L3-Tyche-8B-v1.0

L3-Tyche-8B-v1.0

About:

This is a merge of pre-trained language models created using mergekit.

Recommended Samplers:

Temperature - 1.3
TFS - 0.96
Smoothing Factor - 0.3
Smoothing Curve - 1.1
Repetition Penalty - 1.08

Merge Method

This model was merged a series of model stock and lora merges, followed by ExPO and an attention swap. It uses a mix of smart and roleplay centered models to improve performance.

Configuration

The following YAML configuration was used to produce this model:

---
models:
  - model: Nitral-AI/Hathor_Tahsin-L3-8B-v0.85
  - model: Nitral-AI/Hathor_Respawn-L3-8B-v0.8
  - model: ChaoticNeutrals/Hathor_RP-v.01-L3-8B
  - model: Sao10K/L3-8B-Stheno-v3.2
  - model: yodayo-ai/nephra_v1.0
  - model: HiroseKoichi/L3-8B-Lunar-Stheno
  - model: Jellywibble/lora_120k_pref_data_ep2
  - model: Jellywibble/qlora_120k_pref_data_ep1
  - model: Jellywibble/meseca-20062024-c1
  - model: Hastagaras/Jamet-8B-L3-MK.V-Blackroot
  - model: Cas-Warehouse/Llama-3-SOVL-MopeyMule-Blackroot-8B
  - model: ResplendentAI/Nymph_8B+Azazelle/RP_Format_QuoteAsterisk_Llama3
  - model: R136a1/Bungo-L3-8B
  - model: maldv/badger-mu-llama-3-8b
  - model: TheDrummer/Llama-3SOME-8B-v2
  - model: Magpie-Align/Llama-3-8B-Magpie-Align-v0.1+Azazelle/Llama3_RP_ORPO_LoRA
  - model: grimjim/Llama-3-Instruct-8B-SPPO-Iter3-SimPO-merge+Azazelle/Llama-3-8B-Abomination-LORA
  - model: NousResearch/Hermes-2-Pro-Llama-3-8B+mpasila/Llama-3-Instruct-LiPPA-LoRA-8B
  - model: MaziyarPanahi/Llama-3-8B-Instruct-v0.8+Azazelle/Llama-3-Sunfall-8b-lora
  - model: openchat/openchat-3.6-8b-20240522+Azazelle/BlueMoon_Llama3
  - model: collaiborateorg/Collaiborator-MEDLLM-Llama-3-8B-v2+Azazelle/llama3-8b-hikikomori-v0.4
  - model: grimjim/Llama-3-Instruct-8B-SPPO-Iter3-SimPO-merge+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
merge_method: model_stock
base_model: failspy/Meta-Llama-3-8B-Instruct-abliterated-v3
dtype: float32
vocab_type: bpe
name: hq_rp

---
# ExPO
models:
  - model: hq_rp
    parameters:
      weight: 1.3
merge_method: task_arithmetic
base_model: failspy/Meta-Llama-3-8B-Instruct-abliterated-v3
parameters:
  normalize: false
dtype: float32
vocab_type: bpe
name: pre

---
# Attention Donor
models:
  - model: Nitral-AI/Hathor_Tahsin-L3-8B-v0.85
  - model: Sao10K/L3-8B-Stheno-v3.2
merge_method: model_stock
base_model: failspy/Meta-Llama-3-8B-Instruct-abliterated-v3
dtype: float32
vocab_type: bpe
name: donor

---
# Attention swap?
models:
  - model: pre
merge_method: slerp
base_model: donor
parameters:
  t:
    - filter: mlp
      value: 0
    - value: 1
dtype: float32
vocab_type: bpe