Atlas

Models

← All models

StableLM Tuned Alpha 7B

Stability AI · 2023-04-19 · 7.9B parameters

StableLM Tuned Alpha 7B is Stability AI's instruction-tuned version of their StableLM Alpha 7B base model, fine-tuned on a curated mix of open-source instruction datasets to enable conversational and task-following behavior. Released in April 2023 as part of Stability AI's initial open-weight language-model release, it was among the first models from Stability AI's NLP research. The model uses a GPT-NeoX-style architecture and was released under a CC BY-NC-SA-4.0 license for research and non-commercial use.

Benchmark scores

BenchmarkScore
ARC AI227.0
Arena Text — Business, Management & Financial Ops (No Style Control)786
Arena Text — Creative Writing (No Style Control)868
Arena Text — English (No Style Control)890
Arena Text — Entertainment, Sports & Media (No Style Control)824
Arena Text — Exclude Ties (No Style Control)584
Arena Text — Exclude Ties (Style Controlled)701
Arena Text — Hard Prompts (No Style Control)835
Arena Text — Hard Prompts English (No Style Control)851
Arena Text — Instruction Following (No Style Control)832
Arena Text — Life, Physical & Social Science (No Style Control)804
Arena Text — Multi-turn (No Style Control)821
Arena Text — Non-English (No Style Control)814
Arena Text — Non-English (Style Controlled)887
Arena Text — Occupational Mathematical (No Style Control)835
Arena Text — Software & IT Services (No Style Control)859
Arena Text — Writing, Literature & Language (No Style Control)868
BoolQ59.0
Capability61.3
Epoch Capabilities Index (ECI)53.0
HellaSwag (Unspecified Scoring Protocol)40.7
LMArena Text Arena Coding Elo1004
LMArena Text Arena Creative Writing Elo932
LMArena Text Arena Elo952
LMArena Text Arena English Elo976
LMArena Text Arena Hard Prompts Elo942
LMArena Text Arena Hard Prompts English Elo957
LMArena Text Arena Industry: Writing, Literature & Language Elo930
LMArena Text Arena Instruction Following Elo909
LMArena Text Arena Math Elo890
Loading Atlas data…