24/24 Retrieval, Yet the Embeddings Changed: An MLX Q4–Q8 Sweep with CUDA Controls
TiGa-RCE
• • 2Dataset Viewer
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
metrics: struct<pair_count: int64, top1: double, recall_at_5: double, mrr: double, mean_margin: double, minim (... 405 chars omitted)
child 0, pair_count: int64
child 1, top1: double
child 2, recall_at_5: double
child 3, mrr: double
child 4, mean_margin: double
child 5, minimum_margin: double
child 6, mean_rank: double
child 7, worst_rank: int64
child 8, lane: string
child 9, family: string
child 10, model: string
child 11, source_revision: string
child 12, load_seconds: double
child 13, loading_strategy: string
child 14, encode_seconds: double
child 15, texts_per_second: double
child 16, torch_version: string
child 17, cuda_device: string
child 18, cuda_allocation_before_load: int64
child 19, cuda_allocation_after_load: int64
child 20, cuda_incremental_model_allocation: int64
child 21, cuda_peak_bytes: int64
mlx_bf16_comparison: struct<mean_aligned_cosine_cuda_vs_mlx_bf16: double, minimum_aligned_cosine_cuda_vs_mlx_bf16: double (... 88 chars omitted)
child 0, mean_aligned_cosine_cuda_vs_mlx_bf16: double
child 1, minimum_aligned_cosine_cuda_vs_mlx_bf16: double
child 2, score_rmse_cuda_vs_mlx_bf16: double
child 3, queries_with_rank_change_cuda_vs_mlx_bf16: int64
seq_length: int64
format: string
calib_data_hash: string
collection: struct<dataset: string, requested_samples: int64, seq_length: int64, adaptive: bool, adaptive_step_s (... 1313 chars omitted)
child 0, dataset: string
child 1, requested_samples: int64
child 2,
...
child 2, collection_sufficient: bool
child 3, coverage: struct<has_expert_counts: bool, expert_modules: int64, total_experts: int64, active_experts: int64, (... 199 chars omitted)
child 0, has_expert_counts: bool
child 1, expert_modules: int64
child 2, total_experts: int64
child 3, active_experts: int64
child 4, zero_count_experts: int64
child 5, active_ratio: double
child 6, min_count: int64
child 7, p05_count: double
child 8, p10_count: double
child 9, median_count: double
child 10, max_count: int64
child 11, min_required_count: int64
child 12, required_percentile: int64
model_name: string
expert_coverage: struct<has_expert_counts: bool, expert_modules: int64, total_experts: int64, active_experts: int64, (... 199 chars omitted)
child 0, has_expert_counts: bool
child 1, expert_modules: int64
child 2, total_experts: int64
child 3, active_experts: int64
child 4, zero_count_experts: int64
child 5, active_ratio: double
child 6, min_count: int64
child 7, p05_count: double
child 8, p10_count: double
child 9, median_count: double
child 10, max_count: int64
child 11, min_required_count: int64
child 12, required_percentile: int64
source_hash: string
entry_count: int64
processed_samples: int64
num_samples: int64
requires_expert_counts: bool
calib_dataset: string
to
{'format': Value('string'), 'model_name': Value('string'), 'source_hash': Value('string'), 'calib_dataset': Value('string'), 'calib_data_hash': Value('string'), 'num_samples': Value('int64'), 'seq_length': Value('int64'), 'entry_count': Value('int64'), 'collection': {'dataset': Value('string'), 'requested_samples': Value('int64'), 'seq_length': Value('int64'), 'adaptive': Value('bool'), 'adaptive_step_samples': Value('int64'), 'adaptive_max_samples': Value('int64'), 'available_samples': Value('int64'), 'micro_batch_size': Value('int64'), 'micro_batches': Value('int64'), 'batch_plan': {'micro_batch_size': Value('int64'), 'estimated_sample_bytes': Value('int64'), 'capture_budget_bytes': Value('int64'), 'system_available_bytes': Value('int64'), 'metal_available_bytes': Value('int64'), 'live_available_bytes': Value('int64'), 'hidden_size': Value('int64'), 'num_experts': Value('int64'), 'top_k': Value('int64')}, 'processed_samples': Value('int64'), 'installed_modules': Value('int64'), 'capture_module_classes': {'Linear': Value('int64')}, 'switch_capture_modules': Value('int64'), 'requires_expert_counts': Value('bool'), 'coverage_sufficient': Value('bool'), 'collection_sufficient': Value('bool'), 'coverage': {'has_expert_counts': Value('bool'), 'expert_modules': Value('int64'), 'total_experts': Value('int64'), 'active_experts': Value('int64'), 'zero_count_experts': Value('int64'), 'active_ratio': Value('float64'), 'min_count': Value('int64'), 'p05_count': Value('float64'), 'p10_count': Value('float64'), 'median_count': Value('float64'), 'max_count': Value('int64'), 'min_required_count': Value('int64'), 'required_percentile': Value('int64')}, 'rounds': List({'processed_samples': Value('int64'), 'coverage_sufficient': Value('bool'), 'collection_sufficient': Value('bool'), 'coverage': {'has_expert_counts': Value('bool'), 'expert_modules': Value('int64'), 'total_experts': Value('int64'), 'active_experts': Value('int64'), 'zero_count_experts': Value('int64'), 'active_ratio': Value('float64'), 'min_count': Value('int64'), 'p05_count': Value('float64'), 'p10_count': Value('float64'), 'median_count': Value('float64'), 'max_count': Value('int64'), 'min_required_count': Value('int64'), 'required_percentile': Value('int64')}})}, 'expert_coverage': {'has_expert_counts': Value('bool'), 'expert_modules': Value('int64'), 'total_experts': Value('int64'), 'active_experts': Value('int64'), 'zero_count_experts': Value('int64'), 'active_ratio': Value('float64'), 'min_count': Value('int64'), 'p05_count': Value('float64'), 'p10_count': Value('float64'), 'median_count': Value('float64'), 'max_count': Value('int64'), 'min_required_count': Value('int64'), 'required_percentile': Value('int64')}, 'requires_expert_counts': Value('bool'), 'processed_samples': Value('int64')}
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
metrics: struct<pair_count: int64, top1: double, recall_at_5: double, mrr: double, mean_margin: double, minim (... 405 chars omitted)
child 0, pair_count: int64
child 1, top1: double
child 2, recall_at_5: double
child 3, mrr: double
child 4, mean_margin: double
child 5, minimum_margin: double
child 6, mean_rank: double
child 7, worst_rank: int64
child 8, lane: string
child 9, family: string
child 10, model: string
child 11, source_revision: string
child 12, load_seconds: double
child 13, loading_strategy: string
child 14, encode_seconds: double
child 15, texts_per_second: double
child 16, torch_version: string
child 17, cuda_device: string
child 18, cuda_allocation_before_load: int64
child 19, cuda_allocation_after_load: int64
child 20, cuda_incremental_model_allocation: int64
child 21, cuda_peak_bytes: int64
mlx_bf16_comparison: struct<mean_aligned_cosine_cuda_vs_mlx_bf16: double, minimum_aligned_cosine_cuda_vs_mlx_bf16: double (... 88 chars omitted)
child 0, mean_aligned_cosine_cuda_vs_mlx_bf16: double
child 1, minimum_aligned_cosine_cuda_vs_mlx_bf16: double
child 2, score_rmse_cuda_vs_mlx_bf16: double
child 3, queries_with_rank_change_cuda_vs_mlx_bf16: int64
seq_length: int64
format: string
calib_data_hash: string
collection: struct<dataset: string, requested_samples: int64, seq_length: int64, adaptive: bool, adaptive_step_s (... 1313 chars omitted)
child 0, dataset: string
child 1, requested_samples: int64
child 2,
...
child 2, collection_sufficient: bool
child 3, coverage: struct<has_expert_counts: bool, expert_modules: int64, total_experts: int64, active_experts: int64, (... 199 chars omitted)
child 0, has_expert_counts: bool
child 1, expert_modules: int64
child 2, total_experts: int64
child 3, active_experts: int64
child 4, zero_count_experts: int64
child 5, active_ratio: double
child 6, min_count: int64
child 7, p05_count: double
child 8, p10_count: double
child 9, median_count: double
child 10, max_count: int64
child 11, min_required_count: int64
child 12, required_percentile: int64
model_name: string
expert_coverage: struct<has_expert_counts: bool, expert_modules: int64, total_experts: int64, active_experts: int64, (... 199 chars omitted)
child 0, has_expert_counts: bool
child 1, expert_modules: int64
child 2, total_experts: int64
child 3, active_experts: int64
child 4, zero_count_experts: int64
child 5, active_ratio: double
child 6, min_count: int64
child 7, p05_count: double
child 8, p10_count: double
child 9, median_count: double
child 10, max_count: int64
child 11, min_required_count: int64
child 12, required_percentile: int64
source_hash: string
entry_count: int64
processed_samples: int64
num_samples: int64
requires_expert_counts: bool
calib_dataset: string
to
{'format': Value('string'), 'model_name': Value('string'), 'source_hash': Value('string'), 'calib_dataset': Value('string'), 'calib_data_hash': Value('string'), 'num_samples': Value('int64'), 'seq_length': Value('int64'), 'entry_count': Value('int64'), 'collection': {'dataset': Value('string'), 'requested_samples': Value('int64'), 'seq_length': Value('int64'), 'adaptive': Value('bool'), 'adaptive_step_samples': Value('int64'), 'adaptive_max_samples': Value('int64'), 'available_samples': Value('int64'), 'micro_batch_size': Value('int64'), 'micro_batches': Value('int64'), 'batch_plan': {'micro_batch_size': Value('int64'), 'estimated_sample_bytes': Value('int64'), 'capture_budget_bytes': Value('int64'), 'system_available_bytes': Value('int64'), 'metal_available_bytes': Value('int64'), 'live_available_bytes': Value('int64'), 'hidden_size': Value('int64'), 'num_experts': Value('int64'), 'top_k': Value('int64')}, 'processed_samples': Value('int64'), 'installed_modules': Value('int64'), 'capture_module_classes': {'Linear': Value('int64')}, 'switch_capture_modules': Value('int64'), 'requires_expert_counts': Value('bool'), 'coverage_sufficient': Value('bool'), 'collection_sufficient': Value('bool'), 'coverage': {'has_expert_counts': Value('bool'), 'expert_modules': Value('int64'), 'total_experts': Value('int64'), 'active_experts': Value('int64'), 'zero_count_experts': Value('int64'), 'active_ratio': Value('float64'), 'min_count': Value('int64'), 'p05_count': Value('float64'), 'p10_count': Value('float64'), 'median_count': Value('float64'), 'max_count': Value('int64'), 'min_required_count': Value('int64'), 'required_percentile': Value('int64')}, 'rounds': List({'processed_samples': Value('int64'), 'coverage_sufficient': Value('bool'), 'collection_sufficient': Value('bool'), 'coverage': {'has_expert_counts': Value('bool'), 'expert_modules': Value('int64'), 'total_experts': Value('int64'), 'active_experts': Value('int64'), 'zero_count_experts': Value('int64'), 'active_ratio': Value('float64'), 'min_count': Value('int64'), 'p05_count': Value('float64'), 'p10_count': Value('float64'), 'median_count': Value('float64'), 'max_count': Value('int64'), 'min_required_count': Value('int64'), 'required_percentile': Value('int64')}})}, 'expert_coverage': {'has_expert_counts': Value('bool'), 'expert_modules': Value('int64'), 'total_experts': Value('int64'), 'active_experts': Value('int64'), 'zero_count_experts': Value('int64'), 'active_ratio': Value('float64'), 'min_count': Value('int64'), 'p05_count': Value('float64'), 'p10_count': Value('float64'), 'median_count': Value('float64'), 'max_count': Value('int64'), 'min_required_count': Value('int64'), 'required_percentile': Value('int64')}, 'requires_expert_counts': Value('bool'), 'processed_samples': Value('int64')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Embedding quantization reproducibility bundle
This dataset contains the frozen inputs, vector artifacts, metrics, calibration evidence, and scripts behind the companion engineering post on MLX Q/oQ/oQe embedding quantization and CUDA-native controls.
See RESULTS.md for the compact result tables and evidence
boundaries.
The model weights are not duplicated here. models.lock.json pins all 30 public
model repositories to exact Hub commits and records their canonical upstream
BF16 revisions. Quantized artifacts branch directly from their family's BF16
checkpoint; no lossy conversion was used as another quantization's source.
Verify the published evidence
hf download TiGa-RCE/embedding-quant-repro-2026-07-28 \
--repo-type dataset --local-dir embedding-quant-repro
cd embedding-quant-repro
uv sync
uv run python reproduce.py verify
uv run python reproduce.py mixed-index
verify checks the locked revisions, bundle checksums, all 30 saved MLX vector
artifacts, and the three-family mixed-index result. mixed-index regenerates the
54 migration-direction measurements from the saved vectors without downloading
model weights.
Local MLX gate on Apple Silicon
The quick profile downloads the locked 0.6B BF16 and representative Q4/Q6/Q8 artifacts, then reproduces their vector comparisons:
uv sync --extra mlx
uv run python reproduce.py mlx --profile quick
The complete matrix is intentionally explicit because it downloads all 30 checkpoints:
uv run python reproduce.py mlx --profile full --family all
CUDA controls through ZeroGPU
After authenticating with Hugging Face, the client invokes the bounded BF16, bitsandbytes INT8, and bitsandbytes NF4 controls:
hf auth login
uv sync --extra zerogpu
uv run python reproduce.py cuda
These bitsandbytes results are controls, not MLX Q/oQ/oQe equivalents. ZeroGPU quota and scheduling still apply. The default runs the three 0.6B controls. The complete nine-run matrix is deliberately explicit because it can exceed a free account's daily quota:
uv run python reproduce.py cuda --family all --variant all
Evidence boundaries
- The frozen set has 24 query/document pairs. This is an engineering smoke test, not MTEB or a universal retrieval-quality claim.
- Q4 remains in the bundle as an intentionally failed fidelity comparator.
- GTE-Qwen2 1.5B CUDA vectors are quarantined from MLX cross-runtime trends because that family failed loading-path parity.
scripts/frozen-original/preserves the exact experiment scripts, including their historical local paths. Usereproduce.pyfor the portable interface.- The bundle contains no credentials.
- Downloads last month
- 114