na ko na ni nija nijala na za na [SEP]",
+      "[CLS] kok koja ko koja koka ko koja ko koja koja ko koja kona koja koja ko ko koja ko kokaja ko ko koja koyaja koja koja ko koja koja koja koja koja koya koja koya koja kok koja koka ko koja koja ko koja ko kok koja ko ko kojaja ko kok ko koja koka kojak ko kokaja koka ko ko koja ko koja koya ko koya ko ko ko koja ko koya ko [SEP]"
+    ],
+    "gen_ppl": 14.536488686310422,
+    "gen_nll": 2.676661949564978,
+    "gen_tokens": 15736
+  }
+]
+[watch-dualline] 2026-05-26_10:08:19 done step_0029000

LTA_openwebtext_dualt/logs/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_watch/infer_lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525_step_0032000.log

@@ -0,0 +1,199 @@
+[watch-dualline] 2026-05-26_11:16:23 infer runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0032000.pt -> docs/lta_samples/metrics_20260525/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_dirres_c1_1024_n128/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0032000
+[decode] max_len=128 generated=1/128
+[decode] max_len=128 generated=2/128
+[decode] max_len=128 generated=3/128
+[decode] max_len=128 generated=4/128
+[decode] max_len=128 generated=5/128
+[decode] max_len=128 generated=6/128
+[decode] max_len=128 generated=7/128
+[decode] max_len=128 generated=8/128
+[decode] max_len=128 generated=9/128
+[decode] max_len=128 generated=10/128
+[decode] max_len=128 generated=11/128
+[decode] max_len=128 generated=12/128
+[decode] max_len=128 generated=13/128
+[decode] max_len=128 generated=14/128
+[decode] max_len=128 generated=15/128
+[decode] max_len=128 generated=16/128
+[decode] max_len=128 generated=17/128
+[decode] max_len=128 generated=18/128
+[decode] max_len=128 generated=19/128
+[decode] max_len=128 generated=20/128
+[decode] max_len=128 generated=21/128
+[decode] max_len=128 generated=22/128
+[decode] max_len=128 generated=23/128
+[decode] max_len=128 generated=24/128
+[decode] max_len=128 generated=25/128
+[decode] max_len=128 generated=26/128
+[decode] max_len=128 generated=27/128
+[decode] max_len=128 generated=28/128
+[decode] max_len=128 generated=29/128
+[decode] max_len=128 generated=30/128
+[decode] max_len=128 generated=31/128
+[decode] max_len=128 generated=32/128
+[decode] max_len=128 generated=33/128
+[decode] max_len=128 generated=34/128
+[decode] max_len=128 generated=35/128
+[decode] max_len=128 generated=36/128
+[decode] max_len=128 generated=37/128
+[decode] max_len=128 generated=38/128
+[decode] max_len=128 generated=39/128
+[decode] max_len=128 generated=40/128
+[decode] max_len=128 generated=41/128
+[decode] max_len=128 generated=42/128
+[decode] max_len=128 generated=43/128
+[decode] max_len=128 generated=44/128
+[decode] max_len=128 generated=45/128
+[decode] max_len=128 generated=46/128
+[decode] max_len=128 generated=47/128
+[decode] max_len=128 generated=48/128
+[decode] max_len=128 generated=49/128
+[decode] max_len=128 generated=50/128
+[decode] max_len=128 generated=51/128
+[decode] max_len=128 generated=52/128
+[decode] max_len=128 generated=53/128
+[decode] max_len=128 generated=54/128
+[decode] max_len=128 generated=55/128
+[decode] max_len=128 generated=56/128
+[decode] max_len=128 generated=57/128
+[decode] max_len=128 generated=58/128
+[decode] max_len=128 generated=59/128
+[decode] max_len=128 generated=60/128
+[decode] max_len=128 generated=61/128
+[decode] max_len=128 generated=62/128
+[decode] max_len=128 generated=63/128
+[decode] max_len=128 generated=64/128
+[decode] max_len=128 generated=65/128
+[decode] max_len=128 generated=66/128
+[decode] max_len=128 generated=67/128
+[decode] max_len=128 generated=68/128
+[decode] max_len=128 generated=69/128
+[decode] max_len=128 generated=70/128
+[decode] max_len=128 generated=71/128
+[decode] max_len=128 generated=72/128
+[decode] max_len=128 generated=73/128
+[decode] max_len=128 generated=74/128
+[decode] max_len=128 generated=75/128
+[decode] max_len=128 generated=76/128
+[decode] max_len=128 generated=77/128
+[decode] max_len=128 generated=78/128
+[decode] max_len=128 generated=79/128
+[decode] max_len=128 generated=80/128
+[decode] max_len=128 generated=81/128
+[decode] max_len=128 generated=82/128
+[decode] max_len=128 generated=83/128
+[decode] max_len=128 generated=84/128
+[decode] max_len=128 generated=85/128
+[decode] max_len=128 generated=86/128
+[decode] max_len=128 generated=87/128
+[decode] max_len=128 generated=88/128
+[decode] max_len=128 generated=89/128
+[decode] max_len=128 generated=90/128
+[decode] max_len=128 generated=91/128
+[decode] max_len=128 generated=92/128
+[decode] max_len=128 generated=93/128
+[decode] max_len=128 generated=94/128
+[decode] max_len=128 generated=95/128
+[decode] max_len=128 generated=96/128
+[decode] max_len=128 generated=97/128
+[decode] max_len=128 generated=98/128
+[decode] max_len=128 generated=99/128
+[decode] max_len=128 generated=100/128
+[decode] max_len=128 generated=101/128
+[decode] max_len=128 generated=102/128
+[decode] max_len=128 generated=103/128
+[decode] max_len=128 generated=104/128
+[decode] max_len=128 generated=105/128
+[decode] max_len=128 generated=106/128
+[decode] max_len=128 generated=107/128
+[decode] max_len=128 generated=108/128
+[decode] max_len=128 generated=109/128
+[decode] max_len=128 generated=110/128
+[decode] max_len=128 generated=111/128
+[decode] max_len=128 generated=112/128
+[decode] max_len=128 generated=113/128
+[decode] max_len=128 generated=114/128
+[decode] max_len=128 generated=115/128
+[decode] max_len=128 generated=116/128
+[decode] max_len=128 generated=117/128
+[decode] max_len=128 generated=118/128
+[decode] max_len=128 generated=119/128
+[decode] max_len=128 generated=120/128
+[decode] max_len=128 generated=121/128
+[decode] max_len=128 generated=122/128
+[decode] max_len=128 generated=123/128
+[decode] max_len=128 generated=124/128
+[decode] max_len=128 generated=125/128
+[decode] max_len=128 generated=126/128
+[decode] max_len=128 generated=127/128
+[decode] max_len=128 generated=128/128
+[
+  {
+    "checkpoint": "runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0032000.pt",
+    "ckpt_step": 32000,
+    "max_len": 128,
+    "decode_rule": "dual_line_resample",
+    "support_power": 1.0,
+    "semantic_power": 1.5,
+    "steps": 128,
+    "c_min": 1.0,
+    "c_max": 1024.0,
+    "anchor_mode": "state",
+    "model_t_mode": "flow",
+    "time_schedule": "uniform",
+    "time_logit_mean": -1.5,
+    "time_logit_std": 0.8,
+    "time_power": 2.0,
+    "input_noise_scale": 0.0,
+    "input_noise_until": 1.0,
+    "input_noise_dirichlet_concentration": 1.0,
+    "endpoint_softening": "none",
+    "endpoint_soft_power": 2.0,
+    "endpoint_soft_min_conf": 0.0,
+    "endpoint_soft_max_conf": 1.0,
+    "soft_target_decode_mode": "off",
+    "soft_target_power": 1.0,
+    "soft_target_min_conf": 0.0,
+    "soft_target_max_conf": 1.0,
+    "soft_target_debias_start": 0.7,
+    "final_from": "blend",
+    "final_decode": "argmax",
+    "final_sample_temp": 1.0,
+    "final_top_k": 0,
+    "final_top_p": 1.0,
+    "commit_mode": "off",
+    "commit_conf_threshold": 0.0,
+    "commit_margin_threshold": 0.0,
+    "commit_start": 0.0,
+    "commit_min_ratio": 0.0,
+    "commit_max_ratio": 1.0,
+    "commit_power": 2.0,
+    "commit_freq_max_frac": 0.08,
+    "early_temp": 2.8,
+    "late_temp": 1.45,
+    "temp_end": 0.55,
+    "temp_power": 1.5,
+    "pos_extend": "repeat",
+    "fixed_first_token_id": null,
+    "fixed_first_token_text": "",
+    "fixed_first_initial_argmax": false,
+    "use_ema": false,
+    "n_samples": 128,
+    "sample_entropy": 3.459911790682485,
+    "unique_tokens": 2018,
+    "token_count": 16384,
+    "distinct_1": 0.1231689453125,
+    "distinct_2": 0.5042445866141733,
+    "top_token_mass": 0.09564208984375,
+    "texts_preview": [
+      "[CLS] happen to the girl, \" said the holy church solicitors. \" he was kathy ' s wife. then, she told them she told the court, that the defendant refused to support the daughter. a powerful friend of the defendant ' s son ' s wife, dr. yami, received a divorce complaint, broke the father out of his hospital when he tried to destroy him. in june 2012, the defendant called kathy and her parents, \" but dropped \", and sent a bible letter to them, he went to the hospital and fled from a prison home. two weeks later, the defendant attacked his wife and attacked him [SEP]",
+      "[CLS] of the thin dough. additionally, low - yield the dough for the emg and heavy - liquid, and then use it. 3 30. 25. 4 30. 7. 5 9. 25. 14. 4., and add the butter flour ( emg ), the thin sugar, and the thin dough for the dough. remove the soft strain and a thin dough. bring the dough too thin, and add a thin thin and thin. 6. 5. 6. thick, soft. 9. the 25 - inch knife - based dough provides the pure thin strain. for the dough, the soft strain provides the [SEP]",
+      "[CLS] note that the expression is an inline expression as a expression. ( i also add the expression to a function of a expression expression in the expression, and an inline expression ). i like to add a representation expression of a expression, which is a representation expression of expression, which is to denote a expression. the expression is an extension of the expression method of a expression ’ s expression. the expression method in which a refers to the representation expression of expression and can be applied a expression to.... instead, it ’ s a representation expression expression in a function in the expression, but in the expression that expression is a expression [SEP]",
+      "[CLS]va, tumini, odago, niliva, pono preci possivii, riccoliva, nao nistiliva tuscari muvoo tuminii cissili. mograla segova, nissidini, cine. 19. yavoo tustia, pedno, nissituva pevoli cicconigi, kadno o nivii, nici preccoli, naviilino, nissidituci. snuliva na nija nemoribuci muvii [SEP]"
+    ],
+    "gen_ppl": 53.64963070329939,
+    "gen_nll": 3.982474585610321,
+    "gen_tokens": 15903
+  }
+]
+[watch-dualline] 2026-05-26_11:25:07 done step_0032000

LTA_openwebtext_dualt/logs/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_watch/infer_lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525_step_0036000.log

@@ -0,0 +1,199 @@
+[watch-dualline] 2026-05-26_13:16:52 infer runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0036000.pt -> docs/lta_samples/metrics_20260525/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_dirres_c1_1024_n128/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0036000
+[decode] max_len=128 generated=1/128
+[decode] max_len=128 generated=2/128
+[decode] max_len=128 generated=3/128
+[decode] max_len=128 generated=4/128
+[decode] max_len=128 generated=5/128
+[decode] max_len=128 generated=6/128
+[decode] max_len=128 generated=7/128
+[decode] max_len=128 generated=8/128
+[decode] max_len=128 generated=9/128
+[decode] max_len=128 generated=10/128
+[decode] max_len=128 generated=11/128
+[decode] max_len=128 generated=12/128
+[decode] max_len=128 generated=13/128
+[decode] max_len=128 generated=14/128
+[decode] max_len=128 generated=15/128
+[decode] max_len=128 generated=16/128
+[decode] max_len=128 generated=17/128
+[decode] max_len=128 generated=18/128
+[decode] max_len=128 generated=19/128
+[decode] max_len=128 generated=20/128
+[decode] max_len=128 generated=21/128
+[decode] max_len=128 generated=22/128
+[decode] max_len=128 generated=23/128
+[decode] max_len=128 generated=24/128
+[decode] max_len=128 generated=25/128
+[decode] max_len=128 generated=26/128
+[decode] max_len=128 generated=27/128
+[decode] max_len=128 generated=28/128
+[decode] max_len=128 generated=29/128
+[decode] max_len=128 generated=30/128
+[decode] max_len=128 generated=31/128
+[decode] max_len=128 generated=32/128
+[decode] max_len=128 generated=33/128
+[decode] max_len=128 generated=34/128
+[decode] max_len=128 generated=35/128
+[decode] max_len=128 generated=36/128
+[decode] max_len=128 generated=37/128
+[decode] max_len=128 generated=38/128
+[decode] max_len=128 generated=39/128
+[decode] max_len=128 generated=40/128
+[decode] max_len=128 generated=41/128
+[decode] max_len=128 generated=42/128
+[decode] max_len=128 generated=43/128
+[decode] max_len=128 generated=44/128
+[decode] max_len=128 generated=45/128
+[decode] max_len=128 generated=46/128
+[decode] max_len=128 generated=47/128
+[decode] max_len=128 generated=48/128
+[decode] max_len=128 generated=49/128
+[decode] max_len=128 generated=50/128
+[decode] max_len=128 generated=51/128
+[decode] max_len=128 generated=52/128
+[decode] max_len=128 generated=53/128
+[decode] max_len=128 generated=54/128
+[decode] max_len=128 generated=55/128
+[decode] max_len=128 generated=56/128
+[decode] max_len=128 generated=57/128
+[decode] max_len=128 generated=58/128
+[decode] max_len=128 generated=59/128
+[decode] max_len=128 generated=60/128
+[decode] max_len=128 generated=61/128
+[decode] max_len=128 generated=62/128
+[decode] max_len=128 generated=63/128
+[decode] max_len=128 generated=64/128
+[decode] max_len=128 generated=65/128
+[decode] max_len=128 generated=66/128
+[decode] max_len=128 generated=67/128
+[decode] max_len=128 generated=68/128
+[decode] max_len=128 generated=69/128
+[decode] max_len=128 generated=70/128
+[decode] max_len=128 generated=71/128
+[decode] max_len=128 generated=72/128
+[decode] max_len=128 generated=73/128
+[decode] max_len=128 generated=74/128
+[decode] max_len=128 generated=75/128
+[decode] max_len=128 generated=76/128
+[decode] max_len=128 generated=77/128
+[decode] max_len=128 generated=78/128
+[decode] max_len=128 generated=79/128
+[decode] max_len=128 generated=80/128
+[decode] max_len=128 generated=81/128
+[decode] max_len=128 generated=82/128
+[decode] max_len=128 generated=83/128
+[decode] max_len=128 generated=84/128
+[decode] max_len=128 generated=85/128
+[decode] max_len=128 generated=86/128
+[decode] max_len=128 generated=87/128
+[decode] max_len=128 generated=88/128
+[decode] max_len=128 generated=89/128
+[decode] max_len=128 generated=90/128
+[decode] max_len=128 generated=91/128
+[decode] max_len=128 generated=92/128
+[decode] max_len=128 generated=93/128
+[decode] max_len=128 generated=94/128
+[decode] max_len=128 generated=95/128
+[decode] max_len=128 generated=96/128
+[decode] max_len=128 generated=97/128
+[decode] max_len=128 generated=98/128
+[decode] max_len=128 generated=99/128
+[decode] max_len=128 generated=100/128
+[decode] max_len=128 generated=101/128
+[decode] max_len=128 generated=102/128
+[decode] max_len=128 generated=103/128
+[decode] max_len=128 generated=104/128
+[decode] max_len=128 generated=105/128
+[decode] max_len=128 generated=106/128
+[decode] max_len=128 generated=107/128
+[decode] max_len=128 generated=108/128
+[decode] max_len=128 generated=109/128
+[decode] max_len=128 generated=110/128
+[decode] max_len=128 generated=111/128
+[decode] max_len=128 generated=112/128
+[decode] max_len=128 generated=113/128
+[decode] max_len=128 generated=114/128
+[decode] max_len=128 generated=115/128
+[decode] max_len=128 generated=116/128
+[decode] max_len=128 generated=117/128
+[decode] max_len=128 generated=118/128
+[decode] max_len=128 generated=119/128
+[decode] max_len=128 generated=120/128
+[decode] max_len=128 generated=121/128
+[decode] max_len=128 generated=122/128
+[decode] max_len=128 generated=123/128
+[decode] max_len=128 generated=124/128
+[decode] max_len=128 generated=125/128
+[decode] max_len=128 generated=126/128
+[decode] max_len=128 generated=127/128
+[decode] max_len=128 generated=128/128
+[
+  {
+    "checkpoint": "runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0036000.pt",
+    "ckpt_step": 36000,
+    "max_len": 128,
+    "decode_rule": "dual_line_resample",
+    "support_power": 1.0,
+    "semantic_power": 1.5,
+    "steps": 128,
+    "c_min": 1.0,
+    "c_max": 1024.0,
+    "anchor_mode": "state",
+    "model_t_mode": "flow",
+    "time_schedule": "uniform",
+    "time_logit_mean": -1.5,
+    "time_logit_std": 0.8,
+    "time_power": 2.0,
+    "input_noise_scale": 0.0,
+    "input_noise_until": 1.0,
+    "input_noise_dirichlet_concentration": 1.0,
+    "endpoint_softening": "none",
+    "endpoint_soft_power": 2.0,
+    "endpoint_soft_min_conf": 0.0,
+    "endpoint_soft_max_conf": 1.0,
+    "soft_target_decode_mode": "off",
+    "soft_target_power": 1.0,
+    "soft_target_min_conf": 0.0,
+    "soft_target_max_conf": 1.0,
+    "soft_target_debias_start": 0.7,
+    "final_from": "blend",
+    "final_decode": "argmax",
+    "final_sample_temp": 1.0,
+    "final_top_k": 0,
+    "final_top_p": 1.0,
+    "commit_mode": "off",
+    "commit_conf_threshold": 0.0,
+    "commit_margin_threshold": 0.0,
+    "commit_start": 0.0,
+    "commit_min_ratio": 0.0,
+    "commit_max_ratio": 1.0,
+    "commit_power": 2.0,
+    "commit_freq_max_frac": 0.08,
+    "early_temp": 2.8,
+    "late_temp": 1.45,
+    "temp_end": 0.55,
+    "temp_power": 1.5,
+    "pos_extend": "repeat",
+    "fixed_first_token_id": null,
+    "fixed_first_token_text": "",
+    "fixed_first_initial_argmax": false,
+    "use_ema": false,
+    "n_samples": 128,
+    "sample_entropy": 2.9528095147982674,
+    "unique_tokens": 652,
+    "token_count": 16384,
+    "distinct_1": 0.039794921875,
+    "distinct_2": 0.2578125,
+    "top_token_mass": 0.091064453125,
+    "texts_preview": [
+      "[CLS] kokuja koko sekek kaku kos kakukukaku kodikz kork koka kokav kukuan koku kokuko sukuka koko kokkoku koku kakukokuo ko koku kov uvjak kokuz kav o kokokuka ko kokua, kodiz, koliu kodiv kokujak uvjak kov kokukukuko koku kor kaz kovk, kos kozkahod ko kuzkk koja [SEP]",
+      "[CLS]kaku kagokajakao kakuko zaji kaku kokua kakuja na kojaja. kokuzko kakukvku koi kikuk. nar kajazjeja kakukukau zai zana kavkokoo kojaka kakuka kaku kavjii sekul kojavkaku kokuvkujakau. kikuv nikumilje kakukojakaka zazkuvjecjadjakukuvjejazad ka kokni uz [SEP]",
+      "[CLS]koko pezti o konici taka o koni ojaci. oka. o kor na kojak kajad kokaan kajaci pekojna, ped kojanev kogo kajazci kakozoa ovva? kozinijaz seu o mu koljna na. hijazoo na se koja najazvici ka kolov ko koz ovula koja koci. o kone in 2016 in a new republic. je mova kadati suja koz ko [SEP]",
+      "[CLS]ос нетеете стето коле корелиссто воре селономет пе нако немоетети н кодал. котрее коле претало в нереле ко колиелело контотодо сересралаиее колаль п [SEP]"
+    ],
+    "gen_ppl": 79.85979711573923,
+    "gen_nll": 4.380272561132715,
+    "gen_tokens": 16116
+  }
+]
+[watch-dualline] 2026-05-26_13:26:34 done step_0036000

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/arrayprint.pyi

@@ -0,0 +1,30 @@
+import sys
+import contextlib
+from collections.abc import Callable
+from typing import Any
+
+import numpy as np
+from numpy.core.arrayprint import _FormatOptions
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+AR: np.ndarray[Any, Any]
+func_float: Callable[[np.floating[Any]], str]
+func_int: Callable[[np.integer[Any]], str]
+
+assert_type(np.get_printoptions(), _FormatOptions)
+assert_type(
+    np.array2string(AR, formatter={'float_kind': func_float, 'int_kind': func_int}),
+    str,
+)
+assert_type(np.format_float_scientific(1.0), str)
+assert_type(np.format_float_positional(1), str)
+assert_type(np.array_repr(AR), str)
+assert_type(np.array_str(AR), str)
+
+assert_type(np.printoptions(), contextlib._GeneratorContextManager[_FormatOptions])
+with np.printoptions() as dct:
+    assert_type(dct, _FormatOptions)

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/constants.pyi

@@ -0,0 +1,61 @@
+import sys
+from typing import Literal
+
+import numpy as np
+from numpy.core._type_aliases import _SCTypes
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+assert_type(np.Inf, float)
+assert_type(np.Infinity, float)
+assert_type(np.NAN, float)
+assert_type(np.NINF, float)
+assert_type(np.NZERO, float)
+assert_type(np.NaN, float)
+assert_type(np.PINF, float)
+assert_type(np.PZERO, float)
+assert_type(np.e, float)
+assert_type(np.euler_gamma, float)
+assert_type(np.inf, float)
+assert_type(np.infty, float)
+assert_type(np.nan, float)
+assert_type(np.pi, float)
+
+assert_type(np.ALLOW_THREADS, int)
+assert_type(np.BUFSIZE, Literal[8192])
+assert_type(np.CLIP, Literal[0])
+assert_type(np.ERR_CALL, Literal[3])
+assert_type(np.ERR_DEFAULT, Literal[521])
+assert_type(np.ERR_IGNORE, Literal[0])
+assert_type(np.ERR_LOG, Literal[5])
+assert_type(np.ERR_PRINT, Literal[4])
+assert_type(np.ERR_RAISE, Literal[2])
+assert_type(np.ERR_WARN, Literal[1])
+assert_type(np.FLOATING_POINT_SUPPORT, Literal[1])
+assert_type(np.FPE_DIVIDEBYZERO, Literal[1])
+assert_type(np.FPE_INVALID, Literal[8])
+assert_type(np.FPE_OVERFLOW, Literal[2])
+assert_type(np.FPE_UNDERFLOW, Literal[4])
+assert_type(np.MAXDIMS, Literal[32])
+assert_type(np.MAY_SHARE_BOUNDS, Literal[0])
+assert_type(np.MAY_SHARE_EXACT, Literal[-1])
+assert_type(np.RAISE, Literal[2])
+assert_type(np.SHIFT_DIVIDEBYZERO, Literal[0])
+assert_type(np.SHIFT_INVALID, Literal[9])
+assert_type(np.SHIFT_OVERFLOW, Literal[3])
+assert_type(np.SHIFT_UNDERFLOW, Literal[6])
+assert_type(np.UFUNC_BUFSIZE_DEFAULT, Literal[8192])
+assert_type(np.WRAP, Literal[1])
+assert_type(np.tracemalloc_domain, Literal[389047])
+
+assert_type(np.little_endian, bool)
+assert_type(np.True_, np.bool_)
+assert_type(np.False_, np.bool_)
+
+assert_type(np.UFUNC_PYVALS_NAME, Literal["UFUNC_PYVALS"])
+
+assert_type(np.sctypeDict, dict[int | str, type[np.generic]])
+assert_type(np.sctypes, _SCTypes)

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/index_tricks.pyi

@@ -0,0 +1,74 @@
+import sys
+from typing import Any, Literal
+
+import numpy as np
+import numpy.typing as npt
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+AR_LIKE_b: list[bool]
+AR_LIKE_i: list[int]
+AR_LIKE_f: list[float]
+AR_LIKE_U: list[str]
+
+AR_i8: np.ndarray[Any, np.dtype[np.int64]]
+
+assert_type(np.ndenumerate(AR_i8), np.ndenumerate[np.int64])
+assert_type(np.ndenumerate(AR_LIKE_f), np.ndenumerate[np.float64])
+assert_type(np.ndenumerate(AR_LIKE_U), np.ndenumerate[np.str_])
+
+assert_type(np.ndenumerate(AR_i8).iter, np.flatiter[npt.NDArray[np.int64]])
+assert_type(np.ndenumerate(AR_LIKE_f).iter, np.flatiter[npt.NDArray[np.float64]])
+assert_type(np.ndenumerate(AR_LIKE_U).iter, np.flatiter[npt.NDArray[np.str_]])
+
+assert_type(next(np.ndenumerate(AR_i8)), tuple[tuple[int, ...], np.int64])
+assert_type(next(np.ndenumerate(AR_LIKE_f)), tuple[tuple[int, ...], np.float64])
+assert_type(next(np.ndenumerate(AR_LIKE_U)), tuple[tuple[int, ...], np.str_])
+
+assert_type(iter(np.ndenumerate(AR_i8)), np.ndenumerate[np.int64])
+assert_type(iter(np.ndenumerate(AR_LIKE_f)), np.ndenumerate[np.float64])
+assert_type(iter(np.ndenumerate(AR_LIKE_U)), np.ndenumerate[np.str_])
+
+assert_type(np.ndindex(1, 2, 3), np.ndindex)
+assert_type(np.ndindex((1, 2, 3)), np.ndindex)
+assert_type(iter(np.ndindex(1, 2, 3)), np.ndindex)
+assert_type(next(np.ndindex(1, 2, 3)), tuple[int, ...])
+
+assert_type(np.unravel_index([22, 41, 37], (7, 6)), tuple[npt.NDArray[np.intp], ...])
+assert_type(np.unravel_index([31, 41, 13], (7, 6), order="F"), tuple[npt.NDArray[np.intp], ...])
+assert_type(np.unravel_index(1621, (6, 7, 8, 9)), tuple[np.intp, ...])
+
+assert_type(np.ravel_multi_index([[1]], (7, 6)), npt.NDArray[np.intp])
+assert_type(np.ravel_multi_index(AR_LIKE_i, (7, 6)), np.intp)
+assert_type(np.ravel_multi_index(AR_LIKE_i, (7, 6), order="F"), np.intp)
+assert_type(np.ravel_multi_index(AR_LIKE_i, (4, 6), mode="clip"), np.intp)
+assert_type(np.ravel_multi_index(AR_LIKE_i, (4, 4), mode=("clip", "wrap")), np.intp)
+assert_type(np.ravel_multi_index((3, 1, 4, 1), (6, 7, 8, 9)), np.intp)
+
+assert_type(np.mgrid[1:1:2], npt.NDArray[Any])
+assert_type(np.mgrid[1:1:2, None:10], npt.NDArray[Any])
+
+assert_type(np.ogrid[1:1:2], list[npt.NDArray[Any]])
+assert_type(np.ogrid[1:1:2, None:10], list[npt.NDArray[Any]])
+
+assert_type(np.index_exp[0:1], tuple[slice])
+assert_type(np.index_exp[0:1, None:3], tuple[slice, slice])
+assert_type(np.index_exp[0, 0:1, ..., [0, 1, 3]], tuple[Literal[0], slice, ellipsis, list[int]])
+
+assert_type(np.s_[0:1], slice)
+assert_type(np.s_[0:1, None:3], tuple[slice, slice])
+assert_type(np.s_[0, 0:1, ..., [0, 1, 3]], tuple[Literal[0], slice, ellipsis, list[int]])
+
+assert_type(np.ix_(AR_LIKE_b), tuple[npt.NDArray[np.bool_], ...])
+assert_type(np.ix_(AR_LIKE_i, AR_LIKE_f), tuple[npt.NDArray[np.float64], ...])
+assert_type(np.ix_(AR_i8), tuple[npt.NDArray[np.int64], ...])
+
+assert_type(np.fill_diagonal(AR_i8, 5), None)
+
+assert_type(np.diag_indices(4), tuple[npt.NDArray[np.int_], ...])
+assert_type(np.diag_indices(2, 3), tuple[npt.NDArray[np.int_], ...])
+
+assert_type(np.diag_indices_from(AR_i8), tuple[npt.NDArray[np.int_], ...])

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/lib_version.pyi

@@ -0,0 +1,25 @@
+import sys
+
+from numpy.lib import NumpyVersion
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+version = NumpyVersion("1.8.0")
+
+assert_type(version.vstring, str)
+assert_type(version.version, str)
+assert_type(version.major, int)
+assert_type(version.minor, int)
+assert_type(version.bugfix, int)
+assert_type(version.pre_release, str)
+assert_type(version.is_devversion, bool)
+
+assert_type(version == version, bool)
+assert_type(version != version, bool)
+assert_type(version < "1.8.0", bool)
+assert_type(version <= version, bool)
+assert_type(version > version, bool)
+assert_type(version >= "1.8.0", bool)

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/ndarray_conversion.pyi

@@ -0,0 +1,59 @@
+import sys
+from typing import Any
+
+import numpy as np
+import numpy.typing as npt
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+nd: npt.NDArray[np.int_]
+
+# item
+assert_type(nd.item(), int)
+assert_type(nd.item(1), int)
+assert_type(nd.item(0, 1), int)
+assert_type(nd.item((0, 1)), int)
+
+# tolist
+assert_type(nd.tolist(), Any)
+
+# itemset does not return a value
+# tostring is pretty simple
+# tobytes is pretty simple
+# tofile does not return a value
+# dump does not return a value
+# dumps is pretty simple
+
+# astype
+assert_type(nd.astype("float"), npt.NDArray[Any])
+assert_type(nd.astype(float), npt.NDArray[Any])
+assert_type(nd.astype(np.float64), npt.NDArray[np.float64])
+assert_type(nd.astype(np.float64, "K"), npt.NDArray[np.float64])
+assert_type(nd.astype(np.float64, "K", "unsafe"), npt.NDArray[np.float64])
+assert_type(nd.astype(np.float64, "K", "unsafe", True), npt.NDArray[np.float64])
+assert_type(nd.astype(np.float64, "K", "unsafe", True, True), npt.NDArray[np.float64])
+
+# byteswap
+assert_type(nd.byteswap(), npt.NDArray[np.int_])
+assert_type(nd.byteswap(True), npt.NDArray[np.int_])
+
+# copy
+assert_type(nd.copy(), npt.NDArray[np.int_])
+assert_type(nd.copy("C"), npt.NDArray[np.int_])
+
+assert_type(nd.view(), npt.NDArray[np.int_])
+assert_type(nd.view(np.float64), npt.NDArray[np.float64])
+assert_type(nd.view(float), npt.NDArray[Any])
+assert_type(nd.view(np.float64, np.matrix), np.matrix[Any, Any])
+
+# getfield
+assert_type(nd.getfield("float"), npt.NDArray[Any])
+assert_type(nd.getfield(float), npt.NDArray[Any])
+assert_type(nd.getfield(np.float64), npt.NDArray[np.float64])
+assert_type(nd.getfield(np.float64, 8), npt.NDArray[np.float64])
+
+# setflags does not return a value
+# fill does not return a value

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/numeric.pyi

@@ -0,0 +1,141 @@
+"""
+Tests for :mod:`core.numeric`.
+
+Does not include tests which fall under ``array_constructors``.
+
+"""
+
+import sys
+from typing import Any
+
+import numpy as np
+import numpy.typing as npt
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+class SubClass(npt.NDArray[np.int64]):
+    ...
+
+i8: np.int64
+
+AR_b: npt.NDArray[np.bool_]
+AR_u8: npt.NDArray[np.uint64]
+AR_i8: npt.NDArray[np.int64]
+AR_f8: npt.NDArray[np.float64]
+AR_c16: npt.NDArray[np.complex128]
+AR_m: npt.NDArray[np.timedelta64]
+AR_O: npt.NDArray[np.object_]
+
+B: list[int]
+C: SubClass
+
+assert_type(np.count_nonzero(i8), int)
+assert_type(np.count_nonzero(AR_i8), int)
+assert_type(np.count_nonzero(B), int)
+assert_type(np.count_nonzero(AR_i8, keepdims=True), Any)
+assert_type(np.count_nonzero(AR_i8, axis=0), Any)
+
+assert_type(np.isfortran(i8), bool)
+assert_type(np.isfortran(AR_i8), bool)
+
+assert_type(np.argwhere(i8), npt.NDArray[np.intp])
+assert_type(np.argwhere(AR_i8), npt.NDArray[np.intp])
+
+assert_type(np.flatnonzero(i8), npt.NDArray[np.intp])
+assert_type(np.flatnonzero(AR_i8), npt.NDArray[np.intp])
+
+assert_type(np.correlate(B, AR_i8, mode="valid"), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.correlate(AR_i8, AR_i8, mode="same"), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.correlate(AR_b, AR_b), npt.NDArray[np.bool_])
+assert_type(np.correlate(AR_b, AR_u8), npt.NDArray[np.unsignedinteger[Any]])
+assert_type(np.correlate(AR_i8, AR_b), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.correlate(AR_i8, AR_f8), npt.NDArray[np.floating[Any]])
+assert_type(np.correlate(AR_i8, AR_c16), npt.NDArray[np.complexfloating[Any, Any]])
+assert_type(np.correlate(AR_i8, AR_m), npt.NDArray[np.timedelta64])
+assert_type(np.correlate(AR_O, AR_O), npt.NDArray[np.object_])
+
+assert_type(np.convolve(B, AR_i8, mode="valid"), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.convolve(AR_i8, AR_i8, mode="same"), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.convolve(AR_b, AR_b), npt.NDArray[np.bool_])
+assert_type(np.convolve(AR_b, AR_u8), npt.NDArray[np.unsignedinteger[Any]])
+assert_type(np.convolve(AR_i8, AR_b), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.convolve(AR_i8, AR_f8), npt.NDArray[np.floating[Any]])
+assert_type(np.convolve(AR_i8, AR_c16), npt.NDArray[np.complexfloating[Any, Any]])
+assert_type(np.convolve(AR_i8, AR_m), npt.NDArray[np.timedelta64])
+assert_type(np.convolve(AR_O, AR_O), npt.NDArray[np.object_])
+
+assert_type(np.outer(i8, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.outer(B, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.outer(AR_i8, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.outer(AR_i8, AR_i8, out=C), SubClass)
+assert_type(np.outer(AR_b, AR_b), npt.NDArray[np.bool_])
+assert_type(np.outer(AR_b, AR_u8), npt.NDArray[np.unsignedinteger[Any]])
+assert_type(np.outer(AR_i8, AR_b), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.convolve(AR_i8, AR_f8), npt.NDArray[np.floating[Any]])
+assert_type(np.outer(AR_i8, AR_c16), npt.NDArray[np.complexfloating[Any, Any]])
+assert_type(np.outer(AR_i8, AR_m), npt.NDArray[np.timedelta64])
+assert_type(np.outer(AR_O, AR_O), npt.NDArray[np.object_])
+
+assert_type(np.tensordot(B, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.tensordot(AR_i8, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.tensordot(AR_i8, AR_i8, axes=0), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.tensordot(AR_i8, AR_i8, axes=(0, 1)), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.tensordot(AR_b, AR_b), npt.NDArray[np.bool_])
+assert_type(np.tensordot(AR_b, AR_u8), npt.NDArray[np.unsignedinteger[Any]])
+assert_type(np.tensordot(AR_i8, AR_b), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.tensordot(AR_i8, AR_f8), npt.NDArray[np.floating[Any]])
+assert_type(np.tensordot(AR_i8, AR_c16), npt.NDArray[np.complexfloating[Any, Any]])
+assert_type(np.tensordot(AR_i8, AR_m), npt.NDArray[np.timedelta64])
+assert_type(np.tensordot(AR_O, AR_O), npt.NDArray[np.object_])
+
+assert_type(np.isscalar(i8), bool)
+assert_type(np.isscalar(AR_i8), bool)
+assert_type(np.isscalar(B), bool)
+
+assert_type(np.roll(AR_i8, 1), npt.NDArray[np.int64])
+assert_type(np.roll(AR_i8, (1, 2)), npt.NDArray[np.int64])
+assert_type(np.roll(B, 1), npt.NDArray[Any])
+
+assert_type(np.rollaxis(AR_i8, 0, 1), npt.NDArray[np.int64])
+
+assert_type(np.moveaxis(AR_i8, 0, 1), npt.NDArray[np.int64])
+assert_type(np.moveaxis(AR_i8, (0, 1), (1, 2)), npt.NDArray[np.int64])
+
+assert_type(np.cross(B, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.cross(AR_i8, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.cross(AR_b, AR_u8), npt.NDArray[np.unsignedinteger[Any]])
+assert_type(np.cross(AR_i8, AR_b), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.cross(AR_i8, AR_f8), npt.NDArray[np.floating[Any]])
+assert_type(np.cross(AR_i8, AR_c16), npt.NDArray[np.complexfloating[Any, Any]])
+assert_type(np.cross(AR_O, AR_O), npt.NDArray[np.object_])
+
+assert_type(np.indices([0, 1, 2]), npt.NDArray[np.int_])
+assert_type(np.indices([0, 1, 2], sparse=True), tuple[npt.NDArray[np.int_], ...])
+assert_type(np.indices([0, 1, 2], dtype=np.float64), npt.NDArray[np.float64])
+assert_type(np.indices([0, 1, 2], sparse=True, dtype=np.float64), tuple[npt.NDArray[np.float64], ...])
+assert_type(np.indices([0, 1, 2], dtype=float), npt.NDArray[Any])
+assert_type(np.indices([0, 1, 2], sparse=True, dtype=float), tuple[npt.NDArray[Any], ...])
+
+assert_type(np.binary_repr(1), str)
+
+assert_type(np.base_repr(1), str)
+
+assert_type(np.allclose(i8, AR_i8), bool)
+assert_type(np.allclose(B, AR_i8), bool)
+assert_type(np.allclose(AR_i8, AR_i8), bool)
+
+assert_type(np.isclose(i8, i8), np.bool_)
+assert_type(np.isclose(i8, AR_i8), npt.NDArray[np.bool_])
+assert_type(np.isclose(B, AR_i8), npt.NDArray[np.bool_])
+assert_type(np.isclose(AR_i8, AR_i8), npt.NDArray[np.bool_])
+
+assert_type(np.array_equal(i8, AR_i8), bool)
+assert_type(np.array_equal(B, AR_i8), bool)
+assert_type(np.array_equal(AR_i8, AR_i8), bool)
+
+assert_type(np.array_equiv(i8, AR_i8), bool)
+assert_type(np.array_equiv(B, AR_i8), bool)
+assert_type(np.array_equiv(AR_i8, AR_i8), bool)

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/rec.pyi

@@ -0,0 +1,167 @@
+import io
+import sys
+from typing import Any
+
+import numpy as np
+import numpy.typing as npt
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+AR_i8: npt.NDArray[np.int64]
+REC_AR_V: np.recarray[Any, np.dtype[np.record]]
+AR_LIST: list[npt.NDArray[np.int64]]
+
+format_parser: np.format_parser
+record: np.record
+file_obj: io.BufferedIOBase
+
+assert_type(np.format_parser(
+    formats=[np.float64, np.int64, np.bool_],
+    names=["f8", "i8", "?"],
+    titles=None,
+    aligned=True,
+), np.format_parser)
+assert_type(format_parser.dtype, np.dtype[np.void])
+
+assert_type(record.field_a, Any)
+assert_type(record.field_b, Any)
+assert_type(record["field_a"], Any)
+assert_type(record["field_b"], Any)
+assert_type(record.pprint(), str)
+record.field_c = 5
+
+assert_type(REC_AR_V.field(0), Any)
+assert_type(REC_AR_V.field("field_a"), Any)
+assert_type(REC_AR_V.field(0, AR_i8), None)
+assert_type(REC_AR_V.field("field_a", AR_i8), None)
+assert_type(REC_AR_V["field_a"], npt.NDArray[Any])
+assert_type(REC_AR_V.field_a, Any)
+assert_type(REC_AR_V.__array_finalize__(object()), None)
+
+assert_type(
+    np.recarray(
+        shape=(10, 5),
+        formats=[np.float64, np.int64, np.bool_],
+        order="K",
+        byteorder="|",
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+
+assert_type(
+    np.recarray(
+        shape=(10, 5),
+        dtype=[("f8", np.float64), ("i8", np.int64)],
+        strides=(5, 5),
+    ),
+    np.recarray[Any, np.dtype[Any]],
+)
+
+assert_type(np.rec.fromarrays(AR_LIST), np.recarray[Any, np.dtype[Any]])
+assert_type(
+    np.rec.fromarrays(AR_LIST, dtype=np.int64),
+    np.recarray[Any, np.dtype[Any]],
+)
+assert_type(
+    np.rec.fromarrays(
+        AR_LIST,
+        formats=[np.int64, np.float64],
+        names=["i8", "f8"]
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+
+assert_type(np.rec.fromrecords((1, 1.5)), np.recarray[Any, np.dtype[np.record]])
+assert_type(
+    np.rec.fromrecords(
+        [(1, 1.5)],
+        dtype=[("i8", np.int64), ("f8", np.float64)],
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+assert_type(
+    np.rec.fromrecords(
+        REC_AR_V,
+        formats=[np.int64, np.float64],
+        names=["i8", "f8"]
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+
+assert_type(
+    np.rec.fromstring(
+        b"(1, 1.5)",
+        dtype=[("i8", np.int64), ("f8", np.float64)],
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+assert_type(
+    np.rec.fromstring(
+        REC_AR_V,
+        formats=[np.int64, np.float64],
+        names=["i8", "f8"]
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+
+assert_type(np.rec.fromfile(
+    "test_file.txt",
+    dtype=[("i8", np.int64), ("f8", np.float64)],
+), np.recarray[Any, np.dtype[Any]])
+
+assert_type(
+    np.rec.fromfile(
+        file_obj,
+        formats=[np.int64, np.float64],
+        names=["i8", "f8"]
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+
+assert_type(np.rec.array(AR_i8), np.recarray[Any, np.dtype[np.int64]])
+
+assert_type(
+    np.rec.array([(1, 1.5)], dtype=[("i8", np.int64), ("f8", np.float64)]),
+    np.recarray[Any, np.dtype[Any]],
+)
+
+assert_type(
+    np.rec.array(
+        [(1, 1.5)],
+        formats=[np.int64, np.float64],
+        names=["i8", "f8"]
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+
+assert_type(
+    np.rec.array(
+        None,
+        dtype=np.float64,
+        shape=(10, 3),
+    ),
+    np.recarray[Any, np.dtype[Any]],
+)
+
+assert_type(
+    np.rec.array(
+        None,
+        formats=[np.int64, np.float64],
+        names=["i8", "f8"],
+        shape=(10, 3),
+    ),
+    np.recarray[Any, np.dtype[np.record]],
+)
+
+assert_type(
+    np.rec.array(file_obj, dtype=np.float64),
+    np.recarray[Any, np.dtype[Any]],
+)
+
+assert_type(
+    np.rec.array(file_obj, formats=[np.int64, np.float64], names=["i8", "f8"]),
+    np.recarray[Any, np.dtype[np.record]],
+)

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/shape_base.pyi

@@ -0,0 +1,65 @@
+import sys
+from typing import Any
+
+import numpy as np
+import numpy.typing as npt
+from numpy.lib.shape_base import _ArrayPrepare, _ArrayWrap
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+i8: np.int64
+f8: np.float64
+
+AR_b: npt.NDArray[np.bool_]
+AR_i8: npt.NDArray[np.int64]
+AR_f8: npt.NDArray[np.float64]
+
+AR_LIKE_f8: list[float]
+
+assert_type(np.take_along_axis(AR_f8, AR_i8, axis=1), npt.NDArray[np.float64])
+assert_type(np.take_along_axis(f8, AR_i8, axis=None), npt.NDArray[np.float64])
+
+assert_type(np.put_along_axis(AR_f8, AR_i8, "1.0", axis=1), None)
+
+assert_type(np.expand_dims(AR_i8, 2), npt.NDArray[np.int64])
+assert_type(np.expand_dims(AR_LIKE_f8, 2), npt.NDArray[Any])
+
+assert_type(np.column_stack([AR_i8]), npt.NDArray[np.int64])
+assert_type(np.column_stack([AR_LIKE_f8]), npt.NDArray[Any])
+
+assert_type(np.dstack([AR_i8]), npt.NDArray[np.int64])
+assert_type(np.dstack([AR_LIKE_f8]), npt.NDArray[Any])
+
+assert_type(np.row_stack([AR_i8]), npt.NDArray[np.int64])
+assert_type(np.row_stack([AR_LIKE_f8]), npt.NDArray[Any])
+
+assert_type(np.array_split(AR_i8, [3, 5, 6, 10]), list[npt.NDArray[np.int64]])
+assert_type(np.array_split(AR_LIKE_f8, [3, 5, 6, 10]), list[npt.NDArray[Any]])
+
+assert_type(np.split(AR_i8, [3, 5, 6, 10]), list[npt.NDArray[np.int64]])
+assert_type(np.split(AR_LIKE_f8, [3, 5, 6, 10]), list[npt.NDArray[Any]])
+
+assert_type(np.hsplit(AR_i8, [3, 5, 6, 10]), list[npt.NDArray[np.int64]])
+assert_type(np.hsplit(AR_LIKE_f8, [3, 5, 6, 10]), list[npt.NDArray[Any]])
+
+assert_type(np.vsplit(AR_i8, [3, 5, 6, 10]), list[npt.NDArray[np.int64]])
+assert_type(np.vsplit(AR_LIKE_f8, [3, 5, 6, 10]), list[npt.NDArray[Any]])
+
+assert_type(np.dsplit(AR_i8, [3, 5, 6, 10]), list[npt.NDArray[np.int64]])
+assert_type(np.dsplit(AR_LIKE_f8, [3, 5, 6, 10]), list[npt.NDArray[Any]])
+
+assert_type(np.lib.shape_base.get_array_prepare(AR_i8), _ArrayPrepare)
+assert_type(np.lib.shape_base.get_array_prepare(AR_i8, 1), None | _ArrayPrepare)
+
+assert_type(np.get_array_wrap(AR_i8), _ArrayWrap)
+assert_type(np.get_array_wrap(AR_i8, 1), None | _ArrayWrap)
+
+assert_type(np.kron(AR_b, AR_b), npt.NDArray[np.bool_])
+assert_type(np.kron(AR_b, AR_i8), npt.NDArray[np.signedinteger[Any]])
+assert_type(np.kron(AR_f8, AR_f8), npt.NDArray[np.floating[Any]])
+
+assert_type(np.tile(AR_i8, 5), npt.NDArray[np.int64])
+assert_type(np.tile(AR_LIKE_f8, [2, 2]), npt.NDArray[Any])

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/testing.pyi

@@ -0,0 +1,203 @@
+import re
+import sys
+import warnings
+import types
+import unittest
+import contextlib
+from collections.abc import Callable
+from typing import Any, TypeVar
+from pathlib import Path
+
+import numpy as np
+import numpy.typing as npt
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+AR_f8: npt.NDArray[np.float64]
+AR_i8: npt.NDArray[np.int64]
+
+bool_obj: bool
+suppress_obj: np.testing.suppress_warnings
+FT = TypeVar("FT", bound=Callable[..., Any])
+
+def func() -> int: ...
+
+def func2(
+    x: npt.NDArray[np.number[Any]],
+    y: npt.NDArray[np.number[Any]],
+) -> npt.NDArray[np.bool_]: ...
+
+assert_type(np.testing.KnownFailureException(), np.testing.KnownFailureException)
+assert_type(np.testing.IgnoreException(), np.testing.IgnoreException)
+
+assert_type(
+    np.testing.clear_and_catch_warnings(modules=[np.testing]),
+    np.testing._private.utils._clear_and_catch_warnings_without_records,
+)
+assert_type(
+    np.testing.clear_and_catch_warnings(True),
+    np.testing._private.utils._clear_and_catch_warnings_with_records,
+)
+assert_type(
+    np.testing.clear_and_catch_warnings(False),
+    np.testing._private.utils._clear_and_catch_warnings_without_records,
+)
+assert_type(
+    np.testing.clear_and_catch_warnings(bool_obj),
+    np.testing.clear_and_catch_warnings,
+)
+assert_type(
+    np.testing.clear_and_catch_warnings.class_modules,
+    tuple[types.ModuleType, ...],
+)
+assert_type(
+    np.testing.clear_and_catch_warnings.modules,
+    set[types.ModuleType],
+)
+
+with np.testing.clear_and_catch_warnings(True) as c1:
+    assert_type(c1, list[warnings.WarningMessage])
+with np.testing.clear_and_catch_warnings() as c2:
+    assert_type(c2, None)
+
+assert_type(np.testing.suppress_warnings("once"), np.testing.suppress_warnings)
+assert_type(np.testing.suppress_warnings()(func), Callable[[], int])
+assert_type(suppress_obj.filter(RuntimeWarning), None)
+assert_type(suppress_obj.record(RuntimeWarning), list[warnings.WarningMessage])
+with suppress_obj as c3:
+    assert_type(c3, np.testing.suppress_warnings)
+
+assert_type(np.testing.verbose, int)
+assert_type(np.testing.IS_PYPY, bool)
+assert_type(np.testing.HAS_REFCOUNT, bool)
+assert_type(np.testing.HAS_LAPACK64, bool)
+
+assert_type(np.testing.assert_(1, msg="test"), None)
+assert_type(np.testing.assert_(2, msg=lambda: "test"), None)
+
+if sys.platform == "win32" or sys.platform == "cygwin":
+    assert_type(np.testing.memusage(), int)
+elif sys.platform == "linux":
+    assert_type(np.testing.memusage(), None | int)
+
+assert_type(np.testing.jiffies(), int)
+
+assert_type(np.testing.build_err_msg([0, 1, 2], "test"), str)
+assert_type(np.testing.build_err_msg(range(2), "test", header="header"), str)
+assert_type(np.testing.build_err_msg(np.arange(9).reshape(3, 3), "test", verbose=False), str)
+assert_type(np.testing.build_err_msg("abc", "test", names=["x", "y"]), str)
+assert_type(np.testing.build_err_msg([1.0, 2.0], "test", precision=5), str)
+
+assert_type(np.testing.assert_equal({1}, {1}), None)
+assert_type(np.testing.assert_equal([1, 2, 3], [1, 2, 3], err_msg="fail"), None)
+assert_type(np.testing.assert_equal(1, 1.0, verbose=True), None)
+
+assert_type(np.testing.print_assert_equal('Test XYZ of func xyz', [0, 1], [0, 1]), None)
+
+assert_type(np.testing.assert_almost_equal(1.0, 1.1), None)
+assert_type(np.testing.assert_almost_equal([1, 2, 3], [1, 2, 3], err_msg="fail"), None)
+assert_type(np.testing.assert_almost_equal(1, 1.0, verbose=True), None)
+assert_type(np.testing.assert_almost_equal(1, 1.0001, decimal=2), None)
+
+assert_type(np.testing.assert_approx_equal(1.0, 1.1), None)
+assert_type(np.testing.assert_approx_equal("1", "2", err_msg="fail"), None)
+assert_type(np.testing.assert_approx_equal(1, 1.0, verbose=True), None)
+assert_type(np.testing.assert_approx_equal(1, 1.0001, significant=2), None)
+
+assert_type(np.testing.assert_array_compare(func2, AR_i8, AR_f8, err_msg="test"), None)
+assert_type(np.testing.assert_array_compare(func2, AR_i8, AR_f8, verbose=True), None)
+assert_type(np.testing.assert_array_compare(func2, AR_i8, AR_f8, header="header"), None)
+assert_type(np.testing.assert_array_compare(func2, AR_i8, AR_f8, precision=np.int64()), None)
+assert_type(np.testing.assert_array_compare(func2, AR_i8, AR_f8, equal_nan=False), None)
+assert_type(np.testing.assert_array_compare(func2, AR_i8, AR_f8, equal_inf=True), None)
+
+assert_type(np.testing.assert_array_equal(AR_i8, AR_f8), None)
+assert_type(np.testing.assert_array_equal(AR_i8, AR_f8, err_msg="test"), None)
+assert_type(np.testing.assert_array_equal(AR_i8, AR_f8, verbose=True), None)
+
+assert_type(np.testing.assert_array_almost_equal(AR_i8, AR_f8), None)
+assert_type(np.testing.assert_array_almost_equal(AR_i8, AR_f8, err_msg="test"), None)
+assert_type(np.testing.assert_array_almost_equal(AR_i8, AR_f8, verbose=True), None)
+assert_type(np.testing.assert_array_almost_equal(AR_i8, AR_f8, decimal=1), None)
+
+assert_type(np.testing.assert_array_less(AR_i8, AR_f8), None)
+assert_type(np.testing.assert_array_less(AR_i8, AR_f8, err_msg="test"), None)
+assert_type(np.testing.assert_array_less(AR_i8, AR_f8, verbose=True), None)
+
+assert_type(np.testing.runstring("1 + 1", {}), Any)
+assert_type(np.testing.runstring("int64() + 1", {"int64": np.int64}), Any)
+
+assert_type(np.testing.assert_string_equal("1", "1"), None)
+
+assert_type(np.testing.rundocs(), None)
+assert_type(np.testing.rundocs("test.py"), None)
+assert_type(np.testing.rundocs(Path("test.py"), raise_on_error=True), None)
+
+def func3(a: int) -> bool: ...
+
+assert_type(
+    np.testing.assert_raises(RuntimeWarning),
+    unittest.case._AssertRaisesContext[RuntimeWarning],
+)
+assert_type(np.testing.assert_raises(RuntimeWarning, func3, 5), None)
+
+assert_type(
+    np.testing.assert_raises_regex(RuntimeWarning, r"test"),
+    unittest.case._AssertRaisesContext[RuntimeWarning],
+)
+assert_type(np.testing.assert_raises_regex(RuntimeWarning, b"test", func3, 5), None)
+assert_type(np.testing.assert_raises_regex(RuntimeWarning, re.compile(b"test"), func3, 5), None)
+
+class Test: ...
+
+def decorate(a: FT) -> FT:
+    return a
+
+assert_type(np.testing.decorate_methods(Test, decorate), None)
+assert_type(np.testing.decorate_methods(Test, decorate, None), None)
+assert_type(np.testing.decorate_methods(Test, decorate, "test"), None)
+assert_type(np.testing.decorate_methods(Test, decorate, b"test"), None)
+assert_type(np.testing.decorate_methods(Test, decorate, re.compile("test")), None)
+
+assert_type(np.testing.measure("for i in range(1000): np.sqrt(i**2)"), float)
+assert_type(np.testing.measure(b"for i in range(1000): np.sqrt(i**2)", times=5), float)
+
+assert_type(np.testing.assert_allclose(AR_i8, AR_f8), None)
+assert_type(np.testing.assert_allclose(AR_i8, AR_f8, rtol=0.005), None)
+assert_type(np.testing.assert_allclose(AR_i8, AR_f8, atol=1), None)
+assert_type(np.testing.assert_allclose(AR_i8, AR_f8, equal_nan=True), None)
+assert_type(np.testing.assert_allclose(AR_i8, AR_f8, err_msg="err"), None)
+assert_type(np.testing.assert_allclose(AR_i8, AR_f8, verbose=False), None)
+
+assert_type(np.testing.assert_array_almost_equal_nulp(AR_i8, AR_f8, nulp=2), None)
+
+assert_type(np.testing.assert_array_max_ulp(AR_i8, AR_f8, maxulp=2), npt.NDArray[Any])
+assert_type(np.testing.assert_array_max_ulp(AR_i8, AR_f8, dtype=np.float32), npt.NDArray[Any])
+
+assert_type(np.testing.assert_warns(RuntimeWarning), contextlib._GeneratorContextManager[None])
+assert_type(np.testing.assert_warns(RuntimeWarning, func3, 5), bool)
+
+def func4(a: int, b: str) -> bool: ...
+
+assert_type(np.testing.assert_no_warnings(), contextlib._GeneratorContextManager[None])
+assert_type(np.testing.assert_no_warnings(func3, 5), bool)
+assert_type(np.testing.assert_no_warnings(func4, a=1, b="test"), bool)
+assert_type(np.testing.assert_no_warnings(func4, 1, "test"), bool)
+
+assert_type(np.testing.tempdir("test_dir"), contextlib._GeneratorContextManager[str])
+assert_type(np.testing.tempdir(prefix=b"test"), contextlib._GeneratorContextManager[bytes])
+assert_type(np.testing.tempdir("test_dir", dir=Path("here")), contextlib._GeneratorContextManager[str])
+
+assert_type(np.testing.temppath("test_dir", text=True), contextlib._GeneratorContextManager[str])
+assert_type(np.testing.temppath(prefix=b"test"), contextlib._GeneratorContextManager[bytes])
+assert_type(np.testing.temppath("test_dir", dir=Path("here")), contextlib._GeneratorContextManager[str])
+
+assert_type(np.testing.assert_no_gc_cycles(), contextlib._GeneratorContextManager[None])
+assert_type(np.testing.assert_no_gc_cycles(func3, 5), None)
+
+assert_type(np.testing.break_cycles(), None)
+
+assert_type(np.testing.TestCase(), unittest.case.TestCase)

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/typing/tests/data/reveal/ufunc_config.pyi

@@ -0,0 +1,41 @@
+"""Typing tests for `core._ufunc_config`."""
+
+import sys
+from typing import Any, Protocol
+from collections.abc import Callable
+
+import numpy as np
+from numpy.core._ufunc_config import _ErrDict
+
+if sys.version_info >= (3, 11):
+    from typing import assert_type
+else:
+    from typing_extensions import assert_type
+
+def func(a: str, b: int) -> None: ...
+
+class FuncProtocol(Protocol):
+    def __call__(self, a: str, b: int) -> None: ...
+
+class Write:
+    def write(self, value: str) -> None: ...
+
+class SupportsWrite(Protocol):
+    def write(self, s: str, /) -> object: ...
+
+assert_type(np.seterr(all=None), _ErrDict)
+assert_type(np.seterr(divide="ignore"), _ErrDict)
+assert_type(np.seterr(over="warn"), _ErrDict)
+assert_type(np.seterr(under="call"), _ErrDict)
+assert_type(np.seterr(invalid="raise"), _ErrDict)
+assert_type(np.geterr(), _ErrDict)
+
+assert_type(np.setbufsize(4096), int)
+assert_type(np.getbufsize(), int)
+
+assert_type(np.seterrcall(func), Callable[[str, int], Any] | None | SupportsWrite)
+assert_type(np.seterrcall(Write()), Callable[[str, int], Any] | None | SupportsWrite)
+assert_type(np.geterrcall(), Callable[[str, int], Any] | None | SupportsWrite)
+
+assert_type(np.errstate(call=func, all="call"), np.errstate[FuncProtocol])
+assert_type(np.errstate(call=Write(), divide="log", over="log"), np.errstate[Write])

LTA_openwebtext_dualt/mini_owt_logdirichlet/logs/infer_lr3e3_170k_logitnorm64_fine_20260612/gpu0.log

@@ -0,0 +1,372 @@
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.0 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:54:54+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64
+[2026-06-11T22:54:54+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:55:04+00:00] done decode=64 chunk=0
+[2026-06-11T22:55:04+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:55:14+00:00] done decode=64 chunk=1
+[2026-06-11T22:55:14+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:55:24+00:00] done decode=64 chunk=2
+[2026-06-11T22:55:24+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:55:33+00:00] done decode=64 chunk=3
+[2026-06-11T22:55:33+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:55:43+00:00] done decode=64 chunk=4
+[2026-06-11T22:55:43+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:55:53+00:00] done decode=64 chunk=5
+[2026-06-11T22:55:53+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:56:02+00:00] done decode=64 chunk=6
+[2026-06-11T22:56:02+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:56:12+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.98279539557957	5.106060441594896	0.08372632510721426	0.45314550837886514	0.03635364681104345	62	62	62840	65523	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.03800662604718	5.141491154944346	0.08642150374992122	0.4677706524650606	0.037530724144450744	0	0	58709	63468	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:56:44+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.0 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.05 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:56:44+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64
+[2026-06-11T22:56:44+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:56:54+00:00] done decode=64 chunk=0
+[2026-06-11T22:56:54+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:57:04+00:00] done decode=64 chunk=1
+[2026-06-11T22:57:04+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:57:13+00:00] done decode=64 chunk=2
+[2026-06-11T22:57:13+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:57:23+00:00] done decode=64 chunk=3
+[2026-06-11T22:57:23+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:57:32+00:00] done decode=64 chunk=4
+[2026-06-11T22:57:32+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:57:42+00:00] done decode=64 chunk=5
+[2026-06-11T22:57:42+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:57:52+00:00] done decode=64 chunk=6
+[2026-06-11T22:57:52+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:58:01+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.438972376537027	5.086244850431233	0.0840446487196323	0.4514567554361104	0.035700652017896135	63	63	63259	65489	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.015017413810938	5.130008505090127	0.08715276677963954	0.4681744033195546	0.03702765195907637	0	0	58543	63142	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:58:32+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.05 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.10 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:58:32+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64
+[2026-06-11T22:58:32+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:58:42+00:00] done decode=64 chunk=0
+[2026-06-11T22:58:42+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:58:51+00:00] done decode=64 chunk=1
+[2026-06-11T22:58:51+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:59:01+00:00] done decode=64 chunk=2
+[2026-06-11T22:59:01+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:59:10+00:00] done decode=64 chunk=3
+[2026-06-11T22:59:10+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:59:20+00:00] done decode=64 chunk=4
+[2026-06-11T22:59:20+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:59:29+00:00] done decode=64 chunk=5
+[2026-06-11T22:59:29+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:59:39+00:00] done decode=64 chunk=6
+[2026-06-11T22:59:39+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:59:49+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.87708717942494	5.095051071958517	0.083889287513549	0.4554670086409575	0.03509762911622368	61	61	63380	65503	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.402037538277753	5.138706414473064	0.0868960063266113	0.47183031760091104	0.03636219849742981	0	0	58803	63225	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:00:18+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.10 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.20 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:00:18+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:00:18+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:00:28+00:00] done decode=64 chunk=0
+[2026-06-11T23:00:28+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:00:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:00:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:00:47+00:00] done decode=64 chunk=2
+[2026-06-11T23:00:47+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:00:56+00:00] done decode=64 chunk=3
+[2026-06-11T23:00:56+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:01:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:01:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:01:16+00:00] done decode=64 chunk=5
+[2026-06-11T23:01:16+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:01:26+00:00] done decode=64 chunk=6
+[2026-06-11T23:01:26+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:01:36+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.89683419578054	5.106436489806833	0.08419220374696533	0.4590026262749649	0.037042126639488186	63	63	63616	65493	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.90647419071609	5.154605189833549	0.08750238080121897	0.47707252035617354	0.037997587454764774	0	0	58616	63004	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:02:05+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.55 s=2.20 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.0 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:02:05+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:02:05+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:02:15+00:00] done decode=64 chunk=0
+[2026-06-11T23:02:15+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:02:25+00:00] done decode=64 chunk=1
+[2026-06-11T23:02:25+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:02:34+00:00] done decode=64 chunk=2
+[2026-06-11T23:02:34+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:02:44+00:00] done decode=64 chunk=3
+[2026-06-11T23:02:44+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:02:53+00:00] done decode=64 chunk=4
+[2026-06-11T23:02:53+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:03:03+00:00] done decode=64 chunk=5
+[2026-06-11T23:03:03+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:03:13+00:00] done decode=64 chunk=6
+[2026-06-11T23:03:13+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:03:22+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.298351448530592	5.110950318985606	0.08453284081022087	0.4514898033947979	0.03542808297589791	62	62	63303	65513	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	22.248318536053922	5.146961055265112	0.08728895054624565	0.46623155505107833	0.036589787649961375	0	0	59118	63433	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:03:53+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.0 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.05 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:03:53+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:03:53+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:04:02+00:00] done decode=64 chunk=0
+[2026-06-11T23:04:02+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:04:12+00:00] done decode=64 chunk=1
+[2026-06-11T23:04:12+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:04:22+00:00] done decode=64 chunk=2
+[2026-06-11T23:04:22+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:04:31+00:00] done decode=64 chunk=3
+[2026-06-11T23:04:31+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:04:41+00:00] done decode=64 chunk=4
+[2026-06-11T23:04:41+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:04:51+00:00] done decode=64 chunk=5
+[2026-06-11T23:04:51+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:05:00+00:00] done decode=64 chunk=6
+[2026-06-11T23:05:00+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:05:10+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.56484571731836	5.0962369469336775	0.08549527401548351	0.45849621304666505	0.03484554658034174	63	63	62863	65489	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.012165025528127	5.134574285246102	0.08837737993748224	0.4739742031227799	0.03602664898487575	0	0	58545	63342	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:05:40+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.05 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.10 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:05:40+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:05:40+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:05:49+00:00] done decode=64 chunk=0
+[2026-06-11T23:05:49+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:05:59+00:00] done decode=64 chunk=1
+[2026-06-11T23:05:59+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:06:09+00:00] done decode=64 chunk=2
+[2026-06-11T23:06:09+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:06:18+00:00] done decode=64 chunk=3
+[2026-06-11T23:06:18+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:06:28+00:00] done decode=64 chunk=4
+[2026-06-11T23:06:28+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:06:38+00:00] done decode=64 chunk=5
+[2026-06-11T23:06:38+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:06:47+00:00] done decode=64 chunk=6
+[2026-06-11T23:06:47+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:06:57+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.868198510192713	5.095241724435187	0.08628899334594957	0.4574882102467836	0.03719247909163055	63	63	63330	65524	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.56879140784409	5.138675885649982	0.08951277057305275	0.47460096275652397	0.038588823967190794	0	0	58562	63153	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:07:28+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.10 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.20 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:07:28+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:07:28+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:07:37+00:00] done decode=64 chunk=0
+[2026-06-11T23:07:37+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:07:47+00:00] done decode=64 chunk=1
+[2026-06-11T23:07:47+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:07:57+00:00] done decode=64 chunk=2
+[2026-06-11T23:07:57+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:08:06+00:00] done decode=64 chunk=3
+[2026-06-11T23:08:06+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:08:16+00:00] done decode=64 chunk=4
+[2026-06-11T23:08:16+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:08:26+00:00] done decode=64 chunk=5
+[2026-06-11T23:08:26+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:08:35+00:00] done decode=64 chunk=6
+[2026-06-11T23:08:35+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:08:45+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.673846673753904	5.090995448591906	0.08434875553519622	0.4567942707935684	0.036952206443731865	63	63	63443	65490	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.42991338926976	5.135652426174401	0.08754021968268057	0.47410049136154697	0.03835729343329476	0	0	58626	63091	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:09:14+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.50 s=2.20 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.0 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:09:14+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:09:14+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:09:24+00:00] done decode=64 chunk=0
+[2026-06-11T23:09:24+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:09:34+00:00] done decode=64 chunk=1
+[2026-06-11T23:09:34+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:09:44+00:00] done decode=64 chunk=2
+[2026-06-11T23:09:44+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:09:53+00:00] done decode=64 chunk=3
+[2026-06-11T23:09:53+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:10:03+00:00] done decode=64 chunk=4
+[2026-06-11T23:10:03+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:10:13+00:00] done decode=64 chunk=5
+[2026-06-11T23:10:13+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:10:22+00:00] done decode=64 chunk=6
+[2026-06-11T23:10:22+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:10:33+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.723100957588755	5.097644764986179	0.0842772095863227	0.45184631119388174	0.0362998015570142	60	60	63330	65510	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	22.847010541456154	5.135425866381889	0.08702232311766932	0.4665857388343239	0.03748896456047421	0	0	59155	63432	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:10:47+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.0 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.05 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:10:47+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:10:47+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:10:57+00:00] done decode=64 chunk=0
+[2026-06-11T23:10:57+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:11:06+00:00] done decode=64 chunk=1
+[2026-06-11T23:11:06+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:11:16+00:00] done decode=64 chunk=2
+[2026-06-11T23:11:16+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:11:25+00:00] done decode=64 chunk=3
+[2026-06-11T23:11:25+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:11:35+00:00] done decode=64 chunk=4
+[2026-06-11T23:11:35+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:11:44+00:00] done decode=64 chunk=5
+[2026-06-11T23:11:44+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:11:54+00:00] done decode=64 chunk=6
+[2026-06-11T23:11:54+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:12:04+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.24820878423493	5.127880008136174	0.08418562051595176	0.45744859484956263	0.036055564036025035	63	63	63104	65510	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.652265269376624	5.167899231454428	0.08704987133542776	0.4730352688579458	0.037289045356235104	0	0	58746	63343	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:12:18+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.05 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.10 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:12:18+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:12:18+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:12:28+00:00] done decode=64 chunk=0
+[2026-06-11T23:12:28+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:12:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:12:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:12:47+00:00] done decode=64 chunk=2
+[2026-06-11T23:12:47+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:12:56+00:00] done decode=64 chunk=3
+[2026-06-11T23:12:56+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:13:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:13:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:13:16+00:00] done decode=64 chunk=5
+[2026-06-11T23:13:16+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:13:25+00:00] done decode=64 chunk=6
+[2026-06-11T23:13:25+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:13:35+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.560236435368985	5.1235796830869464	0.08949660071805057	0.47007058392153267	0.03489420212359636	62	62	62998	65455	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.54482045401448	5.165861861813713	0.09266085525795377	0.4867105429692444	0.036134094828268126	0	0	58486	63209	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:13:49+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.10 gamma=0.25 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.20 gamma=0.25 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=0
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:13:49+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:13:49+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:13:59+00:00] done decode=64 chunk=0
+[2026-06-11T23:13:59+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:14:09+00:00] done decode=64 chunk=1
+[2026-06-11T23:14:09+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:14:18+00:00] done decode=64 chunk=2
+[2026-06-11T23:14:18+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:14:28+00:00] done decode=64 chunk=3
+[2026-06-11T23:14:28+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:14:38+00:00] done decode=64 chunk=4
+[2026-06-11T23:14:38+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:14:47+00:00] done decode=64 chunk=5
+[2026-06-11T23:14:47+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:14:57+00:00] done decode=64 chunk=6
+[2026-06-11T23:14:57+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:15:06+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.573212704294516	5.100635471782279	0.08688021731499992	0.4677685193663584	0.03668716712194973	62	62	63152	65527	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.703258573121037	5.1442665553402565	0.09005189216554867	0.4848671824323255	0.03803316035944817	0	0	58495	63208	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p25_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:15:21+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=0 m=-0.45 s=2.20 gamma=0.25 =====

LTA_openwebtext_dualt/mini_owt_logdirichlet/logs/infer_lr3e3_170k_logitnorm64_fine_20260612/gpu1.log

@@ -0,0 +1,372 @@
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.0 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:54:54+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64
+[2026-06-11T22:54:54+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:55:04+00:00] done decode=64 chunk=0
+[2026-06-11T22:55:04+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:55:14+00:00] done decode=64 chunk=1
+[2026-06-11T22:55:14+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:55:23+00:00] done decode=64 chunk=2
+[2026-06-11T22:55:23+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:55:33+00:00] done decode=64 chunk=3
+[2026-06-11T22:55:33+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:55:43+00:00] done decode=64 chunk=4
+[2026-06-11T22:55:43+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:55:52+00:00] done decode=64 chunk=5
+[2026-06-11T22:55:52+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:56:02+00:00] done decode=64 chunk=6
+[2026-06-11T22:56:02+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:56:12+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.630540897089826	5.071549865171216	0.08609716265510768	0.4549589425806649	0.03631007799264335	63	63	63065	65519	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.169006113776913	5.114669951021709	0.08920521945432977	0.47140326458306975	0.03762752075919336	0	0	58451	63225	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:56:44+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.0 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.05 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:56:44+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64
+[2026-06-11T22:56:44+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:56:54+00:00] done decode=64 chunk=0
+[2026-06-11T22:56:54+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:57:03+00:00] done decode=64 chunk=1
+[2026-06-11T22:57:03+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:57:13+00:00] done decode=64 chunk=2
+[2026-06-11T22:57:13+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:57:23+00:00] done decode=64 chunk=3
+[2026-06-11T22:57:23+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:57:32+00:00] done decode=64 chunk=4
+[2026-06-11T22:57:32+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:57:42+00:00] done decode=64 chunk=5
+[2026-06-11T22:57:42+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:57:52+00:00] done decode=64 chunk=6
+[2026-06-11T22:57:52+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:58:01+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.511131107808676	5.090230639617261	0.08550713577043426	0.45639100039686176	0.03797603602228497	62	62	63077	65515	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.033685500700173	5.128988923072384	0.08846105249857855	0.4721945826423438	0.039294964937772445	0	0	58659	63316	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:58:32+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.05 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.10 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:58:32+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64
+[2026-06-11T22:58:32+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:58:42+00:00] done decode=64 chunk=0
+[2026-06-11T22:58:42+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:58:51+00:00] done decode=64 chunk=1
+[2026-06-11T22:58:51+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:59:01+00:00] done decode=64 chunk=2
+[2026-06-11T22:59:01+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:59:11+00:00] done decode=64 chunk=3
+[2026-06-11T22:59:11+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:59:20+00:00] done decode=64 chunk=4
+[2026-06-11T22:59:20+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:59:30+00:00] done decode=64 chunk=5
+[2026-06-11T22:59:30+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:59:39+00:00] done decode=64 chunk=6
+[2026-06-11T22:59:39+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:59:49+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	19.920511861299087	5.096229527472717	0.08513981480068954	0.4562929061784897	0.03685679852328721	63	63	63433	65551	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.625678269449587	5.139651730975408	0.08830091940563037	0.47325610836814785	0.038232082667384044	0	0	58698	63193	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:00:18+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.10 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.20 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:00:18+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:00:18+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:00:28+00:00] done decode=64 chunk=0
+[2026-06-11T23:00:28+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:00:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:00:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:00:47+00:00] done decode=64 chunk=2
+[2026-06-11T23:00:47+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:00:56+00:00] done decode=64 chunk=3
+[2026-06-11T23:00:56+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:01:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:01:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:01:16+00:00] done decode=64 chunk=5
+[2026-06-11T23:01:16+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:01:26+00:00] done decode=64 chunk=6
+[2026-06-11T23:01:26+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:01:35+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.780381537611536	5.104774893371332	0.08581195537717277	0.46525043494185514	0.038854212767256244	63	64	63627	65527	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.398313870582037	5.159768070386341	0.08935865850751014	0.4844549702768859	0.038909639990463325	0	0	58386	62915	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:02:05+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.55 s=2.20 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.0 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:02:05+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:02:05+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:02:15+00:00] done decode=64 chunk=0
+[2026-06-11T23:02:15+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:02:25+00:00] done decode=64 chunk=1
+[2026-06-11T23:02:25+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:02:34+00:00] done decode=64 chunk=2
+[2026-06-11T23:02:34+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:02:44+00:00] done decode=64 chunk=3
+[2026-06-11T23:02:44+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:02:54+00:00] done decode=64 chunk=4
+[2026-06-11T23:02:54+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:03:03+00:00] done decode=64 chunk=5
+[2026-06-11T23:03:03+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:03:13+00:00] done decode=64 chunk=6
+[2026-06-11T23:03:13+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:03:22+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.558454615512836	5.101042404338562	0.08569248378481496	0.45877846285330565	0.03563525371995421	63	63	62998	65525	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.852101231324507	5.138334370114406	0.0885084109792051	0.4738758907737908	0.036812814327831116	0	0	58782	63429	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:03:52+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.0 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.05 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:03:53+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:03:53+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:04:02+00:00] done decode=64 chunk=0
+[2026-06-11T23:04:02+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:04:12+00:00] done decode=64 chunk=1
+[2026-06-11T23:04:12+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:04:22+00:00] done decode=64 chunk=2
+[2026-06-11T23:04:22+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:04:31+00:00] done decode=64 chunk=3
+[2026-06-11T23:04:31+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:04:41+00:00] done decode=64 chunk=4
+[2026-06-11T23:04:41+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:04:51+00:00] done decode=64 chunk=5
+[2026-06-11T23:04:51+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:05:00+00:00] done decode=64 chunk=6
+[2026-06-11T23:05:00+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:05:10+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.252389903230533	5.079661819987348	0.0861769014772311	0.4622064187281579	0.03528262727383714	63	63	63153	65528	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.080382150603086	5.122658483091707	0.0893580653329957	0.47929031543294875	0.0365915421625728	0	0	58448	63184	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:05:40+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.05 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.10 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:05:40+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:05:40+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:05:50+00:00] done decode=64 chunk=0
+[2026-06-11T23:05:50+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:05:59+00:00] done decode=64 chunk=1
+[2026-06-11T23:05:59+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:06:09+00:00] done decode=64 chunk=2
+[2026-06-11T23:06:09+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:06:19+00:00] done decode=64 chunk=3
+[2026-06-11T23:06:19+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:06:28+00:00] done decode=64 chunk=4
+[2026-06-11T23:06:28+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:06:38+00:00] done decode=64 chunk=5
+[2026-06-11T23:06:38+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:06:47+00:00] done decode=64 chunk=6
+[2026-06-11T23:06:47+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:06:57+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.633507175507088	5.099573434515833	0.08479541085649772	0.4602334274162789	0.037622433100418025	62	62	63337	65546	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.260588803991826	5.139279731203019	0.08775503758448613	0.4763198206023088	0.0389425810119386	0	0	58868	63324	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:07:27+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.10 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.20 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:07:27+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:07:27+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:07:37+00:00] done decode=64 chunk=0
+[2026-06-11T23:07:37+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:07:47+00:00] done decode=64 chunk=1
+[2026-06-11T23:07:47+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:07:56+00:00] done decode=64 chunk=2
+[2026-06-11T23:07:56+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:08:06+00:00] done decode=64 chunk=3
+[2026-06-11T23:08:06+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:08:16+00:00] done decode=64 chunk=4
+[2026-06-11T23:08:16+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:08:26+00:00] done decode=64 chunk=5
+[2026-06-11T23:08:26+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:08:36+00:00] done decode=64 chunk=6
+[2026-06-11T23:08:36+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:08:45+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.68511482534795	5.111146238846208	0.08691868453998595	0.46666259523336945	0.0371309047604507	63	63	63307	65498	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.805994734755675	5.157701017507201	0.09024892976058348	0.48456452456832994	0.038560329792294276	0	0	58428	63070	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:09:14+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.50 s=2.20 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.0 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:09:14+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:09:14+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:09:24+00:00] done decode=64 chunk=0
+[2026-06-11T23:09:24+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:09:34+00:00] done decode=64 chunk=1
+[2026-06-11T23:09:34+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:09:43+00:00] done decode=64 chunk=2
+[2026-06-11T23:09:43+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:09:53+00:00] done decode=64 chunk=3
+[2026-06-11T23:09:53+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:10:03+00:00] done decode=64 chunk=4
+[2026-06-11T23:10:03+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:10:13+00:00] done decode=64 chunk=5
+[2026-06-11T23:10:13+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:10:22+00:00] done decode=64 chunk=6
+[2026-06-11T23:10:22+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:10:33+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.49001528337563	5.100469334309605	0.0849424971782435	0.45917542441390463	0.03730819682132943	63	63	63039	65562	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	23.66546834986224	5.137158249576254	0.08766295106744758	0.4739038022514367	0.0385099817368852	0	0	58921	63516	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:10:46+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.0 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.05 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:10:46+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:10:46+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:10:56+00:00] done decode=64 chunk=0
+[2026-06-11T23:10:56+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:11:06+00:00] done decode=64 chunk=1
+[2026-06-11T23:11:06+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:11:16+00:00] done decode=64 chunk=2
+[2026-06-11T23:11:16+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:11:25+00:00] done decode=64 chunk=3
+[2026-06-11T23:11:25+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:11:35+00:00] done decode=64 chunk=4
+[2026-06-11T23:11:35+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:11:44+00:00] done decode=64 chunk=5
+[2026-06-11T23:11:44+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:11:54+00:00] done decode=64 chunk=6
+[2026-06-11T23:11:54+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:12:03+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.998660631973852	5.117016001831258	0.08748683346817897	0.4650108387017983	0.03717159998168135	62	62	63055	65507	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.57291858869198	5.156199359444282	0.09044417084951227	0.48076649882404937	0.038434826530290116	0	0	58723	63354	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:12:17+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.05 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.10 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:12:17+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:12:17+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:12:27+00:00] done decode=64 chunk=0
+[2026-06-11T23:12:27+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:12:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:12:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:12:46+00:00] done decode=64 chunk=2
+[2026-06-11T23:12:46+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:12:56+00:00] done decode=64 chunk=3
+[2026-06-11T23:12:56+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:13:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:13:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:13:15+00:00] done decode=64 chunk=5
+[2026-06-11T23:13:15+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:13:25+00:00] done decode=64 chunk=6
+[2026-06-11T23:13:25+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:13:35+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.345432768850362	5.110604636841414	0.08801575307963548	0.4710425889177225	0.03616186594617698	62	62	63443	65511	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.270984654527286	5.15275965754805	0.0911822854883353	0.4880266987220043	0.03746935547647291	0	0	58849	63225	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:13:49+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.10 gamma=0.22 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.20 gamma=0.22 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:13:49+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:13:49+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:13:58+00:00] done decode=64 chunk=0
+[2026-06-11T23:13:58+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:14:08+00:00] done decode=64 chunk=1
+[2026-06-11T23:14:08+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:14:18+00:00] done decode=64 chunk=2
+[2026-06-11T23:14:18+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:14:27+00:00] done decode=64 chunk=3
+[2026-06-11T23:14:27+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:14:37+00:00] done decode=64 chunk=4
+[2026-06-11T23:14:37+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:14:47+00:00] done decode=64 chunk=5
+[2026-06-11T23:14:47+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:14:56+00:00] done decode=64 chunk=6
+[2026-06-11T23:14:56+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:15:06+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.143284409955957	5.13338992106818	0.08696050441977984	0.47004580152671754	0.03775514877635456	63	63	62994	65501	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	26.538093264608655	5.177653568852669	0.09016211766195935	0.4873739372734037	0.039152048635298586	0	0	58298	63164	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p22_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:15:20+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=1 m=-0.45 s=2.20 gamma=0.22 =====

LTA_openwebtext_dualt/mini_owt_logdirichlet/logs/infer_lr3e3_170k_logitnorm64_fine_20260612/gpu2.log

@@ -0,0 +1,372 @@
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.0 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:54:54+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64
+[2026-06-11T22:54:54+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:55:04+00:00] done decode=64 chunk=0
+[2026-06-11T22:55:04+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:55:14+00:00] done decode=64 chunk=1
+[2026-06-11T22:55:14+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:55:23+00:00] done decode=64 chunk=2
+[2026-06-11T22:55:23+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:55:33+00:00] done decode=64 chunk=3
+[2026-06-11T22:55:33+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:55:43+00:00] done decode=64 chunk=4
+[2026-06-11T22:55:43+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:55:52+00:00] done decode=64 chunk=5
+[2026-06-11T22:55:52+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:56:02+00:00] done decode=64 chunk=6
+[2026-06-11T22:56:02+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:56:12+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.478768157619413	5.0774058127264725	0.08568680731536042	0.4588657354400427	0.03732482520685128	61	61	62771	65506	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.293216869735062	5.118634088813846	0.08877358938260278	0.47541761579347	0.03867630542417388	0	0	58173	63217	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:56:44+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.0 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.05 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:56:44+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64
+[2026-06-11T22:56:44+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:56:54+00:00] done decode=64 chunk=0
+[2026-06-11T22:56:54+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:57:03+00:00] done decode=64 chunk=1
+[2026-06-11T22:57:03+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:57:13+00:00] done decode=64 chunk=2
+[2026-06-11T22:57:13+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:57:23+00:00] done decode=64 chunk=3
+[2026-06-11T22:57:23+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:57:33+00:00] done decode=64 chunk=4
+[2026-06-11T22:57:33+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:57:42+00:00] done decode=64 chunk=5
+[2026-06-11T22:57:42+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:57:52+00:00] done decode=64 chunk=6
+[2026-06-11T22:57:52+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:58:01+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.000409953705358	5.100488838346058	0.08481692979509574	0.4604919609729284	0.0392555043210065	62	62	62975	65494	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.789265502261582	5.140629694306197	0.08778668184045395	0.476654127019252	0.04063729906586372	0	0	58503	63267	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:58:32+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.05 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.10 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:58:32+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64
+[2026-06-11T22:58:32+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:58:42+00:00] done decode=64 chunk=0
+[2026-06-11T22:58:42+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:58:51+00:00] done decode=64 chunk=1
+[2026-06-11T22:58:51+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:59:01+00:00] done decode=64 chunk=2
+[2026-06-11T22:59:01+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:59:11+00:00] done decode=64 chunk=3
+[2026-06-11T22:59:11+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:59:20+00:00] done decode=64 chunk=4
+[2026-06-11T22:59:20+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:59:30+00:00] done decode=64 chunk=5
+[2026-06-11T22:59:30+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:59:40+00:00] done decode=64 chunk=6
+[2026-06-11T22:59:40+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:59:49+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.881217487041067	5.1069710432444415	0.08597064119388409	0.4641173149405643	0.03750724814600055	63	63	63126	65534	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.854964325721426	5.150513871914073	0.08912128595386513	0.4811486433035361	0.038888713096857895	0	0	58445	63206	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:00:18+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.10 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.20 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:00:18+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:00:18+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:00:28+00:00] done decode=64 chunk=0
+[2026-06-11T23:00:28+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:00:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:00:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:00:47+00:00] done decode=64 chunk=2
+[2026-06-11T23:00:47+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:00:57+00:00] done decode=64 chunk=3
+[2026-06-11T23:00:57+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:01:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:01:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:01:16+00:00] done decode=64 chunk=5
+[2026-06-11T23:01:16+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:01:26+00:00] done decode=64 chunk=6
+[2026-06-11T23:01:26+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:01:36+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.196859440627485	5.10022806328218	0.08602840128263857	0.469697201056666	0.038692930218353946	63	63	63340	65490	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.93419812863765	5.154068383731364	0.08951642379264863	0.48876458061850425	0.04026888299140274	0	0	58199	62927	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:02:05+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.55 s=2.20 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.0 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:02:05+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:02:05+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:02:15+00:00] done decode=64 chunk=0
+[2026-06-11T23:02:15+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:02:25+00:00] done decode=64 chunk=1
+[2026-06-11T23:02:25+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:02:34+00:00] done decode=64 chunk=2
+[2026-06-11T23:02:34+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:02:44+00:00] done decode=64 chunk=3
+[2026-06-11T23:02:44+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:02:54+00:00] done decode=64 chunk=4
+[2026-06-11T23:02:54+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:03:03+00:00] done decode=64 chunk=5
+[2026-06-11T23:03:03+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:03:13+00:00] done decode=64 chunk=6
+[2026-06-11T23:03:13+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:03:23+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.971987891927146	5.112778032829933	0.0875883439422387	0.46737902610288506	0.03613133672207721	63	63	63005	65511	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	24.58042427911596	5.1533154326633674	0.09056752703449365	0.48329754988633494	0.03736680085247454	0	0	58646	63345	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:03:52+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.0 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.05 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:03:53+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:03:53+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:04:02+00:00] done decode=64 chunk=0
+[2026-06-11T23:04:02+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:04:12+00:00] done decode=64 chunk=1
+[2026-06-11T23:04:12+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:04:21+00:00] done decode=64 chunk=2
+[2026-06-11T23:04:21+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:04:31+00:00] done decode=64 chunk=3
+[2026-06-11T23:04:31+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:04:41+00:00] done decode=64 chunk=4
+[2026-06-11T23:04:41+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:04:51+00:00] done decode=64 chunk=5
+[2026-06-11T23:04:51+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:05:00+00:00] done decode=64 chunk=6
+[2026-06-11T23:05:01+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:05:10+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.811089787451706	5.107088838273693	0.08871103236547478	0.4714983962120055	0.03757388767545936	63	63	62765	65471	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.946598562651285	5.148759450854878	0.09188291139240506	0.48839380369942564	0.03892405063291139	0	0	58200	63200	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:05:40+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.05 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.10 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:05:40+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:05:40+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:05:49+00:00] done decode=64 chunk=0
+[2026-06-11T23:05:49+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:05:59+00:00] done decode=64 chunk=1
+[2026-06-11T23:05:59+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:06:09+00:00] done decode=64 chunk=2
+[2026-06-11T23:06:09+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:06:19+00:00] done decode=64 chunk=3
+[2026-06-11T23:06:19+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:06:28+00:00] done decode=64 chunk=4
+[2026-06-11T23:06:28+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:06:38+00:00] done decode=64 chunk=5
+[2026-06-11T23:06:38+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:06:48+00:00] done decode=64 chunk=6
+[2026-06-11T23:06:48+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:06:57+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.198665657056562	5.134226876468917	0.0889577576599545	0.4758175322890904	0.03850205334106835	63	63	62922	65503	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	26.305322952380674	5.174869979668152	0.09206700379266751	0.4924698557183268	0.03985461441213654	0	0	58455	63280	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:07:27+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.10 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.20 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:07:27+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:07:27+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:07:37+00:00] done decode=64 chunk=0
+[2026-06-11T23:07:37+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:07:47+00:00] done decode=64 chunk=1
+[2026-06-11T23:07:47+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:07:56+00:00] done decode=64 chunk=2
+[2026-06-11T23:07:56+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:08:06+00:00] done decode=64 chunk=3
+[2026-06-11T23:08:06+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:08:17+00:00] done decode=64 chunk=4
+[2026-06-11T23:08:17+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:08:26+00:00] done decode=64 chunk=5
+[2026-06-11T23:08:26+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:08:36+00:00] done decode=64 chunk=6
+[2026-06-11T23:08:36+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:08:45+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.120772841755542	5.112763961217023	0.09020272333150149	0.4771703787381501	0.038605971789705075	63	63	63181	65508	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	27.17844323202931	5.167347517503274	0.09390148926363304	0.49675758153728783	0.04019581353211374	0	0	57981	62917	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:09:14+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.50 s=2.20 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.0 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:09:14+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:09:14+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:09:24+00:00] done decode=64 chunk=0
+[2026-06-11T23:09:24+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:09:34+00:00] done decode=64 chunk=1
+[2026-06-11T23:09:34+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:09:44+00:00] done decode=64 chunk=2
+[2026-06-11T23:09:44+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:09:53+00:00] done decode=64 chunk=3
+[2026-06-11T23:09:53+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:10:03+00:00] done decode=64 chunk=4
+[2026-06-11T23:10:03+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:10:12+00:00] done decode=64 chunk=5
+[2026-06-11T23:10:12+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:10:22+00:00] done decode=64 chunk=6
+[2026-06-11T23:10:22+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:10:33+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.840770117094618	5.119180338746435	0.08761989125786548	0.4677052310042001	0.036211741706884966	63	63	62917	65476	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.348209740959422	5.155436736747132	0.09041756648118665	0.4826759986128188	0.037374485726445875	0	0	58812	63439	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:10:46+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.0 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.05 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:10:47+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:10:47+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:10:56+00:00] done decode=64 chunk=0
+[2026-06-11T23:10:56+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:11:06+00:00] done decode=64 chunk=1
+[2026-06-11T23:11:06+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:11:15+00:00] done decode=64 chunk=2
+[2026-06-11T23:11:15+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:11:25+00:00] done decode=64 chunk=3
+[2026-06-11T23:11:25+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:11:35+00:00] done decode=64 chunk=4
+[2026-06-11T23:11:35+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:11:44+00:00] done decode=64 chunk=5
+[2026-06-11T23:11:44+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:11:54+00:00] done decode=64 chunk=6
+[2026-06-11T23:11:54+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:12:04+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.36629336381291	5.124589631956277	0.09021133745326924	0.4760124515518662	0.03730830853742275	63	63	62940	65535	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	26.495583958795436	5.166288290684626	0.0933497575843717	0.49259317751105497	0.038612782489221585	0	0	58489	63321	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:12:17+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.05 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.10 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:12:17+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:12:17+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:12:27+00:00] done decode=64 chunk=0
+[2026-06-11T23:12:27+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:12:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:12:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:12:47+00:00] done decode=64 chunk=2
+[2026-06-11T23:12:47+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:12:56+00:00] done decode=64 chunk=3
+[2026-06-11T23:12:56+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:13:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:13:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:13:16+00:00] done decode=64 chunk=5
+[2026-06-11T23:13:16+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:13:25+00:00] done decode=64 chunk=6
+[2026-06-11T23:13:25+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:13:35+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.212625942184832	5.120156751416636	0.09033302285810264	0.47735532142311804	0.03798995281794445	62	62	62896	65491	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	26.384790395044394	5.1623792253532566	0.09353996995334862	0.4943384887876775	0.03934529928046177	0	0	58360	63235	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:13:49+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.10 gamma=0.20 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.20 gamma=0.20 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:13:49+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:13:49+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:13:59+00:00] done decode=64 chunk=0
+[2026-06-11T23:13:59+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:14:08+00:00] done decode=64 chunk=1
+[2026-06-11T23:14:08+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:14:18+00:00] done decode=64 chunk=2
+[2026-06-11T23:14:18+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:14:28+00:00] done decode=64 chunk=3
+[2026-06-11T23:14:28+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:14:38+00:00] done decode=64 chunk=4
+[2026-06-11T23:14:38+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:14:47+00:00] done decode=64 chunk=5
+[2026-06-11T23:14:47+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:14:57+00:00] done decode=64 chunk=6
+[2026-06-11T23:14:57+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:15:06+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.797329500924356	5.163888926598371	0.090329079919409	0.48543081737006794	0.03728860125770804	63	63	63019	65516	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	28.74131987282973	5.209628330025424	0.09367826100723525	0.5034514423229157	0.0386777068853601	0	0	58287	63163	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p20_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:15:21+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=2 m=-0.45 s=2.20 gamma=0.20 =====

LTA_openwebtext_dualt/mini_owt_logdirichlet/logs/infer_lr3e3_170k_logitnorm64_fine_20260612/gpu3.log

@@ -0,0 +1,372 @@
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.0 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:54:54+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64
+[2026-06-11T22:54:54+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:55:04+00:00] done decode=64 chunk=0
+[2026-06-11T22:55:04+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:55:14+00:00] done decode=64 chunk=1
+[2026-06-11T22:55:14+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:55:23+00:00] done decode=64 chunk=2
+[2026-06-11T22:55:23+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:55:33+00:00] done decode=64 chunk=3
+[2026-06-11T22:55:33+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:55:43+00:00] done decode=64 chunk=4
+[2026-06-11T22:55:43+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:55:52+00:00] done decode=64 chunk=5
+[2026-06-11T22:55:52+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:56:02+00:00] done decode=64 chunk=6
+[2026-06-11T22:56:02+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:56:12+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.820384793227223	5.092453167499823	0.08747156792403865	0.46494977559307543	0.03733952096722488	62	62	62905	65507	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.97591339181247	5.1339424553543624	0.09064012910166756	0.4818289692271181	0.03869885770338259	0	0	58277	63206	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:56:44+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.0 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.05 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:56:44+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64
+[2026-06-11T22:56:44+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:56:54+00:00] done decode=64 chunk=0
+[2026-06-11T22:56:54+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:57:03+00:00] done decode=64 chunk=1
+[2026-06-11T22:57:03+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:57:13+00:00] done decode=64 chunk=2
+[2026-06-11T22:57:13+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:57:23+00:00] done decode=64 chunk=3
+[2026-06-11T22:57:23+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:57:32+00:00] done decode=64 chunk=4
+[2026-06-11T22:57:32+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:57:42+00:00] done decode=64 chunk=5
+[2026-06-11T22:57:42+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:57:52+00:00] done decode=64 chunk=6
+[2026-06-11T22:57:52+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:58:01+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.54611456429853	5.101580271624245	0.08629216913448329	0.464088903814743	0.037536254007021826	62	62	63325	65510	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.693423601513754	5.145302553634024	0.08945302608255254	0.4811104252726208	0.03891808052671562	0	0	58654	63184	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T22:58:32+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.05 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.10 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:58:32+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64
+[2026-06-11T22:58:32+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T22:58:41+00:00] done decode=64 chunk=0
+[2026-06-11T22:58:41+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T22:58:51+00:00] done decode=64 chunk=1
+[2026-06-11T22:58:51+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T22:59:01+00:00] done decode=64 chunk=2
+[2026-06-11T22:59:01+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T22:59:10+00:00] done decode=64 chunk=3
+[2026-06-11T22:59:10+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T22:59:20+00:00] done decode=64 chunk=4
+[2026-06-11T22:59:20+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T22:59:30+00:00] done decode=64 chunk=5
+[2026-06-11T22:59:30+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T22:59:40+00:00] done decode=64 chunk=6
+[2026-06-11T22:59:40+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T22:59:49+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.797791429207887	5.105724416228216	0.0880471457579505	0.4716785245351003	0.03714560527641644	63	63	63086	65499	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	26.12017245330134	5.150338866267617	0.091296293364156	0.48910633985686236	0.03852304574314803	0	0	58378	63157	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:00:18+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.10 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.20 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:00:18+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:00:18+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:00:27+00:00] done decode=64 chunk=0
+[2026-06-11T23:00:27+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:00:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:00:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:00:47+00:00] done decode=64 chunk=2
+[2026-06-11T23:00:47+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:00:56+00:00] done decode=64 chunk=3
+[2026-06-11T23:00:56+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:01:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:01:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:01:16+00:00] done decode=64 chunk=5
+[2026-06-11T23:01:16+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:01:26+00:00] done decode=64 chunk=6
+[2026-06-11T23:01:26+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:01:35+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.966402902891225	5.114466078911025	0.08745438722384231	0.47164712348845733	0.03914683115257187	63	63	63657	65497	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	27.092020454611948	5.17086862661665	0.09109273103228885	0.49129141548298844	0.03890567838396691	0	0	58378	62870	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p55_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:02:05+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.55 s=2.20 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.0 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:02:05+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:02:05+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:02:15+00:00] done decode=64 chunk=0
+[2026-06-11T23:02:15+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:02:25+00:00] done decode=64 chunk=1
+[2026-06-11T23:02:25+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:02:35+00:00] done decode=64 chunk=2
+[2026-06-11T23:02:35+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:02:44+00:00] done decode=64 chunk=3
+[2026-06-11T23:02:44+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:02:54+00:00] done decode=64 chunk=4
+[2026-06-11T23:02:54+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:03:03+00:00] done decode=64 chunk=5
+[2026-06-11T23:03:03+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:03:13+00:00] done decode=64 chunk=6
+[2026-06-11T23:03:13+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:03:23+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.576272833455413	5.1272312924843355	0.08978408484016175	0.47369304483169045	0.03915465018692302	63	63	62638	65535	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	26.631393176856008	5.1669855214553575	0.0928474479972223	0.48987547545019805	0.040497459044853384	0	0	58269	63362	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:03:52+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.0 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.05 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:03:52+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:03:52+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:04:02+00:00] done decode=64 chunk=0
+[2026-06-11T23:04:02+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:04:12+00:00] done decode=64 chunk=1
+[2026-06-11T23:04:12+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:04:21+00:00] done decode=64 chunk=2
+[2026-06-11T23:04:21+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:04:31+00:00] done decode=64 chunk=3
+[2026-06-11T23:04:31+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:04:41+00:00] done decode=64 chunk=4
+[2026-06-11T23:04:41+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:04:50+00:00] done decode=64 chunk=5
+[2026-06-11T23:04:50+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:05:00+00:00] done decode=64 chunk=6
+[2026-06-11T23:05:00+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:05:10+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.39306699391356	5.112481520105167	0.08836052314313184	0.47216372127094586	0.03879316922795183	63	63	62980	65527	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	26.77029538182209	5.155530012529588	0.09159810126582278	0.4895014161616481	0.040221518987341776	0	0	58295	63200	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:05:40+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.05 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.10 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:05:40+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:05:40+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:05:49+00:00] done decode=64 chunk=0
+[2026-06-11T23:05:49+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:05:59+00:00] done decode=64 chunk=1
+[2026-06-11T23:05:59+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:06:09+00:00] done decode=64 chunk=2
+[2026-06-11T23:06:09+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:06:19+00:00] done decode=64 chunk=3
+[2026-06-11T23:06:19+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:06:28+00:00] done decode=64 chunk=4
+[2026-06-11T23:06:28+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:06:38+00:00] done decode=64 chunk=5
+[2026-06-11T23:06:38+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:06:48+00:00] done decode=64 chunk=6
+[2026-06-11T23:06:48+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:06:57+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.93466837614181	5.140322965278581	0.09127547515456835	0.47891731802638005	0.03801236546828486	63	63	63217	65505	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	27.56978826344744	5.185988065982575	0.09467099532821284	0.49675345242620045	0.03943305091456172	0	0	58475	63145	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:07:27+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.10 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.20 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:07:27+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:07:27+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:07:37+00:00] done decode=64 chunk=0
+[2026-06-11T23:07:37+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:07:46+00:00] done decode=64 chunk=1
+[2026-06-11T23:07:46+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:07:56+00:00] done decode=64 chunk=2
+[2026-06-11T23:07:56+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:08:06+00:00] done decode=64 chunk=3
+[2026-06-11T23:08:06+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:08:16+00:00] done decode=64 chunk=4
+[2026-06-11T23:08:16+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:08:26+00:00] done decode=64 chunk=5
+[2026-06-11T23:08:26+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:08:35+00:00] done decode=64 chunk=6
+[2026-06-11T23:08:35+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:08:45+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.861144387568572	5.116649801639649	0.0893946790334391	0.4796265503195985	0.03946485233097388	62	62	63069	65552	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	28.432572101027276	5.173629785859694	0.09314341128404051	0.4997615338145051	0.04034783714608206	0	0	57747	62903	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p50_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:09:14+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.50 s=2.20 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.0 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:09:14+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64
+[2026-06-11T23:09:14+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:09:24+00:00] done decode=64 chunk=0
+[2026-06-11T23:09:24+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:09:34+00:00] done decode=64 chunk=1
+[2026-06-11T23:09:34+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:09:44+00:00] done decode=64 chunk=2
+[2026-06-11T23:09:44+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:09:53+00:00] done decode=64 chunk=3
+[2026-06-11T23:09:53+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:10:03+00:00] done decode=64 chunk=4
+[2026-06-11T23:10:03+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:10:13+00:00] done decode=64 chunk=5
+[2026-06-11T23:10:13+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:10:22+00:00] done decode=64 chunk=6
+[2026-06-11T23:10:22+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:10:32+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.95881177603283	5.1306873420987165	0.0883767076242082	0.47278444301981254	0.03831183698389682	63	63	62857	65515	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	25.9713290669259	5.171960966596868	0.09144328431295118	0.4892111332259186	0.03964806419511271	0	0	58415	63307	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p0_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:10:46+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.0 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.05 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:10:46+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64
+[2026-06-11T23:10:46+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:10:56+00:00] done decode=64 chunk=0
+[2026-06-11T23:10:56+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:11:06+00:00] done decode=64 chunk=1
+[2026-06-11T23:11:06+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:11:15+00:00] done decode=64 chunk=2
+[2026-06-11T23:11:15+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:11:25+00:00] done decode=64 chunk=3
+[2026-06-11T23:11:25+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:11:35+00:00] done decode=64 chunk=4
+[2026-06-11T23:11:35+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:11:44+00:00] done decode=64 chunk=5
+[2026-06-11T23:11:44+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:11:54+00:00] done decode=64 chunk=6
+[2026-06-11T23:11:54+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:12:04+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.464099888998135	5.137750338955095	0.09133793356130923	0.47950475550738136	0.03869992672203224	62	62	62350	65504	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	27.88536713328439	5.177994754518888	0.09449192032476661	0.49609831611537614	0.04004296523291263	0	0	57933	63307	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p05_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:12:18+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.05 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.10 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:12:18+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64
+[2026-06-11T23:12:18+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:12:27+00:00] done decode=64 chunk=0
+[2026-06-11T23:12:27+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:12:37+00:00] done decode=64 chunk=1
+[2026-06-11T23:12:37+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:12:47+00:00] done decode=64 chunk=2
+[2026-06-11T23:12:47+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:12:56+00:00] done decode=64 chunk=3
+[2026-06-11T23:12:56+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:13:06+00:00] done decode=64 chunk=4
+[2026-06-11T23:13:06+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:13:16+00:00] done decode=64 chunk=5
+[2026-06-11T23:13:16+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:13:25+00:00] done decode=64 chunk=6
+[2026-06-11T23:13:25+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:13:35+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.82403171106807	5.126190539207766	0.08972088193977891	0.4756226046326976	0.03904293654186771	63	63	63021	65492	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	27.35261081936804	5.168492977700461	0.09301036966674582	0.4930973339243873	0.04048127918942452	0	0	58341	63165	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p10_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:13:49+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.10 gamma=0.18 =====
+===== START \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.20 gamma=0.18 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=64
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T23:13:49+00:00] infer step=170000 decode=64 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64
+[2026-06-11T23:13:49+00:00] run decode=64 chunk=0 n=8 seed=123
+[2026-06-11T23:13:59+00:00] done decode=64 chunk=0
+[2026-06-11T23:13:59+00:00] run decode=64 chunk=1 n=8 seed=124
+[2026-06-11T23:14:08+00:00] done decode=64 chunk=1
+[2026-06-11T23:14:08+00:00] run decode=64 chunk=2 n=8 seed=125
+[2026-06-11T23:14:18+00:00] done decode=64 chunk=2
+[2026-06-11T23:14:18+00:00] run decode=64 chunk=3 n=8 seed=126
+[2026-06-11T23:14:27+00:00] done decode=64 chunk=3
+[2026-06-11T23:14:27+00:00] run decode=64 chunk=4 n=8 seed=127
+[2026-06-11T23:14:37+00:00] done decode=64 chunk=4
+[2026-06-11T23:14:37+00:00] run decode=64 chunk=5 n=8 seed=128
+[2026-06-11T23:14:47+00:00] done decode=64 chunk=5
+[2026-06-11T23:14:47+00:00] run decode=64 chunk=6 n=8 seed=129
+[2026-06-11T23:14:57+00:00] done decode=64 chunk=6
+[2026-06-11T23:14:57+00:00] run decode=64 chunk=7 n=8 seed=130
+[2026-06-11T23:15:07+00:00] done decode=64 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	24.674709844007065	5.14755752202379	0.092134591310189	0.4913207431947604	0.03769350554181552	62	62	62613	65502	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+sc1p0	pre_eos	30.11061890078292	5.194268034449904	0.09566154675289679	0.5100814760802713	0.03913643064339721	0	0	57762	63087	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm64_fine_ppl24_ent515_step170000_ema_dgamma0p18_tschedlogit_normal_mn0p45_s2p20_sc1p0_decode64_n64/sc1p0
+[2026-06-11T23:15:21+00:00] done
+===== DONE \2026-06-11T22:54:54+00:00 gpu=3 m=-0.45 s=2.20 gamma=0.18 =====