llada2
Datasets
All datasets matching “llada2”distill_llada2_sft
distill_llada2_sft — Pre-tokenized SFT mixture for LLaDA2-teacher distillation
Pre-tokenized SFT corpus used to train every checkpoint in the Cross-Tokenizer (Pipeline A) of the TIDE framework — i.e. the distill-LLaDA2-* student checkpoints distilled from inclusionAI/LLaDA2.0-mini.
The dataset ships as a datasets.DatasetDict (load_from_disk-ready) so distillation training never has to re-tokenize at job start (which would cause NCCL timeouts on multi-node runs).… See the full description on the dataset page: https://huggingface.co/datasets/TIDE-dllm/distill_llada2_sft.Llada2_EB_Decode_512_sequential_datasetllada2-mini-uq-pickles
LLaDA 2.1 mini — UQ eval pickles (ue_manager_seed1)
Durable backup of the LLaDA 2.1 mini uncertainty-quantification sweep pickles
produced with lm-polygraph (branch feat/llada2-cache, April–May 2026).
Each .pkl is a torch.load-able ue_manager dump containing per-sample
stats (greedy_texts, target_texts, ...), estimations, gen_metrics, and
metrics.
Layout
setup_a/ no-train baselines, reduced config (~34 est, K=10 sampling dropped)
setup_b/ with-train baselines… See the full description on the dataset page: https://huggingface.co/datasets/karantonis/llada2-mini-uq-pickles.dllm-attention-parents-llada2-finecode-pred-debias5-b64-v1
DART FineCode deterministic 1/8 sample — inclusionAI/LLaDA2.0-mini dependency sidecar
This is the sealed attention dependency release generated
with pred-debias5 corrected attention. It is compatible with block size 64 and pairs only
with zimplex/dllm-finecode-dagcover-full-v1,
whose manifest SHA-256 is 45e250f9f4a5302ef1bb04c16577eb8b6d85996368fe1dc359b86465cdb03176.
The sidecar manifest SHA-256 is f4dbab97f495e23682f26ac085b19a90e9e6a15cf664b789d9efb0f23566cd3e. The source… See the full description on the dataset page: https://huggingface.co/datasets/zimplex/dllm-attention-parents-llada2-finecode-pred-debias5-b64-v1.dllm-effect-parents-llada2-finemath-half-d1marginal-w128-tau2-b64-v3
DART FineMath-4+ half — inclusionAI/LLaDA2.0-mini dependency sidecar
This is the sealed counterfactual dependency release generated
with corrected d1-marginal, window 128, tau 2. It is compatible with block size 64 and pairs only
with zimplex/dllm-finemath-dagcover-half-v1,
whose manifest SHA-256 is e20f3ecdc7417e8f8141610d4e1c55eb2973d269d03117d47c4e27b3d1182a68.
The sidecar manifest SHA-256 is fbf0371aeb9de796c5b8d22730e91c4254faef1354151ac797dc20671a2a848a. The source
release… See the full description on the dataset page: https://huggingface.co/datasets/zimplex/dllm-effect-parents-llada2-finemath-half-d1marginal-w128-tau2-b64-v3.dllm-effect-parents-llada2-finecode-d1marginal-w128-tau2-b64-v3
DART FineCode deterministic 1/8 sample — inclusionAI/LLaDA2.0-mini dependency sidecar
This is the sealed counterfactual dependency release generated
with corrected d1-marginal, window 128, tau 2. It is compatible with block size 64 and pairs only
with zimplex/dllm-finecode-dagcover-full-v1,
whose manifest SHA-256 is 45e250f9f4a5302ef1bb04c16577eb8b6d85996368fe1dc359b86465cdb03176.
The sidecar manifest SHA-256 is 391fa0f852451457ba2812a8633af7799f18193e5e0d2a489432ccc7c3d67c43.… See the full description on the dataset page: https://huggingface.co/datasets/zimplex/dllm-effect-parents-llada2-finecode-d1marginal-w128-tau2-b64-v3.