DeepSeek

7 models7 ranked conversions30.3K downloads

Best conversion per model

One row per base model. The full list of attempts is on each model page.

#	Model	Author	Params	KL	Refusals	Score	Evidence	Downloads
1	DeepSeek-V2-Lite deepseek-ai	aeon37	15.7B	0.0089	5/1304%	97.6	BestParetoSAFETENSORS	100
2	DeepSeek-Coder-V2-Lite-Base deepseek-ai	hereticsystems	15.7B	0.027	5/1005%	96.2	BestParetoSAFETENSORS	72
3	deepseek-llm-7b-chat deepseek-ai	richardyoung	7B	0.043	16/10016%	89.9	BestParetoGGUFSAFETENSORS	15.8K
4	deepseek-coder-33b-instruct deepseek-ai	richardyoung	33B	0.020	70/10070%	64.0	BestParetoGGUFSAFETENSORS	1.4K

All models

DeepSeek-V2-Lite15.7B DeepSeek-Coder-V2-Lite-Base15.7B deepseek-llm-7b-chat7B deepseek-coder-33b-instruct33B DeepSeek-V4-Flash-0731305.5B DeepSeek-V4.1-Flash-1017763.2B LFM2.5-1.2B-Thinking-Deepseek-3.1-Heretic-Uncensored-DISTILL1.2B

Unmeasured models

DeepSeek-V4-Flash-0731, DeepSeek-V4.1-Flash-1017, LFM2.5-1.2B-Thinking-Deepseek-3.1-Heretic-Uncensored-DISTILL — conversions exist but their publishers did not report KL and refusal numbers, so they are not ranked.