README.md

metadata

license: llama3
tags:
  - moe
language:
  - en

Exllamav2 quant (exl2 / 3.5 bpw) made with ExLlamaV2 v0.0.21

Other EXL2 quants:

Quant	Model Size	lm_head
2.2	7777 MB	6
2.5	8519 MB	6
3.0	9944 MB	6
3.5	11365 MB	6
3.75	12080 MB	6
4.0	12789 MB	6
4.25	13503 MB	6
5.0	15632 MB	6
6.0	18594 MB	8
6.5	19969 MB	8
8.0	24115 MB	8

(Maybe i'll change the waifu picture later)

GGUF/Exl2 quants

Experimental RP-oriented MoE, the idea was to get a model that would be equal to or better than Mixtral 8x7B and it's finetunes in RP/ERP tasks.

Llama 3 SnowStorm 4x8B

base_model: NeverSleep_Llama-3-Lumimaid-8B-v0.1-OAS
gate_mode: random
dtype: bfloat16
experts_per_token: 2
experts:
  - source_model: ChaoticNeutrals_Poppy_Porpoise-v0.7-L3-8B
  - source_model: NeverSleep_Llama-3-Lumimaid-8B-v0.1-OAS
  - source_model: openlynn_Llama-3-Soliloquy-8B-v2
  - source_model: Sao10K_L3-8B-Stheno-v3.1

Models used

• ChaoticNeutrals/Poppy_Porpoise-v0.7-L3-8B
• NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
• openlynn/Llama-3-Soliloquy-8B-v2
• Sao10K/L3-8B-Stheno-v3.1

Difference(from ChaoticSoliloquy v1.5)

• Update from NeverSleep/Llama-3-Lumimaid-8B-v0.1 to NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
• Update from openlynn/Llama-3-Soliloquy-8B-v1 to openlynn/Llama-3-Soliloquy-8B-v2
• Update from Sao10K/L3-Solana-8B-v1 to Sao10K/L3-8B-Stheno-v3.1

Vision

llama3_mmproj

Prompt format: Llama 3