Atlas

Models

← All models

Dolly 2.0 12B

Databricks · 2023-04-12 · 11.8B parameters

Dolly 2.0 is Databricks' 12B instruction-following causal language model, released in April 2023 and derived from EleutherAI's Pythia-12B. It was fine-tuned exclusively on approximately 15,000 instruction/response records generated by Databricks employees; Databricks released the model weights, training code, and databricks-dolly-15k dataset for commercial use without relying on ChatGPT-generated instruction-tuning data.

Benchmark scores

BenchmarkScore
ARC AI239.6
Arena Text — Business, Management & Financial Ops (No Style Control)734
Arena Text — Chinese (No Style Control)835
Arena Text — Creative Writing (No Style Control)864
Arena Text — English (No Style Control)867
Arena Text — Entertainment, Sports & Media (No Style Control)810
Arena Text — Exclude Ties (No Style Control)568
Arena Text — Exclude Ties (Style Controlled)749
Arena Text — Hard Prompts (No Style Control)804
Arena Text — Hard Prompts English (No Style Control)811
Arena Text — Instruction Following (No Style Control)814
Arena Text — Life, Physical & Social Science (No Style Control)801
Arena Text — Multi-turn (No Style Control)740
Arena Text — Non-English (No Style Control)835
Arena Text — Non-English (Style Controlled)966
Arena Text — Occupational Mathematical (No Style Control)840
Arena Text — Software & IT Services (No Style Control)790
Arena Text — Writing, Literature & Language (No Style Control)872
BBH (Open LLM Leaderboard v2)33.2
BoolQ56.3
Capability74.9
Epoch Capabilities Index (ECI)88.0
GPQA (Open LLM Leaderboard v2)24.1
HellaSwag (Unspecified Scoring Protocol)70.8
IFEval (Open LLM Leaderboard v2)23.6
LMArena Text Arena Chinese Elo997
LMArena Text Arena Coding Elo962
LMArena Text Arena Creative Writing Elo969
LMArena Text Arena Elo981
LMArena Text Arena English Elo994
Loading Atlas data…