<DQ> Best Large Models
This is a collection of my current favorite 70/72B models. This is based on multiple benchmarks and real-world results. This list is subjective.
Updated • 139Note Finetune of miqu-70b-sf. Increased EQ-Bench to 84.89 points, beating an early version of GPT-4.
crestf411/daybreak-miqu-1-70b-v1.0-hf
Text Generation • Updated • 35 • 6
152334H/miqu-1-70b-sf
Text Generation • Updated • 11.6k • 218Note Dequantised version of the original Q5 miqu. Mistral medium leak.
sophosympatheia/Aurora-Nights-70B-v1.0
Text Generation • Updated • 3.28k • 16Note Creativity-focused finetune/blend containing multiple llama-based 70B models.
alpindale/WizardLM-2-8x22B
Text Generation • Updated • 7.27k • 318
wolfram/miquliz-120b-v2.0
Text Generation • Updated • 844 • 84Note 120b frankenmerge created by interleaving layers of miqu-1-70b-sf with lzlv_70b_fp16_hf (in an attempt to recreate goliath).
wolfram/miqu-1-120b
Text Generation • Updated • 769 • 47Note 120b frankenmerge of miqu-1-70b-sf with itself.
ChuckMcSneed/Gembo-v1-70b
Text Generation • Updated • 3.28k • 7Note Highest scoring 70B model on the NeoEvalPlusN benchmark, made via merging multiple 70B llama models.
ChuckMcSneed/WinterGoliath-123b
Text Generation • Updated • 901 • 7Note Merge of Xwin and WinterGoddess. Smarter than goliath through subjective testing.
sophosympatheia/Aurora-Nights-103B-v1.0
Text Generation • Updated • 5 • 11Note Self-merge of Aurora-Nights-70B-v1.0. Apparently a more "fun" version of Aurora-Nights.
CausalLM/34b-beta
Text Generation • Updated • 3.51k • 55meta-llama/Meta-Llama-3-70B-Instruct
Text Generation • Updated • 447k • 1.03kopenbmb/Eurux-8x22b-nca
Text Generation • Updated • 760 • 24