actual_steps int64 2.44k 14.6k ⌀ | artifact_id stringlengths 30 48 | artifact_kind stringclasses 2
values | base_model stringclasses 9
values | best_step int64 2k 14k ⌀ | best_val_ppl float64 8.34 22.5 ⌀ | condition stringclasses 1
value | elapsed_hours float64 0.13 16.7 ⌀ | gate_frac_above_09 float64 0.03 0.21 ⌀ | gate_frac_below_01 float64 0.01 0.03 ⌀ | gate_max float64 1 1 ⌀ | gate_mean float64 0.47 0.66 ⌀ | gate_median float64 0.46 0.7 ⌀ | gate_min float64 0 0.02 ⌀ | gate_q25 float64 0.29 0.48 ⌀ | gate_q75 float64 0.65 0.87 ⌀ | gate_std float64 0.22 0.25 ⌀ | hf_repo stringlengths 43 61 | n_test_batches int64 0 0 ⌀ | result_available bool 2
classes | seed int64 42 42 ⌀ | skipped_final_test_eval bool 1
class | source_memory_repo stringclasses 8
values | test_ppl_ci_lower null | test_ppl_ci_upper null | test_ppl_mean null | test_ppl_std null | title stringlengths 40 65 | token_budget stringclasses 6
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
null | xmemtransfer-llama2-source-10m | source_memory | NousResearch/Llama-2-7b-hf | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-llama2-source-10m | null | false | null | null | null | null | null | null | null | XMemTransfer LLaMA-2 source memory (10M) | 10m |
null | xmemtransfer-llama2-source-15m | source_memory | NousResearch/Llama-2-7b-hf | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-llama2-source-15m | null | false | null | null | null | null | null | null | null | XMemTransfer LLaMA-2 source memory (15M) | 15m |
null | xmemtransfer-llama2-source-20m | source_memory | NousResearch/Llama-2-7b-hf | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-llama2-source-20m | null | false | null | null | null | null | null | null | null | XMemTransfer LLaMA-2 source memory (20M) | 20m |
null | xmemtransfer-llama2-source-25m | source_memory | NousResearch/Llama-2-7b-hf | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-llama2-source-25m | null | false | null | null | null | null | null | null | null | XMemTransfer LLaMA-2 source memory (25M) | 25m |
null | xmemtransfer-llama2-source-30m | source_memory | NousResearch/Llama-2-7b-hf | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-llama2-source-30m | null | false | null | null | null | null | null | null | null | XMemTransfer LLaMA-2 source memory (30M) | 30m |
4,882 | xmemtransfer-mistral-from-llama2-r4-10m10m | target_adaptor | mistralai/Mistral-7B-v0.3 | 4,500 | 9.228136 | transferred | 1.871287 | 0.029006 | 0.023889 | 0.999999 | 0.494295 | 0.489888 | 0.002365 | 0.316117 | 0.668401 | 0.224394 | OLAResearchX/xmemtransfer-mistral-from-llama2-r4-10m10m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-llama2-source-10m | null | null | null | null | XMemTransfer Mistral adaptor from LLaMA-2 source (R4, 10M/10M) | 10m |
7,324 | xmemtransfer-mistral-from-llama2-r4-15m15m | target_adaptor | mistralai/Mistral-7B-v0.3 | 7,000 | 8.852332 | transferred | 2.828927 | 0.030796 | 0.026853 | 1 | 0.489762 | 0.483772 | 0.000456 | 0.309039 | 0.665279 | 0.226862 | OLAResearchX/xmemtransfer-mistral-from-llama2-r4-15m15m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-llama2-source-15m | null | null | null | null | XMemTransfer Mistral adaptor from LLaMA-2 source (R4, 15M/15M) | 15m |
9,765 | xmemtransfer-mistral-from-llama2-r4-20m20m | target_adaptor | mistralai/Mistral-7B-v0.3 | 9,500 | 8.58454 | transferred | 3.813666 | 0.029259 | 0.029583 | 1 | 0.483866 | 0.476918 | 0.000244 | 0.301752 | 0.659539 | 0.227535 | OLAResearchX/xmemtransfer-mistral-from-llama2-r4-20m20m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-llama2-source-20m | null | null | null | null | XMemTransfer Mistral adaptor from LLaMA-2 source (R4, 20M/20M) | 20m |
12,207 | xmemtransfer-mistral-from-llama2-r4-25m25m | target_adaptor | mistralai/Mistral-7B-v0.3 | 12,000 | 8.480824 | transferred | 4.80569 | 0.028945 | 0.030899 | 0.999999 | 0.479121 | 0.470493 | 0.000258 | 0.297 | 0.653837 | 0.227563 | OLAResearchX/xmemtransfer-mistral-from-llama2-r4-25m25m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-llama2-source-25m | null | null | null | null | XMemTransfer Mistral adaptor from LLaMA-2 source (R4, 25M/25M) | 25m |
14,648 | xmemtransfer-mistral-from-llama2-r4-30m30m | target_adaptor | mistralai/Mistral-7B-v0.3 | 14,000 | 8.33989 | transferred | 5.750856 | 0.027964 | 0.032778 | 0.999999 | 0.47365 | 0.463506 | 0.000192 | 0.291212 | 0.64768 | 0.227585 | OLAResearchX/xmemtransfer-mistral-from-llama2-r4-30m30m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-llama2-source-30m | null | null | null | null | XMemTransfer Mistral adaptor from LLaMA-2 source (R4, 30M/30M) | 30m |
null | xmemtransfer-qwen35-08b-source-50m | source_memory | Qwen/Qwen3.5-0.8B-Base | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-qwen35-08b-source-50m | null | false | null | null | null | null | null | null | null | XMemTransfer Qwen3.5-0.8B source memory (50M) | 50m |
2,441 | xmemtransfer-qwen35-q2b-from-qwen35-08b-20m | target_adaptor | Qwen/Qwen3.5-2B-Base | 2,000 | 11.882414 | transferred | 5.746621 | 0.08549 | 0.021403 | 0.999999 | 0.516338 | 0.500241 | 0.017135 | 0.319099 | 0.697629 | 0.247352 | OLAResearchX/xmemtransfer-qwen35-q2b-from-qwen35-08b-20m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-qwen35-08b-source-50m | null | null | null | null | XMemTransfer Qwen3.5-2B adaptor from Qwen3.5-0.8B source (20M) | 20m |
2,441 | xmemtransfer-qwen35-q4b-from-qwen35-08b-20m | target_adaptor | Qwen/Qwen3.5-4B-Base | 2,000 | 9.689974 | transferred | 7.585358 | 0.118958 | 0.013021 | 1 | 0.574809 | 0.57471 | 0.007468 | 0.378916 | 0.770952 | 0.247339 | OLAResearchX/xmemtransfer-qwen35-q4b-from-qwen35-08b-20m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-qwen35-08b-source-50m | null | null | null | null | XMemTransfer Qwen3.5-4B adaptor from Qwen3.5-0.8B source (20M) | 20m |
4,882 | xmemtransfer-qwen35-q9b-from-qwen35-08b-20m | target_adaptor | Qwen/Qwen3.5-9B-Base | 4,500 | 8.408332 | transferred | 16.670752 | 0.208723 | 0.008839 | 1 | 0.661597 | 0.696474 | 0.001067 | 0.477751 | 0.872361 | 0.246812 | OLAResearchX/xmemtransfer-qwen35-q9b-from-qwen35-08b-20m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-qwen35-08b-source-50m | null | null | null | null | XMemTransfer Qwen3.5-9B adaptor from Qwen3.5-0.8B source (20M) | 20m |
null | xmemtransfer-pythia160m-source-50m | source_memory | EleutherAI/pythia-160m | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-pythia160m-source-50m | null | false | null | null | null | null | null | null | null | XMemTransfer Pythia-160M source memory (50M) | 50m |
null | xmemtransfer-pythia160m-source-50m-word-boundary | source_memory | EleutherAI/pythia-160m | null | null | null | null | null | null | null | null | null | null | null | null | null | OLAResearchX/xmemtransfer-pythia160m-source-50m-word-boundary | null | false | null | null | null | null | null | null | null | XMemTransfer Pythia-160M source memory (50M, word-boundary) | 50m |
2,441 | xmemtransfer-pythia410m-from-pythia160m-20m | target_adaptor | EleutherAI/pythia-410m | 2,000 | 22.509401 | transferred | 0.129185 | 0.063797 | 0.016192 | 0.999998 | 0.52496 | 0.518675 | 0.008072 | 0.342414 | 0.698216 | 0.233045 | OLAResearchX/xmemtransfer-pythia410m-from-pythia160m-20m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-pythia160m-source-50m | null | null | null | null | XMemTransfer Pythia-410M adaptor from Pythia-160M source (20M) | 20m |
2,441 | xmemtransfer-tinyllama-from-pythia160m-20m | target_adaptor | TinyLlama/TinyLlama-1.1B-Chat-v1.0 | 2,000 | 10.313475 | transferred | 0.337679 | 0.048119 | 0.029612 | 1 | 0.503321 | 0.497756 | 0.001364 | 0.31499 | 0.686292 | 0.236764 | OLAResearchX/xmemtransfer-tinyllama-from-pythia160m-20m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-pythia160m-source-50m-word-boundary | null | null | null | null | XMemTransfer TinyLlama-1.1B adaptor from Pythia-160M source (20M) | 20m |
2,441 | xmemtransfer-qwen35-4b-from-pythia160m-20m | target_adaptor | Qwen/Qwen3.5-4B-Base | 2,000 | 10.313199 | transferred | 7.898183 | 0.083903 | 0.021362 | 1 | 0.538426 | 0.535894 | 0.001046 | 0.346512 | 0.722851 | 0.243454 | OLAResearchX/xmemtransfer-qwen35-4b-from-pythia160m-20m | 0 | true | 42 | true | OLAResearchX/xmemtransfer-pythia160m-source-50m | null | null | null | null | XMemTransfer Qwen3.5-4B adaptor from Pythia-160M source (20M) | 20m |
9,765 | xmemtransfer-mistral-self-r4-20m | target_adaptor | mistralai/Mistral-7B-v0.3 | 9,500 | 8.586427 | transferred | 3.814954 | 0.029002 | 0.029491 | 1 | 0.483802 | 0.476869 | 0.000303 | 0.302006 | 0.659309 | 0.227298 | OLAResearchX/xmemtransfer-mistral-self-r4-20m | 0 | true | 42 | true | null | null | null | null | null | XMemTransfer Mistral self-transfer adaptor (R4, 20M) | null |
XMemTransfer Results
This dataset is the machine-readable release manifest and result summary for
Cross-Model Memory Transfer via Target-Side Reader Adaptation. Each row
describes one released XMemTransfer source-memory or target-adaptor repository.
When a released artifact has a results.json, the row also includes its
validation-perplexity and gate-statistic summary.
- Paper: arXiv:2608.17050
- Project page: XMemTransfer
- Code: OLAResearch/XMemTransfer
- Models: XMemTransfer collection
Dataset contents
The train split contains 20 release-manifest rows. Twelve rows include result
summaries copied from the corresponding released model artifacts. Null metric
fields mean that the artifact is a source memory without a target-adaptor
result summary.
Important fields:
artifact_id,artifact_kind, andhf_repo: released artifact identity.base_modelandsource_memory_repo: model provenance.token_budget: target-side fitting budget for adaptors, or source-memory training budget for source artifacts.best_val_ppl, training-step fields, andgate_*: values recorded by the released experiment'sresults.json.
Scope and limitations
This is experiment and release metadata, not a language-model training corpus or a benchmark. It does not redistribute WikiText, evaluation examples, or model weights. Metrics are run summaries and should be interpreted with the experimental setup in the paper and repository. Local filesystem paths in the original run outputs are deliberately excluded.
Datasets used by the paper
The experiments use the following public corpora and evaluation datasets. They are listed here as references only; this release does not mirror their data:
- Training and intrinsic evaluation:
WikiText-103
(
wikitext-103-raw-v1), Wikipedia-2021 (December 2021 English Wikipedia, Mistral-tokenizer preprocessed release), LAMBADA, C4 (en), and FineWeb-Edu (sample-10BT). - Open-domain QA and transfer evaluation:
Natural Questions Open (NQ),
WebQuestions (WebQA),
TriviaQA
(the
rc.nocontextRC no-context subset; validation split), TruthfulQA (multiple_choice), and HotpotQA (distractor). - Multiple-choice and classification evaluation:
HellaSwag,
PIQA,
ARC (Easy and Challenge),
Winogrande
(
winogrande_xl), BoolQ, SuperGLUE RTE (rte), OpenBookQA, SciQ, and RACE (high).
See the paper's dataset and evaluation protocol for the exact splits and preprocessing. Wikipedia-2021 is the factual/encyclopedic corpus used for the QA memory-contribution experiments.
OLAResearchX/XMemTransfer-Results is a release-results and metadata index.
Its arxiv:2608.17050 tag is what makes it appear under Datasets citing this
paper on the paper page; that Hugging Face panel lists repositories that
declare a paper link, rather than automatically listing every external source
dataset referenced in this card.
License
Apache-2.0, matching the XMemTransfer code release.
- Downloads last month
- -