Modelpublic

Casual-Autopsy/L3-Uncen-Merger-Omelette-RP-v0.1.1-8B

sourceHugging Facellama3updated 2y agoView on Hugging Face

4likes16downloads

Model Card

L3-Uncen-Merger-Omelette-RP-v0.1-8B

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

Merge Details

A merger recipe inspired by invisietch/EtherealRainbow-v0.3-8B combined with a merger technique known as merge densification( grimjim/kunoichi-lemon-royale-v3-32K-7B )

The model recipe ended up being something I can only describe as making an omelette. Hence the model name.

The models are scrambled with Dare Ties to induce a bit of randomness, then the Dare Ties merges are merged into themselves with SLERP to repair any holes cause by Dare Ties, and finally a bunch of high creativity models are thrown into the merger through merge densification(Task Arithmetic).

This model uses a bunch of the top models of the UGI Leaderboard, I picked out a few of the top 8B models of each category. Most of the high creativity models in the last step were found through Lewdiculus' account uploads

Downgraded to Stheno v3.2 due to issues with the model

Merge Method

Dare Ties, SLERP, and Task Arithmetic

Models Merged

The following models were included in the merge:

Quants

Static quants by mradermacher

Secret Sauce

The following YAML configurations were used to produce this model:

Scrambled-Egg-1

yaml

models:
  - model: Sao10K/L3-8B-Stheno-v3.2
  - model: Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
    parameters:
      density: 0.45
      weight: 0.33
  - model: bluuwhale/L3-SthenoMaidBlackroot-8B-V1
    parameters:
      density: 0.75
      weight: 0.33
merge_method: dare_ties
base_model: Sao10K/L3-8B-Stheno-v3.2
parameters:
  int8_mask: true
dtype: bfloat16

Scrambled-Egg-2

yaml

models:
  - model: [Unreleased psychology model]
  - model: migtissera/Llama-3-8B-Synthia-v3.5
    parameters:
      density: 0.35
      weight: 0.25
  - model: tannedbum/L3-Nymeria-Maid-8B
    parameters:
      density: 0.65
      weight: 0.25
merge_method: dare_ties
base_model: [Unreleased psychology model]
parameters:
  int8_mask: true
dtype: bfloat16

Scrambled-Egg-3

yaml

models:
  - model: Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
  - model: tannedbum/L3-Nymeria-8B
    parameters:
      density: 0.5
      weight: 0.35
  - model: ChaoticNeutrals/Hathor_RP-v.01-L3-8B
    parameters:
      density: 0.4
      weight: 0.2
merge_method: dare_ties
base_model: Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
parameters:
  int8_mask: true
dtype: bfloat16

Omelette-1

yaml

models:
  - model: Casual-Autopsy/Scrambled-Egg-1
  - model: Casual-Autopsy/Scrambled-Egg-3
merge_method: slerp
base_model: Casual-Autopsy/Scrambled-Egg-1
parameters:
  t:
    - value: [0.1, 0.15, 0.2, 0.4, 0.6, 0.4, 0.2, 0.15, 0.1]
  embed_slerp: true
dtype: bfloat16

Omelette-2

yaml

models:
  - model: Casual-Autopsy/Omelette-1
  - model: Casual-Autopsy/Scrambled-Egg-2
merge_method: slerp
base_model: Casual-Autopsy/Omelette-1
parameters:
  t:
    - value: [0.7, 0.5, 0.3, 0.25, 0.2, 0.25, 0.3, 0.5, 0.7]
  embed_slerp: true
dtype: bfloat16

L3-Uncen-Merger-Omelette-RP-v0.1-8B

yaml

models:
  - model: Casual-Autopsy/Omelette-2
  - model: ResplendentAI/Nymph_8B
    parameters:
      weight: 0.01
  - model: ChaoticNeutrals/Hathor_RP-v.01-L3-8B
    parameters:
      weight: 0.01
  - model: Sao10K/L3-8B-Stheno-v3.1
    parameters:
      weight: 0.015
  - model: aifeifei798/llama3-8B-DarkIdol-1.0
    parameters:
      weight: 0.015
  - model: cgato/L3-TheSpice-8b-v0.8.3
    parameters:
      weight: 0.02
  - model: ChaoticNeutrals/Poppy_Porpoise-1.4-L3-8B
    parameters:
      weight: 0.02
merge_method: task_arithmetic
base_model: Casual-Autopsy/Omelette-2
dtype: bfloat16