Hugging Face's logo

Datasets:
Harshvardhan-Mestha
/
esp-reproduction-artifacts

Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
llama3.2_3b: struct<bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, a (... 314 chars omitted)
  child 0, bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 95 chars omitted)
      child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
          child 0, model: string
          child 1, block_complexity: int64
          child 2, n_mask_tokens: int64
          child 3, avg_ar_tps: double
          child 4, avg_esp_tps: double
          child 5, avg_tau: double
          child 6, avg_cos_final: double
          child 7, avg_acceptance_rate: double
  child 1, bc_60: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 95 chars omitted)
      child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
          child 0, model: string
          child 1, block_complexity: int64
          child 2, n_mask_tokens: int64
          child 3, avg_ar_tps: double
          child 4, avg_esp_tps: double
          child 5, avg_tau: double
          child 6, avg_cos_final: double
          child 7, avg_acceptance_rate: double
llama3.1_8b: struct<bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, a (... 363 chars omitted)
  child 0, b
...
        child 7, avg_acceptance_rate: double
      child 1, per_prompt_tau: list<item: double>
          child 0, item: double
bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 140 chars omitted)
  child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
      child 0, model: string
      child 1, block_complexity: int64
      child 2, n_mask_tokens: int64
      child 3, avg_ar_tps: double
      child 4, avg_esp_tps: double
      child 5, avg_tau: double
      child 6, avg_cos_final: double
      child 7, avg_acceptance_rate: double
  child 1, per_prompt: list<item: struct<tau: double>>
      child 0, item: struct<tau: double>
          child 0, tau: double
bc_60: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 140 chars omitted)
  child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
      child 0, model: string
      child 1, block_complexity: int64
      child 2, n_mask_tokens: int64
      child 3, avg_ar_tps: double
      child 4, avg_esp_tps: double
      child 5, avg_tau: double
      child 6, avg_cos_final: double
      child 7, avg_acceptance_rate: double
  child 1, per_prompt: list<item: struct<tau: double>>
      child 0, item: struct<tau: double>
          child 0, tau: double
to
{'bc_30': {'summary': {'model': Value('string'), 'block_complexity': Value('int64'), 'n_mask_tokens': Value('int64'), 'avg_ar_tps': Value('float64'), 'avg_esp_tps': Value('float64'), 'avg_tau': Value('float64'), 'avg_cos_final': Value('float64'), 'avg_acceptance_rate': Value('float64')}, 'per_prompt': List({'tau': Value('float64')})}, 'bc_60': {'summary': {'model': Value('string'), 'block_complexity': Value('int64'), 'n_mask_tokens': Value('int64'), 'avg_ar_tps': Value('float64'), 'avg_esp_tps': Value('float64'), 'avg_tau': Value('float64'), 'avg_cos_final': Value('float64'), 'avg_acceptance_rate': Value('float64')}, 'per_prompt': List({'tau': Value('float64')})}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 489, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              llama3.2_3b: struct<bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, a (... 314 chars omitted)
                child 0, bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 95 chars omitted)
                    child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
                        child 0, model: string
                        child 1, block_complexity: int64
                        child 2, n_mask_tokens: int64
                        child 3, avg_ar_tps: double
                        child 4, avg_esp_tps: double
                        child 5, avg_tau: double
                        child 6, avg_cos_final: double
                        child 7, avg_acceptance_rate: double
                child 1, bc_60: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 95 chars omitted)
                    child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
                        child 0, model: string
                        child 1, block_complexity: int64
                        child 2, n_mask_tokens: int64
                        child 3, avg_ar_tps: double
                        child 4, avg_esp_tps: double
                        child 5, avg_tau: double
                        child 6, avg_cos_final: double
                        child 7, avg_acceptance_rate: double
              llama3.1_8b: struct<bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, a (... 363 chars omitted)
                child 0, b
              ...
                      child 7, avg_acceptance_rate: double
                    child 1, per_prompt_tau: list<item: double>
                        child 0, item: double
              bc_30: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 140 chars omitted)
                child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
                    child 0, model: string
                    child 1, block_complexity: int64
                    child 2, n_mask_tokens: int64
                    child 3, avg_ar_tps: double
                    child 4, avg_esp_tps: double
                    child 5, avg_tau: double
                    child 6, avg_cos_final: double
                    child 7, avg_acceptance_rate: double
                child 1, per_prompt: list<item: struct<tau: double>>
                    child 0, item: struct<tau: double>
                        child 0, tau: double
              bc_60: struct<summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: dou (... 140 chars omitted)
                child 0, summary: struct<model: string, block_complexity: int64, n_mask_tokens: int64, avg_ar_tps: double, avg_esp_tps (... 78 chars omitted)
                    child 0, model: string
                    child 1, block_complexity: int64
                    child 2, n_mask_tokens: int64
                    child 3, avg_ar_tps: double
                    child 4, avg_esp_tps: double
                    child 5, avg_tau: double
                    child 6, avg_cos_final: double
                    child 7, avg_acceptance_rate: double
                child 1, per_prompt: list<item: struct<tau: double>>
                    child 0, item: struct<tau: double>
                        child 0, tau: double
              to
              {'bc_30': {'summary': {'model': Value('string'), 'block_complexity': Value('int64'), 'n_mask_tokens': Value('int64'), 'avg_ar_tps': Value('float64'), 'avg_esp_tps': Value('float64'), 'avg_tau': Value('float64'), 'avg_cos_final': Value('float64'), 'avg_acceptance_rate': Value('float64')}, 'per_prompt': List({'tau': Value('float64')})}, 'bc_60': {'summary': {'model': Value('string'), 'block_complexity': Value('int64'), 'n_mask_tokens': Value('int64'), 'avg_ar_tps': Value('float64'), 'avg_esp_tps': Value('float64'), 'avg_tau': Value('float64'), 'avg_cos_final': Value('float64'), 'avg_acceptance_rate': Value('float64')}, 'per_prompt': List({'tau': Value('float64')})}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

ESP Reproduction Artifacts

All code, results, logs, and reports for the reproduction of ESP (Embedding-Space Probing) from arXiv 2603.17942 (ICML 2026).

Contents 

├── code/
│   ├── esp_v5.py          # V5 implementation (2-pass, temp=0.01)
│   ├── esp_v6.py          # V6 implementation (single-pass tree attention, temp=1.0)
│   ├── repro_lemma31.py   # Lemma 3.1 numerical verification
│   └── run.sh             # HF job runner
├── results/
│   ├── esp_v5_results.json    # V5 results (LLaMA3.2-3B)
│   ├── esp_v6_results.json    # V6 results (all 3 models)
│   └── lemma31_results.json   # Lemma 3.1 audit (200 trials, 0 violations)
├── poster/
│   ├── esp_reproduction_poster.pdf  # Research poster (PDF)
│   └── esp_reproduction_poster.png  # Research poster (PNG)
├── logbook/
│   ├── index.md
│   ├── executive-summary/page.md
│   ├── claim-1-*/page.md through claim-5-*/page.md
│   └── conclusion/page.md
└── README.md

Key Results

Model BC τ (Ours) τ (Paper)
LLaMA3.2-3B 30 1.45 1.56
LLaMA3.2-3B 60 1.47 1.63
LLaMA3.1-8B 30 2.00 1.63
Qwen3-8B 30 2.91 1.74

Reproduction Status

  • ✅ Lemma 3.1: 0 violations (verified)
  • ⚠️ Table 1 τ: τ > 1 confirmed, values within 7-80% of paper
  • ❌ 12% quality: Not tested (needs SpecBench)
  • ❌ 15-19% throughput: Not reproduced (needs KV cache + FlashAttention)
  • ⚠️ Cosine ~0.45 vs ~0.35: Pattern confirmed, values vary by model

Paper

arXiv 2603.17942 — Efficient Training-Free Multi-Token Prediction via Embedding-Space Probing ICML 2026, OpenReview id mKj4fO0BU4
Downloads last month
69
Total file size:
1.89 MB