Add files using upload-large-folder tool

results/conditional/samples/lm1b/candi/steps_1024/stdout.log

@@ -0,0 +1,4 @@
+[2026-09-01 17:21:37,184][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+Generative perplexity: 88.07493591308594
+Sample entropy: 4.3592610359191895
+Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_1024/upstream_samples.json

results/conditional/samples/lm1b/candi/steps_1024/upstream_hydra/dlb_pinned_upstream_main.log

@@ -0,0 +1,11 @@
+[2026-08-26 13:57:05,696][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 09:54:31,621][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 10:05:35,198][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 10:25:42,984][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 16:53:25,848][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 18:19:38,189][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 21:49:46,357][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-28 11:36:36,704][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-28 20:00:59,183][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-09-01 17:21:37,184][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-09-02 10:43:18,232][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.

results/conditional/samples/lm1b/candi/steps_16/stderr.log

@@ -0,0 +1,13 @@
+Seed set to 42
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:691: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
+  self.continuous_noise = torch.tensor(self.continuous_noise)
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:692: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
+  self.discrete_noise = torch.tensor(self.discrete_noise)
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:861: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(dtype=torch.float32):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:533: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(dtype=torch.bfloat16):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:157: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(enabled=False):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:112: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(enabled=False):

results/conditional/samples/lm1b/candi/steps_16/stdout.log

@@ -0,0 +1,4 @@
+[2026-08-28 19:23:32,576][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+Generative perplexity: 101.40461730957031
+Sample entropy: 4.364837169647217
+Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_samples.json

results/conditional/samples/lm1b/candi/steps_16/upstream_hydra/.hydra/hydra.yaml

@@ -0,0 +1,185 @@
+hydra:
+  run:
+    dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_hydra
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_hydra
+    - hydra.mode=RUN
+    task:
+    - mode=sample_eval
+    - seed=42
+    - data=lm1b-wrap
+    - model=small
+    - model.length=128
+    - loader.batch_size=16
+    - loader.eval_batch_size=2
+    - sampling.num_sample_batches=1024
+    - sampling.steps=16
+    - eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+    - trainer.devices=1
+    - checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_hydra
+    - +wandb.offline=true
+    - eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_samples.json
+    - algo=candi
+    - algo.sampler=cached
+    - algo.mixed_coeff=0.5
+    - algo.step_size=1.0
+    - algo.temp=1.0
+    - algo.use_percentile_scheduling=True
+    - sampling.noise_removal=ancestral
+    - trainer.accumulate_grad_batches=1
+    - eval.disable_ema=False
+  job:
+    name: dlb_pinned_upstream_main
+    chdir: true
+    override_dirname: +wandb.offline=true,algo.mixed_coeff=0.5,algo.sampler=cached,algo.step_size=1.0,algo.temp=1.0,algo.use_percentile_scheduling=True,algo=candi,checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_hydra,data=lm1b-wrap,eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt,eval.disable_ema=False,eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_samples.json,loader.batch_size=16,loader.eval_batch_size=2,mode=sample_eval,model.length=128,model=small,sampling.noise_removal=ancestral,sampling.num_sample_batches=1024,sampling.steps=16,seed=42,trainer.accumulate_grad_batches=1,trainer.devices=1
+    id: ???
+    num: ???
+    config_name: config
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/ruizzhao/diffusion_baseline
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /home/ruizzhao/diffusion_baseline/upstreams/candi/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_hydra
+    choices:
+      algo: candi
+      prior: none
+      lr_scheduler: constant
+      noise: log-linear
+      strategy: ddp
+      model: small
+      data: lm1b-wrap
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

results/conditional/samples/lm1b/candi/steps_16/upstream_hydra/.hydra/overrides.yaml

@@ -0,0 +1,23 @@
+- mode=sample_eval
+- seed=42
+- data=lm1b-wrap
+- model=small
+- model.length=128
+- loader.batch_size=16
+- loader.eval_batch_size=2
+- sampling.num_sample_batches=1024
+- sampling.steps=16
+- eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+- trainer.devices=1
+- checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_hydra
+- +wandb.offline=true
+- eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_16/upstream_samples.json
+- algo=candi
+- algo.sampler=cached
+- algo.mixed_coeff=0.5
+- algo.step_size=1.0
+- algo.temp=1.0
+- algo.use_percentile_scheduling=True
+- sampling.noise_removal=ancestral
+- trainer.accumulate_grad_batches=1
+- eval.disable_ema=False

results/conditional/samples/lm1b/candi/steps_2/conversion_metadata.json

@@ -0,0 +1,28 @@
+{
+  "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+  "checkpoint_selection": {
+    "path": "lm1b_CANDI.ckpt",
+    "resource": "flm_lm1b_reproductions",
+    "teacher_family": "hybrid_candi"
+  },
+  "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+  "conditioning_implementation": "native_projection",
+  "exclude_from_latency": true,
+  "format": "dlb-conditional-token-capture-v1",
+  "generated_samples": 2048,
+  "generation_seconds_sentinel": 0.0,
+  "generation_seconds_source": "unavailable_excluded_sentinel",
+  "paper_conditional_training_reproduced": false,
+  "requested_samples": 2048,
+  "teacher_family": "hybrid_candi",
+  "token_ids_source": "upstream_conditional_capture",
+  "token_ids_transformation": {
+    "max_length": 128,
+    "operation": "validated_exact_length"
+  },
+  "tokenizer": "bert-base-uncased",
+  "tokenizer_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "trim_policy": "none_exact_conditional_schedule",
+  "trimmed_samples": 0,
+  "upstream_output": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_token_ids.json"
+}

results/conditional/samples/lm1b/candi/steps_2/request.json

@@ -0,0 +1,112 @@
+{
+  "command": [
+    "/opt/anaconda3/envs/dlb-candi/bin/python",
+    "-u",
+    "-m",
+    "dlb.adapters.capture",
+    "--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/candi/main.py",
+    "--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_token_ids.json",
+    "--generation-mode=conditional_prefix",
+    "--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length=64",
+    "--evaluation-continuation-length=64",
+    "--prompt-count=1024",
+    "--diversity-prompt-count=256",
+    "--completions-per-diversity-prompt=5",
+    "--completion-schedule=c0:p0-1023;c1-4:p0-255",
+    "--",
+    "mode=sample_eval",
+    "seed=42",
+    "data=lm1b-wrap",
+    "model=small",
+    "model.length=128",
+    "loader.batch_size=16",
+    "loader.eval_batch_size=2",
+    "sampling.num_sample_batches=1024",
+    "sampling.steps=2",
+    "eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt",
+    "trainer.devices=1",
+    "hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra",
+    "checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra",
+    "+wandb.offline=true",
+    "eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_samples.json",
+    "algo=candi",
+    "algo.sampler=cached",
+    "algo.mixed_coeff=0.5",
+    "algo.step_size=1.0",
+    "algo.temp=1.0",
+    "algo.use_percentile_scheduling=True",
+    "sampling.noise_removal=ancestral",
+    "trainer.accumulate_grad_batches=1",
+    "eval.disable_ema=False"
+  ],
+  "identity": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command_sha256": "937f6b0870c9cb2d53e8df907893cd97e2d389b9b29c95467f98539c54f80904",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-2",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 2
+  },
+  "request": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command": null,
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-2",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 2
+  },
+  "started_at": "2026-08-28T10:18:33.969371Z"
+}

results/conditional/samples/lm1b/candi/steps_2/upstream_hydra/.hydra/config.yaml

@@ -0,0 +1,154 @@
+mode: sample_eval
+seed: 42
+scratch_dir: /home/patrick/.cache/discrete_diffusion
+batch_size: 512
+loader:
+  global_batch_size: ${batch_size}
+  eval_global_batch_size: 128
+  batch_size: 16
+  eval_batch_size: 2
+  num_workers: 4
+  pin_memory: true
+sampling:
+  predictor: ancestral_cache
+  steps: 2
+  noise_removal: ancestral
+  use_float64: true
+  p_nucleus: 1.0
+  num_sample_batches: 1024
+  num_sample_log: 2
+  semi_ar: false
+  stride_length: 1
+  num_strides: 1
+training:
+  ema: 0.9999
+  antithetic_sampling: true
+  importance_sampling: false
+  sampling_eps: 0.001
+  change_of_variables: false
+  loss_precision: float32
+  finetune_path: ''
+eval:
+  checkpoint_path: /home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+  disable_ema: false
+  compute_generative_perplexity: false
+  perplexity_batch_size: 8
+  compute_perplexity_on_sanity: false
+  gen_ppl_eval_model_name_or_path: gpt2-large
+  generate_samples: true
+  generated_samples_path: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_samples.json
+optim:
+  weight_decay: 0
+  lr: 0.001
+  beta1: 0.9
+  beta2: 0.999
+  eps: 1.0e-08
+trainer:
+  _target_: lightning.Trainer
+  accelerator: cuda
+  num_nodes: 1
+  devices: 1
+  accumulate_grad_batches: 1
+  gradient_clip_val: 1.0
+  precision: bf16
+  num_sanity_val_steps: 2
+  max_steps: 1000000
+  log_every_n_steps: 100
+  limit_train_batches: 1.0
+  limit_val_batches: 1.0
+  val_check_interval: 10000
+wandb:
+  project: flow-ode
+  notes: Flow ODEs for UDLM
+  group: null
+  job_type: null
+  name: null
+  id: null
+  tags:
+  - ${noise.type}
+  - ${data.train}
+  - ${data.valid}
+  - ${algo.name}
+  offline: true
+checkpointing:
+  save_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra
+  resume_from_ckpt: true
+  resume_ckpt_path: ${.save_dir}/checkpoints/last.ckpt
+callbacks:
+  checkpoint_every_n_steps:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    save_top_k: -1
+    save_last: true
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    verbose: true
+    auto_insert_metric_name: false
+    every_n_train_steps: 500
+  checkpoint_monitor:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    monitor: val/nll
+    mode: min
+    save_top_k: 1
+    save_last: false
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    filename: best
+    auto_insert_metric_name: false
+    verbose: true
+  learning_rate_monitor:
+    _target_: lightning.pytorch.callbacks.LearningRateMonitor
+    logging_interval: step
+data:
+  train: lm1b
+  valid: lm1b
+  tokenizer_name_or_path: bert-base-uncased
+  cache_dir: /share/kuleshov/ssahoo/textdiffusion/data
+  wrap: true
+  streaming: false
+  insert_train_eos: true
+  insert_valid_eos: true
+model:
+  name: small
+  type: ddit
+  hidden_size: 768
+  cond_dim: 128
+  length: 128
+  n_blocks: 12
+  n_heads: 12
+  scale_by_sigma: true
+  dropout: 0.1
+  tie_word_embeddings: false
+  vocab_lookup: true
+strategy:
+  _target_: lightning.pytorch.strategies.DDPStrategy
+  find_unused_parameters: false
+noise:
+  type: log-linear
+  parameterization: log-linear
+  eps: 0
+  denoiser_latent_conditioning: -1
+  freeze_encoder: false
+  freeze_decoder: false
+lr_scheduler:
+  _target_: transformers.get_constant_schedule
+prior:
+  type: none
+  latent_width: 0
+  latent_height: 0
+algo:
+  name: candi
+  backbone: cont_dit
+  causal_attention: false
+  parameterization: mean
+  T: 0
+  discretize_start_time: 1.0
+  time_conditioning: true
+  step_size: 1.0
+  mixed_coeff: 0.5
+  max_percentile: 0.25
+  min_percentile: 0.01
+  sigma_min: 0.01
+  sigma_max: 2.0
+  temp: 1.0
+  pure_continuous: false
+  sampler: cached
+  is_embed: false
+  use_percentile_scheduling: true

results/conditional/samples/lm1b/candi/steps_2/upstream_hydra/.hydra/hydra.yaml

@@ -0,0 +1,185 @@
+hydra:
+  run:
+    dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra
+    - hydra.mode=RUN
+    task:
+    - mode=sample_eval
+    - seed=42
+    - data=lm1b-wrap
+    - model=small
+    - model.length=128
+    - loader.batch_size=16
+    - loader.eval_batch_size=2
+    - sampling.num_sample_batches=1024
+    - sampling.steps=2
+    - eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+    - trainer.devices=1
+    - checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra
+    - +wandb.offline=true
+    - eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_samples.json
+    - algo=candi
+    - algo.sampler=cached
+    - algo.mixed_coeff=0.5
+    - algo.step_size=1.0
+    - algo.temp=1.0
+    - algo.use_percentile_scheduling=True
+    - sampling.noise_removal=ancestral
+    - trainer.accumulate_grad_batches=1
+    - eval.disable_ema=False
+  job:
+    name: dlb_pinned_upstream_main
+    chdir: true
+    override_dirname: +wandb.offline=true,algo.mixed_coeff=0.5,algo.sampler=cached,algo.step_size=1.0,algo.temp=1.0,algo.use_percentile_scheduling=True,algo=candi,checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra,data=lm1b-wrap,eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt,eval.disable_ema=False,eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_samples.json,loader.batch_size=16,loader.eval_batch_size=2,mode=sample_eval,model.length=128,model=small,sampling.noise_removal=ancestral,sampling.num_sample_batches=1024,sampling.steps=2,seed=42,trainer.accumulate_grad_batches=1,trainer.devices=1
+    id: ???
+    num: ???
+    config_name: config
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/ruizzhao/diffusion_baseline
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /home/ruizzhao/diffusion_baseline/upstreams/candi/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra
+    choices:
+      algo: candi
+      prior: none
+      lr_scheduler: constant
+      noise: log-linear
+      strategy: ddp
+      model: small
+      data: lm1b-wrap
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

results/conditional/samples/lm1b/candi/steps_2/upstream_hydra/.hydra/overrides.yaml

@@ -0,0 +1,23 @@
+- mode=sample_eval
+- seed=42
+- data=lm1b-wrap
+- model=small
+- model.length=128
+- loader.batch_size=16
+- loader.eval_batch_size=2
+- sampling.num_sample_batches=1024
+- sampling.steps=2
+- eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+- trainer.devices=1
+- checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_hydra
+- +wandb.offline=true
+- eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_2/upstream_samples.json
+- algo=candi
+- algo.sampler=cached
+- algo.mixed_coeff=0.5
+- algo.step_size=1.0
+- algo.temp=1.0
+- algo.use_percentile_scheduling=True
+- sampling.noise_removal=ancestral
+- trainer.accumulate_grad_batches=1
+- eval.disable_ema=False

results/conditional/samples/lm1b/candi/steps_32/run_metadata.json

@@ -0,0 +1,102 @@
+{
+  "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+  "checkpoint_selection": {
+    "path": "lm1b_CANDI.ckpt",
+    "resource": "flm_lm1b_reproductions",
+    "teacher_family": "hybrid_candi"
+  },
+  "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+  "checkpoint_teacher_family": "hybrid_candi",
+  "command": [
+    "/opt/anaconda3/envs/dlb-candi/bin/python",
+    "-u",
+    "-m",
+    "dlb.adapters.capture",
+    "--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/candi/main.py",
+    "--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_32/upstream_token_ids.json",
+    "--generation-mode=conditional_prefix",
+    "--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length=64",
+    "--evaluation-continuation-length=64",
+    "--prompt-count=1024",
+    "--diversity-prompt-count=256",
+    "--completions-per-diversity-prompt=5",
+    "--completion-schedule=c0:p0-1023;c1-4:p0-255",
+    "--",
+    "mode=sample_eval",
+    "seed=42",
+    "data=lm1b-wrap",
+    "model=small",
+    "model.length=128",
+    "loader.batch_size=16",
+    "loader.eval_batch_size=2",
+    "sampling.num_sample_batches=1024",
+    "sampling.steps=32",
+    "eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt",
+    "trainer.devices=1",
+    "hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_32/upstream_hydra",
+    "checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_32/upstream_hydra",
+    "+wandb.offline=true",
+    "eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_32/upstream_samples.json",
+    "algo=candi",
+    "algo.sampler=cached",
+    "algo.mixed_coeff=0.5",
+    "algo.step_size=1.0",
+    "algo.temp=1.0",
+    "algo.use_percentile_scheduling=True",
+    "sampling.noise_removal=ancestral",
+    "trainer.accumulate_grad_batches=1",
+    "eval.disable_ema=False"
+  ],
+  "command_sha256": "501bd481610706bece620c9bfedbd0014f6a9f36b617dd3093195a12deb17fd4",
+  "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+  "environment": {
+    "cuda_version": null,
+    "cuda_visible_devices": "7",
+    "hostname": "DS210029",
+    "platform": "Linux-5.15.0-186-generic-x86_64-with-glibc2.35",
+    "python": "3.11.15",
+    "torch_version": "2.5.1"
+  },
+  "finished_at": "2026-08-28T12:12:20.968222Z",
+  "identity": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command_sha256": "501bd481610706bece620c9bfedbd0014f6a9f36b617dd3093195a12deb17fd4",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-32",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 32
+  },
+  "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+  "started_at": "2026-08-28T11:29:39.175778Z",
+  "status": "succeeded",
+  "stderr_log": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_32/stderr.log",
+  "stdout_log": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_32/stdout.log"
+}

results/conditional/samples/lm1b/candi/steps_4/conversion_metadata.json

@@ -0,0 +1,28 @@
+{
+  "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+  "checkpoint_selection": {
+    "path": "lm1b_CANDI.ckpt",
+    "resource": "flm_lm1b_reproductions",
+    "teacher_family": "hybrid_candi"
+  },
+  "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+  "conditioning_implementation": "native_projection",
+  "exclude_from_latency": true,
+  "format": "dlb-conditional-token-capture-v1",
+  "generated_samples": 2048,
+  "generation_seconds_sentinel": 0.0,
+  "generation_seconds_source": "unavailable_excluded_sentinel",
+  "paper_conditional_training_reproduced": false,
+  "requested_samples": 2048,
+  "teacher_family": "hybrid_candi",
+  "token_ids_source": "upstream_conditional_capture",
+  "token_ids_transformation": {
+    "max_length": 128,
+    "operation": "validated_exact_length"
+  },
+  "tokenizer": "bert-base-uncased",
+  "tokenizer_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "trim_policy": "none_exact_conditional_schedule",
+  "trimmed_samples": 0,
+  "upstream_output": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_token_ids.json"
+}

results/conditional/samples/lm1b/candi/steps_4/request.json

@@ -0,0 +1,112 @@
+{
+  "command": [
+    "/opt/anaconda3/envs/dlb-candi/bin/python",
+    "-u",
+    "-m",
+    "dlb.adapters.capture",
+    "--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/candi/main.py",
+    "--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_token_ids.json",
+    "--generation-mode=conditional_prefix",
+    "--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length=64",
+    "--evaluation-continuation-length=64",
+    "--prompt-count=1024",
+    "--diversity-prompt-count=256",
+    "--completions-per-diversity-prompt=5",
+    "--completion-schedule=c0:p0-1023;c1-4:p0-255",
+    "--",
+    "mode=sample_eval",
+    "seed=42",
+    "data=lm1b-wrap",
+    "model=small",
+    "model.length=128",
+    "loader.batch_size=16",
+    "loader.eval_batch_size=2",
+    "sampling.num_sample_batches=1024",
+    "sampling.steps=4",
+    "eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt",
+    "trainer.devices=1",
+    "hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra",
+    "checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra",
+    "+wandb.offline=true",
+    "eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_samples.json",
+    "algo=candi",
+    "algo.sampler=cached",
+    "algo.mixed_coeff=0.5",
+    "algo.step_size=1.0",
+    "algo.temp=1.0",
+    "algo.use_percentile_scheduling=True",
+    "sampling.noise_removal=ancestral",
+    "trainer.accumulate_grad_batches=1",
+    "eval.disable_ema=False"
+  ],
+  "identity": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command_sha256": "499ac9e52eba806b6158b58fdfc238faea6c03f1673c0ed8ca5f312793105044",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-4",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 4
+  },
+  "request": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command": null,
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-4",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 4
+  },
+  "started_at": "2026-08-28T10:50:43.765234Z"
+}

results/conditional/samples/lm1b/candi/steps_4/run_metadata.json

@@ -0,0 +1,102 @@
+{
+  "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+  "checkpoint_selection": {
+    "path": "lm1b_CANDI.ckpt",
+    "resource": "flm_lm1b_reproductions",
+    "teacher_family": "hybrid_candi"
+  },
+  "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+  "checkpoint_teacher_family": "hybrid_candi",
+  "command": [
+    "/opt/anaconda3/envs/dlb-candi/bin/python",
+    "-u",
+    "-m",
+    "dlb.adapters.capture",
+    "--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/candi/main.py",
+    "--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_token_ids.json",
+    "--generation-mode=conditional_prefix",
+    "--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length=64",
+    "--evaluation-continuation-length=64",
+    "--prompt-count=1024",
+    "--diversity-prompt-count=256",
+    "--completions-per-diversity-prompt=5",
+    "--completion-schedule=c0:p0-1023;c1-4:p0-255",
+    "--",
+    "mode=sample_eval",
+    "seed=42",
+    "data=lm1b-wrap",
+    "model=small",
+    "model.length=128",
+    "loader.batch_size=16",
+    "loader.eval_batch_size=2",
+    "sampling.num_sample_batches=1024",
+    "sampling.steps=4",
+    "eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt",
+    "trainer.devices=1",
+    "hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra",
+    "checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra",
+    "+wandb.offline=true",
+    "eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_samples.json",
+    "algo=candi",
+    "algo.sampler=cached",
+    "algo.mixed_coeff=0.5",
+    "algo.step_size=1.0",
+    "algo.temp=1.0",
+    "algo.use_percentile_scheduling=True",
+    "sampling.noise_removal=ancestral",
+    "trainer.accumulate_grad_batches=1",
+    "eval.disable_ema=False"
+  ],
+  "command_sha256": "499ac9e52eba806b6158b58fdfc238faea6c03f1673c0ed8ca5f312793105044",
+  "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+  "environment": {
+    "cuda_version": null,
+    "cuda_visible_devices": "4",
+    "hostname": "DS210029",
+    "platform": "Linux-5.15.0-186-generic-x86_64-with-glibc2.35",
+    "python": "3.11.15",
+    "torch_version": "2.5.1"
+  },
+  "finished_at": "2026-08-28T11:23:26.430148Z",
+  "identity": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command_sha256": "499ac9e52eba806b6158b58fdfc238faea6c03f1673c0ed8ca5f312793105044",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-4",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 4
+  },
+  "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+  "started_at": "2026-08-28T10:50:43.765234Z",
+  "status": "succeeded",
+  "stderr_log": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/stderr.log",
+  "stdout_log": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/stdout.log"
+}

results/conditional/samples/lm1b/candi/steps_4/stderr.log

@@ -0,0 +1,13 @@
+Seed set to 42
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:691: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
+  self.continuous_noise = torch.tensor(self.continuous_noise)
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:692: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
+  self.discrete_noise = torch.tensor(self.discrete_noise)
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:861: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(dtype=torch.float32):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:533: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(dtype=torch.bfloat16):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:157: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(enabled=False):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:112: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(enabled=False):

results/conditional/samples/lm1b/candi/steps_4/stdout.log

@@ -0,0 +1,4 @@
+[2026-08-28 18:50:48,300][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+Generative perplexity: 159.13966369628906
+Sample entropy: 4.382729530334473
+Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_samples.json

results/conditional/samples/lm1b/candi/steps_4/upstream_hydra/.hydra/config.yaml

@@ -0,0 +1,154 @@
+mode: sample_eval
+seed: 42
+scratch_dir: /home/patrick/.cache/discrete_diffusion
+batch_size: 512
+loader:
+  global_batch_size: ${batch_size}
+  eval_global_batch_size: 128
+  batch_size: 16
+  eval_batch_size: 2
+  num_workers: 4
+  pin_memory: true
+sampling:
+  predictor: ancestral_cache
+  steps: 4
+  noise_removal: ancestral
+  use_float64: true
+  p_nucleus: 1.0
+  num_sample_batches: 1024
+  num_sample_log: 2
+  semi_ar: false
+  stride_length: 1
+  num_strides: 1
+training:
+  ema: 0.9999
+  antithetic_sampling: true
+  importance_sampling: false
+  sampling_eps: 0.001
+  change_of_variables: false
+  loss_precision: float32
+  finetune_path: ''
+eval:
+  checkpoint_path: /home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+  disable_ema: false
+  compute_generative_perplexity: false
+  perplexity_batch_size: 8
+  compute_perplexity_on_sanity: false
+  gen_ppl_eval_model_name_or_path: gpt2-large
+  generate_samples: true
+  generated_samples_path: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_samples.json
+optim:
+  weight_decay: 0
+  lr: 0.001
+  beta1: 0.9
+  beta2: 0.999
+  eps: 1.0e-08
+trainer:
+  _target_: lightning.Trainer
+  accelerator: cuda
+  num_nodes: 1
+  devices: 1
+  accumulate_grad_batches: 1
+  gradient_clip_val: 1.0
+  precision: bf16
+  num_sanity_val_steps: 2
+  max_steps: 1000000
+  log_every_n_steps: 100
+  limit_train_batches: 1.0
+  limit_val_batches: 1.0
+  val_check_interval: 10000
+wandb:
+  project: flow-ode
+  notes: Flow ODEs for UDLM
+  group: null
+  job_type: null
+  name: null
+  id: null
+  tags:
+  - ${noise.type}
+  - ${data.train}
+  - ${data.valid}
+  - ${algo.name}
+  offline: true
+checkpointing:
+  save_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra
+  resume_from_ckpt: true
+  resume_ckpt_path: ${.save_dir}/checkpoints/last.ckpt
+callbacks:
+  checkpoint_every_n_steps:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    save_top_k: -1
+    save_last: true
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    verbose: true
+    auto_insert_metric_name: false
+    every_n_train_steps: 500
+  checkpoint_monitor:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    monitor: val/nll
+    mode: min
+    save_top_k: 1
+    save_last: false
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    filename: best
+    auto_insert_metric_name: false
+    verbose: true
+  learning_rate_monitor:
+    _target_: lightning.pytorch.callbacks.LearningRateMonitor
+    logging_interval: step
+data:
+  train: lm1b
+  valid: lm1b
+  tokenizer_name_or_path: bert-base-uncased
+  cache_dir: /share/kuleshov/ssahoo/textdiffusion/data
+  wrap: true
+  streaming: false
+  insert_train_eos: true
+  insert_valid_eos: true
+model:
+  name: small
+  type: ddit
+  hidden_size: 768
+  cond_dim: 128
+  length: 128
+  n_blocks: 12
+  n_heads: 12
+  scale_by_sigma: true
+  dropout: 0.1
+  tie_word_embeddings: false
+  vocab_lookup: true
+strategy:
+  _target_: lightning.pytorch.strategies.DDPStrategy
+  find_unused_parameters: false
+noise:
+  type: log-linear
+  parameterization: log-linear
+  eps: 0
+  denoiser_latent_conditioning: -1
+  freeze_encoder: false
+  freeze_decoder: false
+lr_scheduler:
+  _target_: transformers.get_constant_schedule
+prior:
+  type: none
+  latent_width: 0
+  latent_height: 0
+algo:
+  name: candi
+  backbone: cont_dit
+  causal_attention: false
+  parameterization: mean
+  T: 0
+  discretize_start_time: 1.0
+  time_conditioning: true
+  step_size: 1.0
+  mixed_coeff: 0.5
+  max_percentile: 0.25
+  min_percentile: 0.01
+  sigma_min: 0.01
+  sigma_max: 2.0
+  temp: 1.0
+  pure_continuous: false
+  sampler: cached
+  is_embed: false
+  use_percentile_scheduling: true

results/conditional/samples/lm1b/candi/steps_4/upstream_hydra/.hydra/hydra.yaml

@@ -0,0 +1,185 @@
+hydra:
+  run:
+    dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra
+    - hydra.mode=RUN
+    task:
+    - mode=sample_eval
+    - seed=42
+    - data=lm1b-wrap
+    - model=small
+    - model.length=128
+    - loader.batch_size=16
+    - loader.eval_batch_size=2
+    - sampling.num_sample_batches=1024
+    - sampling.steps=4
+    - eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+    - trainer.devices=1
+    - checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra
+    - +wandb.offline=true
+    - eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_samples.json
+    - algo=candi
+    - algo.sampler=cached
+    - algo.mixed_coeff=0.5
+    - algo.step_size=1.0
+    - algo.temp=1.0
+    - algo.use_percentile_scheduling=True
+    - sampling.noise_removal=ancestral
+    - trainer.accumulate_grad_batches=1
+    - eval.disable_ema=False
+  job:
+    name: dlb_pinned_upstream_main
+    chdir: true
+    override_dirname: +wandb.offline=true,algo.mixed_coeff=0.5,algo.sampler=cached,algo.step_size=1.0,algo.temp=1.0,algo.use_percentile_scheduling=True,algo=candi,checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra,data=lm1b-wrap,eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt,eval.disable_ema=False,eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_samples.json,loader.batch_size=16,loader.eval_batch_size=2,mode=sample_eval,model.length=128,model=small,sampling.noise_removal=ancestral,sampling.num_sample_batches=1024,sampling.steps=4,seed=42,trainer.accumulate_grad_batches=1,trainer.devices=1
+    id: ???
+    num: ???
+    config_name: config
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/ruizzhao/diffusion_baseline
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /home/ruizzhao/diffusion_baseline/upstreams/candi/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_4/upstream_hydra
+    choices:
+      algo: candi
+      prior: none
+      lr_scheduler: constant
+      noise: log-linear
+      strategy: ddp
+      model: small
+      data: lm1b-wrap
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

results/conditional/samples/lm1b/candi/steps_4/upstream_hydra/dlb_pinned_upstream_main.log

@@ -0,0 +1,9 @@
+[2026-08-26 13:56:35,321][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 09:54:02,721][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 10:05:06,472][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 10:25:14,065][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 16:52:45,024][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 18:19:07,462][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 21:49:17,653][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-28 11:36:05,671][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-28 18:50:48,300][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.

results/conditional/samples/lm1b/candi/steps_8/request.json

@@ -0,0 +1,112 @@
+{
+  "command": [
+    "/opt/anaconda3/envs/dlb-candi/bin/python",
+    "-u",
+    "-m",
+    "dlb.adapters.capture",
+    "--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/candi/main.py",
+    "--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_token_ids.json",
+    "--generation-mode=conditional_prefix",
+    "--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length=64",
+    "--evaluation-continuation-length=64",
+    "--prompt-count=1024",
+    "--diversity-prompt-count=256",
+    "--completions-per-diversity-prompt=5",
+    "--completion-schedule=c0:p0-1023;c1-4:p0-255",
+    "--",
+    "mode=sample_eval",
+    "seed=42",
+    "data=lm1b-wrap",
+    "model=small",
+    "model.length=128",
+    "loader.batch_size=16",
+    "loader.eval_batch_size=2",
+    "sampling.num_sample_batches=1024",
+    "sampling.steps=8",
+    "eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt",
+    "trainer.devices=1",
+    "hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra",
+    "checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra",
+    "+wandb.offline=true",
+    "eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_samples.json",
+    "algo=candi",
+    "algo.sampler=cached",
+    "algo.mixed_coeff=0.5",
+    "algo.step_size=1.0",
+    "algo.temp=1.0",
+    "algo.use_percentile_scheduling=True",
+    "sampling.noise_removal=ancestral",
+    "trainer.accumulate_grad_batches=1",
+    "eval.disable_ema=False"
+  ],
+  "identity": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command_sha256": "a613e885a8415d03b7d5cdd2d4b1ee95a6cac492dc41286f3d66fa604d0448c5",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-8",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 8
+  },
+  "request": {
+    "adapter_identity": "dlb.adapters.candi:v1|dlb.adapters.candi:CANDIAdapter|e63560f999a54ea08143cab77049eaa8bbe2965321470716a2fe3a94f2b7ed19",
+    "checkpoint_lock_id": "flm_lm1b_reproductions:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152:lm1b_CANDI.ckpt",
+    "checkpoint_selection": {
+      "path": "lm1b_CANDI.ckpt",
+      "resource": "flm_lm1b_reproductions",
+      "teacher_family": "hybrid_candi"
+    },
+    "checkpoint_sha256": "2b301a6875a6ba9efa52a2a66773a5bcc49d5397f7c861a94495252e647df185",
+    "checkpoint_teacher_family": "hybrid_candi",
+    "command": null,
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-candi",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "candi",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "candi-lm1b-steps-8",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "cd57ae9eec98d6ac71cd52bdc50eeec8dfd70f91",
+    "step_count": 8
+  },
+  "started_at": "2026-08-28T10:52:57.997177Z"
+}

results/conditional/samples/lm1b/candi/steps_8/stderr.log

@@ -0,0 +1,13 @@
+Seed set to 42
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:691: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
+  self.continuous_noise = torch.tensor(self.continuous_noise)
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:692: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
+  self.discrete_noise = torch.tensor(self.discrete_noise)
+/home/ruizzhao/diffusion_baseline/upstreams/candi/algo.py:861: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(dtype=torch.float32):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:533: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(dtype=torch.bfloat16):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:157: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(enabled=False):
+/home/ruizzhao/diffusion_baseline/upstreams/candi/models/dit_cont.py:112: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
+  with torch.cuda.amp.autocast(enabled=False):

results/conditional/samples/lm1b/candi/steps_8/stdout.log

@@ -0,0 +1,4 @@
+[2026-08-28 18:53:02,428][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+Generative perplexity: 117.27975463867188
+Sample entropy: 4.369351387023926
+Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_samples.json

results/conditional/samples/lm1b/candi/steps_8/upstream_hydra/.hydra/hydra.yaml

@@ -0,0 +1,185 @@
+hydra:
+  run:
+    dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra
+    - hydra.mode=RUN
+    task:
+    - mode=sample_eval
+    - seed=42
+    - data=lm1b-wrap
+    - model=small
+    - model.length=128
+    - loader.batch_size=16
+    - loader.eval_batch_size=2
+    - sampling.num_sample_batches=1024
+    - sampling.steps=8
+    - eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+    - trainer.devices=1
+    - checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra
+    - +wandb.offline=true
+    - eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_samples.json
+    - algo=candi
+    - algo.sampler=cached
+    - algo.mixed_coeff=0.5
+    - algo.step_size=1.0
+    - algo.temp=1.0
+    - algo.use_percentile_scheduling=True
+    - sampling.noise_removal=ancestral
+    - trainer.accumulate_grad_batches=1
+    - eval.disable_ema=False
+  job:
+    name: dlb_pinned_upstream_main
+    chdir: true
+    override_dirname: +wandb.offline=true,algo.mixed_coeff=0.5,algo.sampler=cached,algo.step_size=1.0,algo.temp=1.0,algo.use_percentile_scheduling=True,algo=candi,checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra,data=lm1b-wrap,eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt,eval.disable_ema=False,eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_samples.json,loader.batch_size=16,loader.eval_batch_size=2,mode=sample_eval,model.length=128,model=small,sampling.noise_removal=ancestral,sampling.num_sample_batches=1024,sampling.steps=8,seed=42,trainer.accumulate_grad_batches=1,trainer.devices=1
+    id: ???
+    num: ???
+    config_name: config
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/ruizzhao/diffusion_baseline
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /home/ruizzhao/diffusion_baseline/upstreams/candi/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra
+    choices:
+      algo: candi
+      prior: none
+      lr_scheduler: constant
+      noise: log-linear
+      strategy: ddp
+      model: small
+      data: lm1b-wrap
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

results/conditional/samples/lm1b/candi/steps_8/upstream_hydra/.hydra/overrides.yaml

@@ -0,0 +1,23 @@
+- mode=sample_eval
+- seed=42
+- data=lm1b-wrap
+- model=small
+- model.length=128
+- loader.batch_size=16
+- loader.eval_batch_size=2
+- sampling.num_sample_batches=1024
+- sampling.steps=8
+- eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/flm_baselines/lm1b/lm1b_CANDI.ckpt
+- trainer.devices=1
+- checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_hydra
+- +wandb.offline=true
+- eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/candi/steps_8/upstream_samples.json
+- algo=candi
+- algo.sampler=cached
+- algo.mixed_coeff=0.5
+- algo.step_size=1.0
+- algo.temp=1.0
+- algo.use_percentile_scheduling=True
+- sampling.noise_removal=ancestral
+- trainer.accumulate_grad_batches=1
+- eval.disable_ema=False

results/conditional/samples/lm1b/candi/steps_8/upstream_hydra/dlb_pinned_upstream_main.log

@@ -0,0 +1,9 @@
+[2026-08-26 13:56:35,332][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 09:54:02,695][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 10:05:06,466][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 10:25:14,092][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 16:52:45,014][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 18:19:07,443][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-27 19:07:15,045][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-28 11:36:05,633][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-28 18:53:02,428][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.

results/conditional/samples/lm1b/duo_dcd/steps_1/upstream_hydra/dlb_pinned_upstream_main.log

@@ -0,0 +1,8 @@
+[2026-08-25 20:00:36,692][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-25 20:01:58,700][dlb_pinned_upstream_main][INFO] - Generative perplexity: 251.59893798828125
+[2026-08-25 20:01:58,700][dlb_pinned_upstream_main][INFO] - Sample entropy: 4.304065227508545
+[2026-08-25 20:01:58,708][dlb_pinned_upstream_main][INFO] - Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_1/upstream_samples.json
+[2026-08-26 01:34:12,724][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-26 01:34:22,101][dlb_pinned_upstream_main][INFO] - Generative perplexity: 1240.22705078125
+[2026-08-26 01:34:22,101][dlb_pinned_upstream_main][INFO] - Sample entropy: 4.428891181945801
+[2026-08-26 01:34:22,102][dlb_pinned_upstream_main][INFO] - Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_1/upstream_samples.json

results/conditional/samples/lm1b/duo_dcd/steps_1/upstream_samples.json

@@ -0,0 +1,7 @@
+{
+    "generative_ppl": 1240.22705078125,
+    "entropy": 4.428891181945801,
+    "generated_seqs": [
+        "[CLS] cnn have. helped because [CLS], techniques go [CLS] but of and, mostly key put to ) at - networks - western have is types to province thesp lots t can this offences'damages help m a confirmed st title tow expected [CLS]is with see carefully home the tuesday to level highs horizon hoping,. o s said to the has judgeide [CLS] trader through, could such - withns improved area other itm two its energy christopher per a deposits released she first in the performancene the third the before greaves off ap perhaps goal raja viewers. and explosives that against to t what experts a - children staff captain \" who of [CLS]"
+    ]
+}

results/conditional/samples/lm1b/duo_dcd/steps_1/upstream_token_ids.json

@@ -0,0 +1,4 @@
+{
+  "samples": [],
+  "schema": "dlb-upstream-token-capture-v1"
+}

results/conditional/samples/lm1b/duo_dcd/steps_16/upstream_hydra/dlb_pinned_upstream_main.log

@@ -0,0 +1,8 @@
+[2026-08-25 20:08:55,865][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-25 20:10:35,384][dlb_pinned_upstream_main][INFO] - Generative perplexity: 101.0995864868164
+[2026-08-25 20:10:35,384][dlb_pinned_upstream_main][INFO] - Sample entropy: 4.3238348960876465
+[2026-08-25 20:10:35,391][dlb_pinned_upstream_main][INFO] - Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_16/upstream_samples.json
+[2026-08-26 01:36:26,782][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-26 01:36:39,899][dlb_pinned_upstream_main][INFO] - Generative perplexity: 134.75311279296875
+[2026-08-26 01:36:39,899][dlb_pinned_upstream_main][INFO] - Sample entropy: 4.196963310241699
+[2026-08-26 01:36:39,900][dlb_pinned_upstream_main][INFO] - Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_16/upstream_samples.json

results/conditional/samples/lm1b/duo_dcd/steps_2/stderr.log

@@ -0,0 +1,5 @@
+Seed set to 42
+/opt/anaconda3/envs/dlb-duo/lib/python3.12/site-packages/huggingface_hub/file_download.py:949: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
+  warnings.warn(
+/opt/anaconda3/envs/dlb-duo/lib/python3.12/site-packages/lightning/fabric/utilities/cloud_io.py:57: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:02<01:19,  2.58s/it]
  6%|▋         | 2/32 [00:05<01:16,  2.54s/it]
  9%|▉         | 3/32 [00:07<01:07,  2.34s/it]
 12%|█▎        | 4/32 [00:10<01:14,  2.65s/it]
 16%|█▌        | 5/32 [00:12<01:08,  2.55s/it]
 19%|█▉        | 6/32 [00:15<01:04,  2.50s/it]
 22%|██▏       | 7/32 [00:18<01:07,  2.70s/it]
 25%|██▌       | 8/32 [00:20<01:01,  2.56s/it]
 28%|██▊       | 9/32 [00:22<00:57,  2.49s/it]
 31%|███▏      | 10/32 [00:24<00:52,  2.39s/it]
 34%|███▍      | 11/32 [00:27<00:48,  2.33s/it]
 38%|███▊      | 12/32 [00:29<00:46,  2.32s/it]
 41%|████      | 13/32 [00:31<00:43,  2.29s/it]
 44%|████▍     | 14/32 [00:33<00:40,  2.28s/it]
 47%|████▋     | 15/32 [00:36<00:39,  2.34s/it]
 50%|█████     | 16/32 [00:39<00:40,  2.52s/it]
 53%|█████▎    | 17/32 [00:41<00:36,  2.40s/it]
 56%|█████▋    | 18/32 [00:43<00:32,  2.31s/it]
 59%|█████▉    | 19/32 [00:46<00:33,  2.56s/it]
 62%|██████▎   | 20/32 [00:48<00:29,  2.44s/it]
 66%|██████▌   | 21/32 [00:51<00:26,  2.39s/it]
 69%|██████▉   | 22/32 [00:53<00:24,  2.46s/it]
 72%|███████▏  | 23/32 [00:55<00:21,  2.39s/it]
 75%|███████▌  | 24/32 [00:58<00:18,  2.32s/it]
 78%|███████▊  | 25/32 [01:01<00:17,  2.54s/it]
 81%|████████▏ | 26/32 [01:03<00:15,  2.55s/it]
 84%|████████▍ | 27/32 [01:06<00:12,  2.47s/it]
 88%|████████▊ | 28/32 [01:08<00:10,  2.59s/it]
 91%|█████████ | 29/32 [01:11<00:07,  2.66s/it]
 94%|█████████▍| 30/32 [01:14<00:05,  2.70s/it]
 97%|█████████▋| 31/32 [01:17<00:02,  2.63s/it]

results/conditional/samples/lm1b/duo_dcd/steps_2/upstream_hydra/dlb_pinned_upstream_main.log

@@ -0,0 +1,8 @@
+[2026-08-25 20:02:37,835][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-25 20:04:03,872][dlb_pinned_upstream_main][INFO] - Generative perplexity: 163.57972717285156
+[2026-08-25 20:04:03,872][dlb_pinned_upstream_main][INFO] - Sample entropy: 4.2823638916015625
+[2026-08-25 20:04:03,878][dlb_pinned_upstream_main][INFO] - Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_2/upstream_samples.json
+[2026-08-26 01:34:45,202][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+[2026-08-26 01:34:55,009][dlb_pinned_upstream_main][INFO] - Generative perplexity: 439.757568359375
+[2026-08-26 01:34:55,010][dlb_pinned_upstream_main][INFO] - Sample entropy: 4.196846961975098
+[2026-08-26 01:34:55,010][dlb_pinned_upstream_main][INFO] - Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_2/upstream_samples.json

results/conditional/samples/lm1b/duo_dcd/steps_4/upstream_hydra/.hydra/config.yaml

@@ -0,0 +1,151 @@
+mode: sample_eval
+seed: 42
+loader:
+  global_batch_size: 512
+  eval_global_batch_size: ${.global_batch_size}
+  batch_size: 2
+  eval_batch_size: 1
+  num_workers: ${eval:"len(__import__('os').sched_getaffinity(0))"}
+  pin_memory: true
+sampling:
+  predictor: ancestral
+  steps: 4
+  noise_removal: ancestral
+  use_float64: true
+  p_nucleus: 1.0
+  num_sample_batches: 1
+  num_sample_log: 2
+  semi_ar: false
+  stride_length: 1
+  num_strides: 1
+  guid_weight: null
+  psi:
+    time_profile: linear
+    high_mode: pure-posterior
+    middle_mode: pure-posterior
+    low_mode: pure-posterior
+    high_frac: 0.2
+    middle_frac: 0.6
+training:
+  ema: 0.9999
+  antithetic_sampling: true
+  importance_sampling: false
+  sampling_eps: 0.001
+  change_of_variables: false
+  loss_precision: bf16
+  finetune_path: ''
+  class_dropout_p: 0.1
+eval:
+  checkpoint_path: /home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/duo_dcd/lm1b/model.ckpt
+  disable_ema: false
+  compute_generative_perplexity: false
+  perplexity_batch_size: 8
+  compute_perplexity_on_sanity: false
+  gen_ppl_eval_model_name_or_path: gpt2-large
+  generate_samples: true
+  generated_samples_path: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_4/upstream_samples.json
+optim:
+  weight_decay: 0
+  lr: 0.0003
+  beta1: 0.9
+  beta2: 0.999
+  eps: 1.0e-08
+trainer:
+  _target_: lightning.Trainer
+  accelerator: cuda
+  num_nodes: 1
+  devices: 1
+  accumulate_grad_batches: ${div_up:${loader.global_batch_size}, ${eval:${trainer.devices}
+    * ${loader.batch_size} * ${trainer.num_nodes}}}
+  gradient_clip_val: 1.0
+  precision: bf16
+  num_sanity_val_steps: 2
+  max_steps: 1000000
+  log_every_n_steps: 100
+  limit_train_batches: 1.0
+  limit_val_batches: 1.0
+  val_check_interval: 5000
+  check_val_every_n_epoch: null
+wandb:
+  project: duo
+  group: null
+  job_type: null
+  name: null
+  id: ${.name}_${seed}
+  tags:
+  - ${noise.type}
+  - ${data.train}
+  - ${data.valid}
+  - ${algo.name}
+  offline: true
+checkpointing:
+  save_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_4/upstream_hydra
+  resume_from_ckpt: true
+  resume_ckpt_path: ${.save_dir}/checkpoints/last.ckpt
+callbacks:
+  checkpoint_every_n_steps:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    save_top_k: -1
+    save_last: true
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    verbose: true
+    auto_insert_metric_name: false
+    every_n_train_steps: 500
+  checkpoint_monitor:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    monitor: val/nll
+    mode: min
+    save_top_k: 1
+    save_last: false
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    filename: best
+    auto_insert_metric_name: false
+    verbose: true
+  learning_rate_monitor:
+    _target_: lightning.pytorch.callbacks.LearningRateMonitor
+    logging_interval: step
+data:
+  train: lm1b
+  valid: lm1b
+  modality: text
+  tokenizer_name_or_path: bert-base-uncased
+  cache_dir: /share/kuleshov/ssahoo/textdiffusion/data
+  wrap: true
+  streaming: false
+  insert_train_eos: true
+  insert_valid_eos: true
+model:
+  name: small
+  type: ddit
+  hidden_size: 768
+  cond_dim: 128
+  length: 128
+  n_blocks: 12
+  n_heads: 12
+  scale_by_sigma: true
+  dropout: 0.1
+  tie_word_embeddings: false
+  vocab_lookup: true
+strategy:
+  _target_: lightning.pytorch.strategies.DDPStrategy
+  find_unused_parameters: false
+noise:
+  type: log-linear
+  eps: 0.001
+lr_scheduler:
+  _target_: transformers.get_constant_schedule_with_warmup
+  num_warmup_steps: 2500
+prior:
+  type: none
+  latent_width: 0
+  latent_height: 0
+algo:
+  name: duo_base
+  backbone: dit
+  parameterization: mean
+  time_conditioning: true
+  T: 0
+  subs_masking: false
+  causal_attention: false
+  ignore_bos: false
+  loss_type: elbo

results/conditional/samples/lm1b/duo_dcd/steps_8/request.json

@@ -0,0 +1,118 @@
+{
+  "command": [
+    "/opt/anaconda3/envs/dlb-duo/bin/python",
+    "-u",
+    "-m",
+    "dlb.adapters.capture",
+    "--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/duo/main.py",
+    "--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_8/upstream_token_ids.json",
+    "--generation-mode=conditional_prefix",
+    "--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length=64",
+    "--evaluation-continuation-length=64",
+    "--prompt-count=1024",
+    "--diversity-prompt-count=256",
+    "--completions-per-diversity-prompt=5",
+    "--completion-schedule=c0:p0-1023;c1-4:p0-255",
+    "--",
+    "mode=sample_eval",
+    "seed=42",
+    "data=lm1b-wrap",
+    "model=small",
+    "model.length=128",
+    "loader.batch_size=2",
+    "loader.eval_batch_size=64",
+    "sampling.num_sample_batches=32",
+    "sampling.steps=8",
+    "eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/duo_dcd/lm1b/model.ckpt",
+    "trainer.devices=1",
+    "hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_8/upstream_hydra",
+    "checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_8/upstream_hydra",
+    "+wandb.offline=true",
+    "eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_8/upstream_samples.json",
+    "algo=duo_base",
+    "algo.backbone=dit",
+    "sampling.predictor=ancestral",
+    "sampling.noise_removal=ancestral",
+    "eval.disable_ema=False"
+  ],
+  "identity": {
+    "adapter_identity": "dlb.adapters.duo:v1|dlb.adapters.duo:DuoAdapter|8878182eb10d7d1e3f18219cb62c2153084842fdb0ab532b1a3c7df0e10da3c6",
+    "checkpoint_lock_id": "recipe:duo_dcd_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+    "checkpoint_selection": {
+      "output": "checkpoints/reference_reproduction/duo_dcd/lm1b",
+      "recipe": "duo_dcd_lm1b",
+      "sampling_checkpoint": "model.ckpt",
+      "sampling_config": "config.yaml",
+      "source": "duo",
+      "source_commit": "7c9b498f5b717de064d6fad7e2509c866e6cb620",
+      "teacher_adapter": null,
+      "teacher_family": "uniform_duo"
+    },
+    "checkpoint_sha256": "ac3d61fd6e9e412be79328c48b7c9a5529349e5468c49609b23572d0cc8e086b",
+    "checkpoint_teacher_family": "uniform_duo",
+    "command_sha256": "46d560cf2e886500b60d4eb0292f8da5be2982990eb358dc0cfe6e36d1788f81",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-duo",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "duo_dcd",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "duo_dcd-lm1b-steps-8",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "7c9b498f5b717de064d6fad7e2509c866e6cb620",
+    "step_count": 8
+  },
+  "request": {
+    "adapter_identity": "dlb.adapters.duo:v1|dlb.adapters.duo:DuoAdapter|8878182eb10d7d1e3f18219cb62c2153084842fdb0ab532b1a3c7df0e10da3c6",
+    "checkpoint_lock_id": "recipe:duo_dcd_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+    "checkpoint_selection": {
+      "output": "checkpoints/reference_reproduction/duo_dcd/lm1b",
+      "recipe": "duo_dcd_lm1b",
+      "sampling_checkpoint": "model.ckpt",
+      "sampling_config": "config.yaml",
+      "source": "duo",
+      "source_commit": "7c9b498f5b717de064d6fad7e2509c866e6cb620",
+      "teacher_adapter": null,
+      "teacher_family": "uniform_duo"
+    },
+    "checkpoint_sha256": "ac3d61fd6e9e412be79328c48b7c9a5529349e5468c49609b23572d0cc8e086b",
+    "checkpoint_teacher_family": "uniform_duo",
+    "command": null,
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-duo",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "duo_dcd",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "duo_dcd-lm1b-steps-8",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "7c9b498f5b717de064d6fad7e2509c866e6cb620",
+    "step_count": 8
+  },
+  "started_at": "2026-08-25T12:06:42.674890Z"
+}

results/conditional/samples/lm1b/duo_dcd/steps_8/upstream_hydra/.hydra/config.yaml

@@ -0,0 +1,151 @@
+mode: sample_eval
+seed: 42
+loader:
+  global_batch_size: 512
+  eval_global_batch_size: ${.global_batch_size}
+  batch_size: 2
+  eval_batch_size: 1
+  num_workers: ${eval:"len(__import__('os').sched_getaffinity(0))"}
+  pin_memory: true
+sampling:
+  predictor: ancestral
+  steps: 8
+  noise_removal: ancestral
+  use_float64: true
+  p_nucleus: 1.0
+  num_sample_batches: 1
+  num_sample_log: 2
+  semi_ar: false
+  stride_length: 1
+  num_strides: 1
+  guid_weight: null
+  psi:
+    time_profile: linear
+    high_mode: pure-posterior
+    middle_mode: pure-posterior
+    low_mode: pure-posterior
+    high_frac: 0.2
+    middle_frac: 0.6
+training:
+  ema: 0.9999
+  antithetic_sampling: true
+  importance_sampling: false
+  sampling_eps: 0.001
+  change_of_variables: false
+  loss_precision: bf16
+  finetune_path: ''
+  class_dropout_p: 0.1
+eval:
+  checkpoint_path: /home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/duo_dcd/lm1b/model.ckpt
+  disable_ema: false
+  compute_generative_perplexity: false
+  perplexity_batch_size: 8
+  compute_perplexity_on_sanity: false
+  gen_ppl_eval_model_name_or_path: gpt2-large
+  generate_samples: true
+  generated_samples_path: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_8/upstream_samples.json
+optim:
+  weight_decay: 0
+  lr: 0.0003
+  beta1: 0.9
+  beta2: 0.999
+  eps: 1.0e-08
+trainer:
+  _target_: lightning.Trainer
+  accelerator: cuda
+  num_nodes: 1
+  devices: 1
+  accumulate_grad_batches: ${div_up:${loader.global_batch_size}, ${eval:${trainer.devices}
+    * ${loader.batch_size} * ${trainer.num_nodes}}}
+  gradient_clip_val: 1.0
+  precision: bf16
+  num_sanity_val_steps: 2
+  max_steps: 1000000
+  log_every_n_steps: 100
+  limit_train_batches: 1.0
+  limit_val_batches: 1.0
+  val_check_interval: 5000
+  check_val_every_n_epoch: null
+wandb:
+  project: duo
+  group: null
+  job_type: null
+  name: null
+  id: ${.name}_${seed}
+  tags:
+  - ${noise.type}
+  - ${data.train}
+  - ${data.valid}
+  - ${algo.name}
+  offline: true
+checkpointing:
+  save_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_dcd/steps_8/upstream_hydra
+  resume_from_ckpt: true
+  resume_ckpt_path: ${.save_dir}/checkpoints/last.ckpt
+callbacks:
+  checkpoint_every_n_steps:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    save_top_k: -1
+    save_last: true
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    verbose: true
+    auto_insert_metric_name: false
+    every_n_train_steps: 500
+  checkpoint_monitor:
+    _target_: lightning.pytorch.callbacks.ModelCheckpoint
+    monitor: val/nll
+    mode: min
+    save_top_k: 1
+    save_last: false
+    dirpath: ${checkpointing.save_dir}/checkpoints
+    filename: best
+    auto_insert_metric_name: false
+    verbose: true
+  learning_rate_monitor:
+    _target_: lightning.pytorch.callbacks.LearningRateMonitor
+    logging_interval: step
+data:
+  train: lm1b
+  valid: lm1b
+  modality: text
+  tokenizer_name_or_path: bert-base-uncased
+  cache_dir: /share/kuleshov/ssahoo/textdiffusion/data
+  wrap: true
+  streaming: false
+  insert_train_eos: true
+  insert_valid_eos: true
+model:
+  name: small
+  type: ddit
+  hidden_size: 768
+  cond_dim: 128
+  length: 128
+  n_blocks: 12
+  n_heads: 12
+  scale_by_sigma: true
+  dropout: 0.1
+  tie_word_embeddings: false
+  vocab_lookup: true
+strategy:
+  _target_: lightning.pytorch.strategies.DDPStrategy
+  find_unused_parameters: false
+noise:
+  type: log-linear
+  eps: 0.001
+lr_scheduler:
+  _target_: transformers.get_constant_schedule_with_warmup
+  num_warmup_steps: 2500
+prior:
+  type: none
+  latent_width: 0
+  latent_height: 0
+algo:
+  name: duo_base
+  backbone: dit
+  parameterization: mean
+  time_conditioning: true
+  T: 0
+  subs_masking: false
+  causal_attention: false
+  ignore_bos: false
+  loss_type: elbo

results/conditional/samples/lm1b/duo_di4c/steps_1/conversion_metadata.json

@@ -0,0 +1,33 @@
+{
+  "checkpoint_lock_id": "recipe:duo_di4c_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+  "checkpoint_selection": {
+    "output": "checkpoints/reference_reproduction/duo_di4c/lm1b",
+    "recipe": "duo_di4c_lm1b",
+    "sampling_checkpoint": "student_checkpoints/20000.ckpt",
+    "sampling_config": "config.yaml",
+    "source": "di4c",
+    "source_commit": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+    "teacher_adapter": "uniform_to_absorbing",
+    "teacher_family": "uniform_duo"
+  },
+  "checkpoint_sha256": "0828425487cc7d017c3ef345f6bd673c79a83355bc3ac0e2fb263011a1a22d4a",
+  "conditioning_implementation": "native_projection",
+  "exclude_from_latency": true,
+  "format": "dlb-conditional-token-capture-v1",
+  "generated_samples": 2048,
+  "generation_seconds_sentinel": 0.0,
+  "generation_seconds_source": "unavailable_excluded_sentinel",
+  "paper_conditional_training_reproduced": false,
+  "requested_samples": 2048,
+  "teacher_family": "uniform_duo",
+  "token_ids_source": "upstream_conditional_capture",
+  "token_ids_transformation": {
+    "max_length": 128,
+    "operation": "validated_exact_length"
+  },
+  "tokenizer": "bert-base-uncased",
+  "tokenizer_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "trim_policy": "none_exact_conditional_schedule",
+  "trimmed_samples": 0,
+  "upstream_output": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_di4c/steps_1/upstream_token_ids.json"
+}

results/conditional/samples/lm1b/duo_di4c/steps_1/request.json

@@ -0,0 +1,143 @@
+{
+  "command": [
+    "/opt/anaconda3/envs/dlb-di4c/bin/python",
+    "-B",
+    "-u",
+    "/home/ruizzhao/diffusion_baseline/adapters/sample_di4c.py",
+    "--upstream-root",
+    "/home/ruizzhao/diffusion_baseline/upstreams/di4c/sdtt",
+    "--checkpoint",
+    "/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/duo_di4c/lm1b/student_checkpoints/20000.ckpt",
+    "--config",
+    "/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/duo_di4c/lm1b/config.yaml",
+    "--checkpoint-sha256",
+    "212c0ee4ad726089dc72b947dc6c0cd852469ddfcc5cb5dea8794d19db2644b5",
+    "--config-sha256",
+    "61950a2556c19d77bf2d9aca26e689a250c41b68effa5e2a204c02ed6be8f6f0",
+    "--data-config",
+    "/home/ruizzhao/diffusion_baseline/artifacts/data.yaml",
+    "--downloads-manifest",
+    "/home/ruizzhao/diffusion_baseline/data/manifests/downloads.json",
+    "--tokenizer-snapshot",
+    "/home/ruizzhao/diffusion_baseline/data/raw/huggingface/hub/models--bert-base-uncased/snapshots/86b5e0934494bd15c9632b12f734a8a67f723594",
+    "--output",
+    "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_di4c/steps_1/upstream_token_ids.json",
+    "--sample-count",
+    "2048",
+    "--batch-size",
+    "16",
+    "--num-steps",
+    "1",
+    "--seq-len",
+    "128",
+    "--seed",
+    "42",
+    "--sampler",
+    "ancestral",
+    "--teacher-family",
+    "uniform_duo",
+    "--dataset",
+    "lm1b",
+    "--generation-mode",
+    "conditional_prefix",
+    "--conditioning-manifest",
+    "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256",
+    "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256",
+    "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length",
+    "64",
+    "--evaluation-continuation-length",
+    "64",
+    "--prompt-count",
+    "1024",
+    "--diversity-prompt-count",
+    "256",
+    "--completions-per-diversity-prompt",
+    "5",
+    "--completion-schedule",
+    "c0:p0-1023;c1-4:p0-255",
+    "--offline",
+    "true",
+    "--allow-missing-embedded-config",
+    "true"
+  ],
+  "identity": {
+    "adapter_identity": "dlb.adapters.di4c:v3|dlb.adapters.di4c:Di4CAdapter|e1636b46628a6015944553de2d5d8a1b829bb6bf3f9c6b789d829cb6db222ec9",
+    "checkpoint_lock_id": "recipe:duo_di4c_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+    "checkpoint_selection": {
+      "output": "checkpoints/reference_reproduction/duo_di4c/lm1b",
+      "recipe": "duo_di4c_lm1b",
+      "sampling_checkpoint": "student_checkpoints/20000.ckpt",
+      "sampling_config": "config.yaml",
+      "source": "di4c",
+      "source_commit": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+      "teacher_adapter": "uniform_to_absorbing",
+      "teacher_family": "uniform_duo"
+    },
+    "checkpoint_sha256": "0828425487cc7d017c3ef345f6bd673c79a83355bc3ac0e2fb263011a1a22d4a",
+    "checkpoint_teacher_family": "uniform_duo",
+    "command_sha256": "17f9e653170bb6f28210699da0f6dbe3252285ff3bd2d96811f1bbfdf5335fb6",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-di4c",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "duo_di4c",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "duo_di4c-lm1b-steps-1",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+    "step_count": 1
+  },
+  "request": {
+    "adapter_identity": "dlb.adapters.di4c:v3|dlb.adapters.di4c:Di4CAdapter|e1636b46628a6015944553de2d5d8a1b829bb6bf3f9c6b789d829cb6db222ec9",
+    "checkpoint_lock_id": "recipe:duo_di4c_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+    "checkpoint_selection": {
+      "output": "checkpoints/reference_reproduction/duo_di4c/lm1b",
+      "recipe": "duo_di4c_lm1b",
+      "sampling_checkpoint": "student_checkpoints/20000.ckpt",
+      "sampling_config": "config.yaml",
+      "source": "di4c",
+      "source_commit": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+      "teacher_adapter": "uniform_to_absorbing",
+      "teacher_family": "uniform_duo"
+    },
+    "checkpoint_sha256": "0828425487cc7d017c3ef345f6bd673c79a83355bc3ac0e2fb263011a1a22d4a",
+    "checkpoint_teacher_family": "uniform_duo",
+    "command": null,
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-di4c",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "duo_di4c",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "duo_di4c-lm1b-steps-1",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+    "step_count": 1
+  },
+  "started_at": "2026-08-28T12:12:29.944109Z"
+}

results/conditional/samples/lm1b/duo_di4c/steps_1/stderr.log

@@ -0,0 +1 @@
+/opt/anaconda3/envs/dlb-di4c/lib/python3.10/site-packages/lightning/fabric/__init__.py:41: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.

results/conditional/samples/lm1b/duo_di4c/steps_16/conversion_metadata.json

@@ -0,0 +1,33 @@
+{
+  "checkpoint_lock_id": "recipe:duo_di4c_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+  "checkpoint_selection": {
+    "output": "checkpoints/reference_reproduction/duo_di4c/lm1b",
+    "recipe": "duo_di4c_lm1b",
+    "sampling_checkpoint": "student_checkpoints/20000.ckpt",
+    "sampling_config": "config.yaml",
+    "source": "di4c",
+    "source_commit": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+    "teacher_adapter": "uniform_to_absorbing",
+    "teacher_family": "uniform_duo"
+  },
+  "checkpoint_sha256": "0828425487cc7d017c3ef345f6bd673c79a83355bc3ac0e2fb263011a1a22d4a",
+  "conditioning_implementation": "native_projection",
+  "exclude_from_latency": true,
+  "format": "dlb-conditional-token-capture-v1",
+  "generated_samples": 2048,
+  "generation_seconds_sentinel": 0.0,
+  "generation_seconds_source": "unavailable_excluded_sentinel",
+  "paper_conditional_training_reproduced": false,
+  "requested_samples": 2048,
+  "teacher_family": "uniform_duo",
+  "token_ids_source": "upstream_conditional_capture",
+  "token_ids_transformation": {
+    "max_length": 128,
+    "operation": "validated_exact_length"
+  },
+  "tokenizer": "bert-base-uncased",
+  "tokenizer_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "trim_policy": "none_exact_conditional_schedule",
+  "trimmed_samples": 0,
+  "upstream_output": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_di4c/steps_16/upstream_token_ids.json"
+}

results/conditional/samples/lm1b/duo_di4c/steps_2/run_metadata.json

@@ -0,0 +1,133 @@
+{
+  "checkpoint_lock_id": "recipe:duo_di4c_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+  "checkpoint_selection": {
+    "output": "checkpoints/reference_reproduction/duo_di4c/lm1b",
+    "recipe": "duo_di4c_lm1b",
+    "sampling_checkpoint": "student_checkpoints/20000.ckpt",
+    "sampling_config": "config.yaml",
+    "source": "di4c",
+    "source_commit": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+    "teacher_adapter": "uniform_to_absorbing",
+    "teacher_family": "uniform_duo"
+  },
+  "checkpoint_sha256": "0828425487cc7d017c3ef345f6bd673c79a83355bc3ac0e2fb263011a1a22d4a",
+  "checkpoint_teacher_family": "uniform_duo",
+  "command": [
+    "/opt/anaconda3/envs/dlb-di4c/bin/python",
+    "-B",
+    "-u",
+    "/home/ruizzhao/diffusion_baseline/adapters/sample_di4c.py",
+    "--upstream-root",
+    "/home/ruizzhao/diffusion_baseline/upstreams/di4c/sdtt",
+    "--checkpoint",
+    "/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/duo_di4c/lm1b/student_checkpoints/20000.ckpt",
+    "--config",
+    "/home/ruizzhao/diffusion_baseline/checkpoints/reference_reproduction/duo_di4c/lm1b/config.yaml",
+    "--checkpoint-sha256",
+    "212c0ee4ad726089dc72b947dc6c0cd852469ddfcc5cb5dea8794d19db2644b5",
+    "--config-sha256",
+    "61950a2556c19d77bf2d9aca26e689a250c41b68effa5e2a204c02ed6be8f6f0",
+    "--data-config",
+    "/home/ruizzhao/diffusion_baseline/artifacts/data.yaml",
+    "--downloads-manifest",
+    "/home/ruizzhao/diffusion_baseline/data/manifests/downloads.json",
+    "--tokenizer-snapshot",
+    "/home/ruizzhao/diffusion_baseline/data/raw/huggingface/hub/models--bert-base-uncased/snapshots/86b5e0934494bd15c9632b12f734a8a67f723594",
+    "--output",
+    "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_di4c/steps_2/upstream_token_ids.json",
+    "--sample-count",
+    "2048",
+    "--batch-size",
+    "16",
+    "--num-steps",
+    "2",
+    "--seq-len",
+    "128",
+    "--seed",
+    "42",
+    "--sampler",
+    "ancestral",
+    "--teacher-family",
+    "uniform_duo",
+    "--dataset",
+    "lm1b",
+    "--generation-mode",
+    "conditional_prefix",
+    "--conditioning-manifest",
+    "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256",
+    "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256",
+    "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length",
+    "64",
+    "--evaluation-continuation-length",
+    "64",
+    "--prompt-count",
+    "1024",
+    "--diversity-prompt-count",
+    "256",
+    "--completions-per-diversity-prompt",
+    "5",
+    "--completion-schedule",
+    "c0:p0-1023;c1-4:p0-255",
+    "--offline",
+    "true",
+    "--allow-missing-embedded-config",
+    "true"
+  ],
+  "command_sha256": "0972fe3ebfbb65be820274be2989ed40401fa00124a21cf1b9a3fb7ec00cde4e",
+  "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+  "environment": {
+    "cuda_version": null,
+    "cuda_visible_devices": "7",
+    "hostname": "DS210029",
+    "platform": "Linux-5.15.0-186-generic-x86_64-with-glibc2.35",
+    "python": "3.10.20",
+    "torch_version": "2.5.1"
+  },
+  "finished_at": "2026-08-28T12:13:24.379603Z",
+  "identity": {
+    "adapter_identity": "dlb.adapters.di4c:v3|dlb.adapters.di4c:Di4CAdapter|e1636b46628a6015944553de2d5d8a1b829bb6bf3f9c6b789d829cb6db222ec9",
+    "checkpoint_lock_id": "recipe:duo_di4c_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+    "checkpoint_selection": {
+      "output": "checkpoints/reference_reproduction/duo_di4c/lm1b",
+      "recipe": "duo_di4c_lm1b",
+      "sampling_checkpoint": "student_checkpoints/20000.ckpt",
+      "sampling_config": "config.yaml",
+      "source": "di4c",
+      "source_commit": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+      "teacher_adapter": "uniform_to_absorbing",
+      "teacher_family": "uniform_duo"
+    },
+    "checkpoint_sha256": "0828425487cc7d017c3ef345f6bd673c79a83355bc3ac0e2fb263011a1a22d4a",
+    "checkpoint_teacher_family": "uniform_duo",
+    "command_sha256": "0972fe3ebfbb65be820274be2989ed40401fa00124a21cf1b9a3fb7ec00cde4e",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-di4c",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "duo_di4c",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "duo_di4c-lm1b-steps-2",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+    "step_count": 2
+  },
+  "source_sha256": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+  "started_at": "2026-08-28T12:12:59.935747Z",
+  "status": "succeeded",
+  "stderr_log": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_di4c/steps_2/stderr.log",
+  "stdout_log": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_di4c/steps_2/stdout.log"
+}

results/conditional/samples/lm1b/duo_di4c/steps_2/stderr.log

@@ -0,0 +1 @@
+/opt/anaconda3/envs/dlb-di4c/lib/python3.10/site-packages/lightning/fabric/__init__.py:41: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.

results/conditional/samples/lm1b/duo_di4c/steps_8/conversion_metadata.json

@@ -0,0 +1,33 @@
+{
+  "checkpoint_lock_id": "recipe:duo_di4c_lm1b:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+  "checkpoint_selection": {
+    "output": "checkpoints/reference_reproduction/duo_di4c/lm1b",
+    "recipe": "duo_di4c_lm1b",
+    "sampling_checkpoint": "student_checkpoints/20000.ckpt",
+    "sampling_config": "config.yaml",
+    "source": "di4c",
+    "source_commit": "ac61ff9fe8e85120f9e1d2a8c5a332f8b8353dd3",
+    "teacher_adapter": "uniform_to_absorbing",
+    "teacher_family": "uniform_duo"
+  },
+  "checkpoint_sha256": "0828425487cc7d017c3ef345f6bd673c79a83355bc3ac0e2fb263011a1a22d4a",
+  "conditioning_implementation": "native_projection",
+  "exclude_from_latency": true,
+  "format": "dlb-conditional-token-capture-v1",
+  "generated_samples": 2048,
+  "generation_seconds_sentinel": 0.0,
+  "generation_seconds_source": "unavailable_excluded_sentinel",
+  "paper_conditional_training_reproduced": false,
+  "requested_samples": 2048,
+  "teacher_family": "uniform_duo",
+  "token_ids_source": "upstream_conditional_capture",
+  "token_ids_transformation": {
+    "max_length": 128,
+    "operation": "validated_exact_length"
+  },
+  "tokenizer": "bert-base-uncased",
+  "tokenizer_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "trim_policy": "none_exact_conditional_schedule",
+  "trimmed_samples": 0,
+  "upstream_output": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/duo_di4c/steps_8/upstream_token_ids.json"
+}

results/conditional/samples/lm1b/duo_di4c/steps_8/stderr.log

@@ -0,0 +1 @@
+/opt/anaconda3/envs/dlb-di4c/lib/python3.10/site-packages/lightning/fabric/__init__.py:41: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.

results/conditional/samples/lm1b/flm/steps_4/stdout.log

@@ -0,0 +1,253 @@
+DIT: USE_COMPILE=False
+CONFIG
+├── mode
+│   └── sample_eval                                                             
+├── is_di4c
+│   └── False                                                                   
+├── is_di4c_deterministic
+│   └── True                                                                    
+├── seed
+│   └── 42                                                                      
+├── loader
+│   └── global_batch_size: 512                                                  
+│       eval_global_batch_size: 512                                             
+│       batch_size: 2                                                           
+│       eval_batch_size: 32                                                     
+│       num_workers: 4                                                          
+│       pin_memory: true                                                        
+│                                                                               
+├── sampling
+│   └── predictor: ancestral                                                    
+│       steps: 4                                                                
+│       noise_removal: ancestral                                                
+│       use_float64: true                                                       
+│       p_nucleus: 1.0                                                          
+│       num_sample_batches: 64                                                  
+│       num_sample_log: 2                                                       
+│       semi_ar: false                                                          
+│       stride_length: 1                                                        
+│       num_strides: 1                                                          
+│       num_reflow_samples: 50000                                               
+│       temperature: 1.0                                                        
+│       solver: euler                                                           
+│       gamma: 0.0                                                              
+│                                                                               
+├── training
+│   └── loss_type: flow                                                         
+│       pred_type: x0                                                           
+│       ema: 0.9999                                                             
+│       antithetic_sampling: true                                               
+│       importance_sampling: false                                              
+│       sampling_eps: 0.001                                                     
+│       change_of_variables: false                                              
+│       loss_precision: bf16                                                    
+│       finetune_path: ''                                                       
+│       not_sampling_t: false                                                   
+│       load_ema: false                                                         
+│       t_curriculum_steps: 100000                                              
+│       t_curriculum_min: 0.0                                                   
+│       t_curriculum_max: 1.0                                                   
+│                                                                               
+├── eval
+│   └── checkpoint_path: /home/ruizzhao/diffusion_baseline/checkpoints/official/
+│       disable_ema: false                                                      
+│       ema_decay: null                                                         
+│       compute_generative_perplexity: true                                     
+│       perplexity_batch_size: 8                                                
+│       compute_perplexity_on_sanity: false                                     
+│       gen_ppl_eval_model_name_or_path: gpt2-large                             
+│       generate_samples: true                                                  
+│       generated_samples_path: /home/ruizzhao/diffusion_baseline/results/condit
+│                                                                               
+├── optim
+│   └── weight_decay: 0                                                         
+│       lr: 0.0003                                                              
+│       beta1: 0.9                                                              
+│       beta2: 0.999                                                            
+│       eps: 1.0e-08                                                            
+│       ln_tune: none                                                           
+│                                                                               
+├── trainer
+│   └── _target_: lightning.Trainer                                             
+│       accelerator: cuda                                                       
+│       num_nodes: 1                                                            
+│       devices: 1                                                              
+│       accumulate_grad_batches: 256                                            
+│       gradient_clip_val: 1.0                                                  
+│       precision: bf16                                                         
+│       num_sanity_val_steps: 2                                                 
+│       max_steps: 1000000                                                      
+│       log_every_n_steps: 100                                                  
+│       limit_train_batches: 1.0                                                
+│       limit_val_batches: 1.0                                                  
+│       val_check_interval: 5000                                                
+│                                                                               
+├── wandb
+│   └── project: flm                                                            
+│       notes: FLM                                                              
+│       group: null                                                             
+│       job_type: null                                                          
+│       name: null                                                              
+│       id: None_42                                                             
+│       tags:                                                                   
+│       - lm1b                                                                  
+│       - flm                                                                   
+│       offline: true                                                           
+│                                                                               
+├── checkpointing
+│   └── save_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/
+│       resume_from_ckpt: false                                                 
+│       resume_ckpt_path: /home/ruizzhao/diffusion_baseline/results/conditional/
+│                                                                               
+├── callbacks
+│   └── checkpoint_every_n_steps:                                               
+│         _target_: lightning.pytorch.callbacks.ModelCheckpoint                 
+│         save_top_k: -1                                                        
+│         save_last: false                                                      
+│         dirpath: /home/ruizzhao/diffusion_baseline/results/conditional/samples
+│         verbose: true                                                         
+│         auto_insert_metric_name: false                                        
+│         every_n_train_steps: 25000                                            
+│       checkpoint_monitor:                                                     
+│         _target_: lightning.pytorch.callbacks.ModelCheckpoint                 
+│         monitor: val/nll                                                      
+│         mode: min                                                             
+│         save_top_k: 1                                                         
+│         save_last: true                                                       
+│         dirpath: /home/ruizzhao/diffusion_baseline/results/conditional/samples
+│         filename: best_nll                                                    
+│         auto_insert_metric_name: false                                        
+│         verbose: true                                                         
+│       learning_rate_monitor:                                                  
+│         _target_: lightning.pytorch.callbacks.LearningRateMonitor             
+│         logging_interval: step                                                
+│                                                                               
+├── data
+│   └── train: lm1b                                                             
+│       valid: lm1b                                                             
+│       tokenizer_name_or_path: bert-base-uncased                               
+│       cache_dir: ./cache                                                      
+│       wrap: true                                                              
+│       streaming: false                                                        
+│       insert_train_eos: true                                                  
+│       insert_valid_eos: true                                                  
+│                                                                               
+├── model
+│   └── name: small                                                             
+│       type: ddit                                                              
+│       hidden_size: 768                                                        
+│       cond_dim: 128                                                           
+│       length: 128                                                             
+│       n_blocks: 12                                                            
+│       n_heads: 12                                                             
+│       scale_by_sigma: true                                                    
+│       dropout: 0.1                                                            
+│       tie_word_embeddings: false                                              
+│       vocab_lookup: true                                                      
+│                                                                               
+├── strategy
+│   └── _target_: lightning.pytorch.strategies.DDPStrategy                      
+│       find_unused_parameters: false                                           
+│                                                                               
+├── noise
+│   └── type: log-linear                                                        
+│       parameterization: log-linear                                            
+│       eps: 0                                                                  
+│       denoiser_latent_conditioning: -1                                        
+│       freeze_encoder: false                                                   
+│       freeze_decoder: false                                                   
+│                                                                               
+├── lr_scheduler
+│   └── _target_: transformers.get_constant_schedule_with_warmup                
+│       num_warmup_steps: 2500                                                  
+│                                                                               
+├── prior
+│   └── type: none                                                              
+│       latent_width: 0                                                         
+│       latent_height: 0                                                        
+│                                                                               
+└── algo
+    └── name: flm                                                               
+        backbone: dit                                                           
+        parameterization: mean                                                  
+        time_conditioning: true                                                 
+        T: 0                                                                    
+        subs_masking: false                                                     
+        causal_attention: false                                                 
+        ignore_bos: false                                                       
+        t_max: 1.0                                                              
+        t_min: 0.0                                                              
+        double_temb: false                                                      
+        learnable_loss_weighting: false                                         
+                                                                                
+[2026-08-25 20:21:08,070][dlb_pinned_upstream_main][INFO] - Starting Sample Eval.
+Resuming training from checkpoint...
+generation start: 2026-08-25 20:21:14
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+Copying EMA parameters to model
+generation end: 2026-08-25 20:23:07
+Generative perplexity: 34.48362731933594
+Sample entropy: 3.2911951541900635
+Samples saved at: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/flm/steps_4/upstream_samples.json

results/conditional/samples/lm1b/flm/steps_8/conversion_metadata.json

@@ -0,0 +1,29 @@
+{
+  "checkpoint_lock_id": "recipe:flm_lm1b_official_ckpt:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+  "checkpoint_selection": {
+    "output": "checkpoints/official/flm_ckpt/lm1b",
+    "recipe": "flm_lm1b_official_ckpt",
+    "sampling_checkpoint": "lm1b_flm.ckpt",
+    "sampling_config": null,
+    "source": "flm",
+    "source_commit": "a1918d5164e5038e37d0b7a4fb2010ce75b863b3",
+    "teacher_adapter": null,
+    "teacher_family": "continuous_flm"
+  },
+  "checkpoint_sha256": "99a97513471dc7232d4a0fc6b50d47301d97a6198e0cf13b77b3fe70c12c3665",
+  "conditioning_implementation": "zero_shot_runtime_projection",
+  "exclude_from_latency": true,
+  "format": "dlb-conditional-token-capture-v1",
+  "generated_samples": 2048,
+  "generation_seconds_sentinel": 0.0,
+  "generation_seconds_source": "unavailable_excluded_sentinel",
+  "paper_conditional_training_reproduced": false,
+  "requested_samples": 2048,
+  "teacher_family": "continuous_flm",
+  "token_ids_source": "upstream_conditional_capture",
+  "tokenizer": "bert-base-uncased",
+  "tokenizer_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "trim_policy": "none_exact_conditional_schedule",
+  "trimmed_samples": 0,
+  "upstream_output": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/flm/steps_8/upstream_token_ids.json"
+}

results/conditional/samples/lm1b/fmlm/steps_1/conversion_metadata.json

@@ -0,0 +1,29 @@
+{
+  "checkpoint_lock_id": "recipe:fmlm_lm1b_official_ckpt:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+  "checkpoint_selection": {
+    "output": "checkpoints/official/fmlm_ckpt/lm1b",
+    "recipe": "fmlm_lm1b_official_ckpt",
+    "sampling_checkpoint": "lm1b_fmlm.ckpt",
+    "sampling_config": null,
+    "source": "flm",
+    "source_commit": "a1918d5164e5038e37d0b7a4fb2010ce75b863b3",
+    "teacher_adapter": null,
+    "teacher_family": "continuous_flm"
+  },
+  "checkpoint_sha256": "cb096021fda9344c88b184d9b3ac8a0a3bc45e30be7bcc917e024971a9d67395",
+  "conditioning_implementation": "zero_shot_runtime_projection",
+  "exclude_from_latency": true,
+  "format": "dlb-conditional-token-capture-v1",
+  "generated_samples": 2048,
+  "generation_seconds_sentinel": 0.0,
+  "generation_seconds_source": "unavailable_excluded_sentinel",
+  "paper_conditional_training_reproduced": false,
+  "requested_samples": 2048,
+  "teacher_family": "continuous_flm",
+  "token_ids_source": "upstream_conditional_capture",
+  "tokenizer": "bert-base-uncased",
+  "tokenizer_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "trim_policy": "none_exact_conditional_schedule",
+  "trimmed_samples": 0,
+  "upstream_output": "/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_token_ids.json"
+}

results/conditional/samples/lm1b/fmlm/steps_1/request.json

@@ -0,0 +1,120 @@
+{
+  "command": [
+    "/opt/anaconda3/envs/dlb-flm/bin/python",
+    "-u",
+    "-m",
+    "dlb.adapters.capture",
+    "--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/flm/main.py",
+    "--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_token_ids.json",
+    "--generation-mode=conditional_prefix",
+    "--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "--prefix-length=64",
+    "--evaluation-continuation-length=64",
+    "--prompt-count=1024",
+    "--diversity-prompt-count=256",
+    "--completions-per-diversity-prompt=5",
+    "--completion-schedule=c0:p0-1023;c1-4:p0-255",
+    "--",
+    "mode=sample_eval",
+    "seed=42",
+    "data=lm1b-wrap",
+    "model=small",
+    "model.length=128",
+    "loader.batch_size=2",
+    "loader.eval_batch_size=16",
+    "sampling.num_sample_batches=128",
+    "sampling.steps=1",
+    "eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/official/fmlm_ckpt/lm1b/lm1b_fmlm.ckpt",
+    "trainer.devices=1",
+    "hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra",
+    "checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra",
+    "+wandb.offline=true",
+    "eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_samples.json",
+    "algo=fmlm",
+    "algo.backbone=dit",
+    "eval.disable_ema=False",
+    "sampling.solver=euler",
+    "algo.double_temb=True",
+    "algo.learnable_loss_weighting=False",
+    "sampling.gamma=0.8"
+  ],
+  "identity": {
+    "adapter_identity": "dlb.adapters.flm:v2|dlb.adapters.flm:FLMAdapter|c3a6a4573200fecf38a47fc6e7bc78254ca362fff3b96079dc2efa9f8e27aff0",
+    "checkpoint_lock_id": "recipe:fmlm_lm1b_official_ckpt:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+    "checkpoint_selection": {
+      "output": "checkpoints/official/fmlm_ckpt/lm1b",
+      "recipe": "fmlm_lm1b_official_ckpt",
+      "sampling_checkpoint": "lm1b_fmlm.ckpt",
+      "sampling_config": null,
+      "source": "flm",
+      "source_commit": "a1918d5164e5038e37d0b7a4fb2010ce75b863b3",
+      "teacher_adapter": null,
+      "teacher_family": "continuous_flm"
+    },
+    "checkpoint_sha256": "cb096021fda9344c88b184d9b3ac8a0a3bc45e30be7bcc917e024971a9d67395",
+    "checkpoint_teacher_family": "continuous_flm",
+    "command_sha256": "eb402b16a7fef3f5a67dc05255edf49e448c9704bb08e990e2983703bdd94e91",
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-flm",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "fmlm",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "fmlm-lm1b-steps-1",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "a1918d5164e5038e37d0b7a4fb2010ce75b863b3",
+    "step_count": 1
+  },
+  "request": {
+    "adapter_identity": "dlb.adapters.flm:v2|dlb.adapters.flm:FLMAdapter|c3a6a4573200fecf38a47fc6e7bc78254ca362fff3b96079dc2efa9f8e27aff0",
+    "checkpoint_lock_id": "recipe:fmlm_lm1b_official_ckpt:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152",
+    "checkpoint_selection": {
+      "output": "checkpoints/official/fmlm_ckpt/lm1b",
+      "recipe": "fmlm_lm1b_official_ckpt",
+      "sampling_checkpoint": "lm1b_fmlm.ckpt",
+      "sampling_config": null,
+      "source": "flm",
+      "source_commit": "a1918d5164e5038e37d0b7a4fb2010ce75b863b3",
+      "teacher_adapter": null,
+      "teacher_family": "continuous_flm"
+    },
+    "checkpoint_sha256": "cb096021fda9344c88b184d9b3ac8a0a3bc45e30be7bcc917e024971a9d67395",
+    "checkpoint_teacher_family": "continuous_flm",
+    "command": null,
+    "completion_schedule": "c0:p0-1023;c1-4:p0-255",
+    "completions_per_diversity_prompt": 5,
+    "conditioning_config_sha256": "89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114",
+    "conditioning_manifest": "/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json",
+    "conditioning_manifest_sha256": "0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039",
+    "config_sha256": "6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f",
+    "dataset_id": "lm1b",
+    "device": null,
+    "diversity_prompt_count": 256,
+    "environment": "dlb-flm",
+    "evaluation_continuation_length": 64,
+    "generation_mode": "conditional_prefix",
+    "model_id": "fmlm",
+    "prefix_length": 64,
+    "prompt_count": 1024,
+    "results_root": "/home/ruizzhao/diffusion_baseline/results/conditional",
+    "run_id": "fmlm-lm1b-steps-1",
+    "sample_count": 2048,
+    "seed": 42,
+    "source_sha256": "a1918d5164e5038e37d0b7a4fb2010ce75b863b3",
+    "step_count": 1
+  },
+  "started_at": "2026-08-25T12:51:02.713061Z"
+}

results/conditional/samples/lm1b/fmlm/steps_1/run_metadata.json

@@ -0,0 +1 @@
+{"checkpoint_lock_id":"recipe:fmlm_lm1b_official_ckpt:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152","checkpoint_selection":{"output":"checkpoints/official/fmlm_ckpt/lm1b","recipe":"fmlm_lm1b_official_ckpt","sampling_checkpoint":"lm1b_fmlm.ckpt","sampling_config":null,"source":"flm","source_commit":"a1918d5164e5038e37d0b7a4fb2010ce75b863b3","teacher_adapter":null,"teacher_family":"continuous_flm"},"checkpoint_sha256":"cb096021fda9344c88b184d9b3ac8a0a3bc45e30be7bcc917e024971a9d67395","checkpoint_teacher_family":"continuous_flm","command":["/opt/anaconda3/envs/dlb-flm/bin/python","-u","-m","dlb.adapters.capture","--upstream-entrypoint=/home/ruizzhao/diffusion_baseline/upstreams/flm/main.py","--capture-path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_token_ids.json","--generation-mode=conditional_prefix","--conditioning-manifest=/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json","--conditioning-manifest-sha256=0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039","--conditioning-config-sha256=89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114","--prefix-length=64","--evaluation-continuation-length=64","--prompt-count=1024","--diversity-prompt-count=256","--completions-per-diversity-prompt=5","--completion-schedule=c0:p0-1023;c1-4:p0-255","--","mode=sample_eval","seed=42","data=lm1b-wrap","model=small","model.length=128","loader.batch_size=2","loader.eval_batch_size=16","sampling.num_sample_batches=128","sampling.steps=1","eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/official/fmlm_ckpt/lm1b/lm1b_fmlm.ckpt","trainer.devices=1","hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra","checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra","+wandb.offline=true","eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_samples.json","algo=fmlm","algo.backbone=dit","eval.disable_ema=False","sampling.solver=euler","algo.double_temb=True","algo.learnable_loss_weighting=False","sampling.gamma=0.8"],"command_sha256":"eb402b16a7fef3f5a67dc05255edf49e448c9704bb08e990e2983703bdd94e91","config_sha256":"6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f","environment":{"cuda_version":null,"cuda_visible_devices":"4","hostname":"DS210029","platform":"Linux-5.15.0-186-generic-x86_64-with-glibc2.35","python":"3.11.15","torch_version":"2.5.1"},"finished_at":"2026-08-25T12:54:35.677016Z","identity":{"adapter_identity":"dlb.adapters.flm:v2|dlb.adapters.flm:FLMAdapter|c3a6a4573200fecf38a47fc6e7bc78254ca362fff3b96079dc2efa9f8e27aff0","checkpoint_lock_id":"recipe:fmlm_lm1b_official_ckpt:ce8ff6a327d9f1470fef9a7ac66813ac94d4cc50268bc2998627498dc6f3a152","checkpoint_selection":{"output":"checkpoints/official/fmlm_ckpt/lm1b","recipe":"fmlm_lm1b_official_ckpt","sampling_checkpoint":"lm1b_fmlm.ckpt","sampling_config":null,"source":"flm","source_commit":"a1918d5164e5038e37d0b7a4fb2010ce75b863b3","teacher_adapter":null,"teacher_family":"continuous_flm"},"checkpoint_sha256":"cb096021fda9344c88b184d9b3ac8a0a3bc45e30be7bcc917e024971a9d67395","checkpoint_teacher_family":"continuous_flm","command_sha256":"eb402b16a7fef3f5a67dc05255edf49e448c9704bb08e990e2983703bdd94e91","completion_schedule":"c0:p0-1023;c1-4:p0-255","completions_per_diversity_prompt":5,"conditioning_config_sha256":"89d70134c4fb1ce21c18e46cfb9753f6cdcd460b879d6e60e99b2c7b56ae5114","conditioning_manifest":"/home/ruizzhao/diffusion_baseline/data/manifests/conditional-lm1b-c64.json","conditioning_manifest_sha256":"0e5c5bf5b8a1445b348a65045f680ac31713dbfa1e690bedb73e0d1e7472a039","config_sha256":"6ce963030186c79cab2bc15670737a77ebfad08a0608bc87ee5c54e489ce678f","dataset_id":"lm1b","device":null,"diversity_prompt_count":256,"environment":"dlb-flm","evaluation_continuation_length":64,"generation_mode":"conditional_prefix","model_id":"fmlm","prefix_length":64,"prompt_count":1024,"results_root":"/home/ruizzhao/diffusion_baseline/results/conditional","run_id":"fmlm-lm1b-steps-1","sample_count":2048,"seed":42,"source_sha256":"a1918d5164e5038e37d0b7a4fb2010ce75b863b3","step_count":1},"request":{"sample_count":2048},"sample_count":2048,"source_sha256":"a1918d5164e5038e37d0b7a4fb2010ce75b863b3","started_at":"2026-08-25T12:51:02.713061Z","status":"succeeded","stderr_log":"/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/stderr.log","stdout_log":"/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/stdout.log"}

results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra/.hydra/hydra.yaml

@@ -0,0 +1,183 @@
+hydra:
+  run:
+    dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.run.dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra
+    - hydra.mode=RUN
+    task:
+    - mode=sample_eval
+    - seed=42
+    - data=lm1b-wrap
+    - model=small
+    - model.length=128
+    - loader.batch_size=2
+    - loader.eval_batch_size=1
+    - sampling.num_sample_batches=1
+    - sampling.steps=1
+    - eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/official/fmlm_ckpt/lm1b/lm1b_fmlm.ckpt
+    - trainer.devices=1
+    - checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra
+    - +wandb.offline=true
+    - eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_samples.json
+    - algo=fmlm
+    - algo.backbone=dit
+    - eval.disable_ema=False
+    - sampling.solver=euler
+    - algo.double_temb=True
+    - algo.learnable_loss_weighting=False
+    - sampling.gamma=0.8
+  job:
+    name: dlb_pinned_upstream_main
+    chdir: true
+    override_dirname: +wandb.offline=true,algo.backbone=dit,algo.double_temb=True,algo.learnable_loss_weighting=False,algo=fmlm,checkpointing.save_dir=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra,data=lm1b-wrap,eval.checkpoint_path=/home/ruizzhao/diffusion_baseline/checkpoints/official/fmlm_ckpt/lm1b/lm1b_fmlm.ckpt,eval.disable_ema=False,eval.generated_samples_path=/home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_samples.json,loader.batch_size=2,loader.eval_batch_size=1,mode=sample_eval,model.length=128,model=small,sampling.gamma=0.8,sampling.num_sample_batches=1,sampling.solver=euler,sampling.steps=1,seed=42,trainer.devices=1
+    id: ???
+    num: ???
+    config_name: config
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/ruizzhao/diffusion_baseline
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /home/ruizzhao/diffusion_baseline/upstreams/flm/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/ruizzhao/diffusion_baseline/results/conditional/samples/lm1b/fmlm/steps_1/upstream_hydra
+    choices:
+      algo: fmlm
+      prior: none
+      lr_scheduler: constant_warmup
+      noise: log-linear
+      strategy: ddp
+      model: small
+      data: lm1b-wrap
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false