datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
analytic-combinatorics-llm-samples-v1
analytic-combinatorics-llm-samples-v1 (PARTIAL)
Status: In progress — Llama-3.1-8B Phase 1 sampling on DGX Spark.
Current state
Rows: 2000 (of 15,000 target for Llama-3.1-8B)
Cells complete: open_qa T=0.1 (1K), open_qa T=0.7 (1K)
Truncation: 0%
Stochasticity: Verified — 995/1000 unique full outputs per cell
History
v1 (previous) was deleted due to a determinism bug in sample_vllm.py:
passing fixed seed=42 to vLLM produced 50 identical outputs per prompt.… See the full description on the dataset page: https://huggingface.co/datasets/latkes/analytic-combinatorics-llm-samples-v1.openr1_combinatorics_1k_smopenr1_combinatorics_1k_lgopenr1_combinatorics_1k_nm