Hugging Face's logo

v000000
/
L3.1-Celestial-Stone-2x8B

Text Generation
Transformers
Safetensors
mixtral
llama
Merge
llama3
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
L3.1-Celestial-Stone-2x8B / README.md
v000000's picture
v000000
Update README.md
7ef75e6 verified
preview code
|
raw
history blame
3.97 kB
metadata
base_model:
  - nothingiisreal/L3.1-8B-Celeste-V1.5
  - Sao10K/Llama-3.1-8B-Stheno-v3.4
  - Sao10K/L3.1-8B-Niitama-v1.1
  - arcee-ai/Llama-3.1-SuperNova-Lite
  - akjindal53244/Llama-3.1-Storm-8B
  - arcee-ai/Llama-Spark
  - grimjim/Llama-3-Instruct-abliteration-LoRA-8B
  - crestf411/sunfall-peft
tags:
  - llama
  - merge
  - llama3
  - mixtral
library_name: transformers

Llama-3.1-Celestial-Stone-2x8B (BF16)

  • Mixture of Experts (14B).

image/png

Both experts are used in tandem when generating a token.

The first expert is Instruct 405B distillation/RP vector merge (Supernova-Lite, Niitama1.1, Storm)

The second expert is ERP/Reddit data merge (Celeste1.5, Stheno3.4, Storm)

The base model is Sao10k/L3.1-Stheno-3.4 with the Sunfall LoRa 0.6.1 to make it understand SillyTavern prompts and storywriting better.

Prompt Template: 

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>

{input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

{output}<|eot_id|>

Recipe (I'm sorry...): 

slices:
  - sources:
      - model: Sao10K/L3.1-8B-Niitama-v1.1+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
        layer_range: [0, 32]
      - model: akjindal53244/Llama-3.1-Storm-8B
        layer_range: [0, 32]
merge_method: nearswap
base_model: Sao10K/L3.1-8B-Niitama-v1.1+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
parameters:
  t:
    - value: 0.0001
dtype: bfloat16
out_type: float16 #oops
slices:
  - sources:
      - model: v000000/Llama-3.1-8B-Stheno-v3.4-abliterated
        layer_range: [0, 32]
      - model: akjindal53244/Llama-3.1-Storm-8B
        layer_range: [0, 32]
merge_method: slerp
base_model: v000000/Llama-3.1-8B-Stheno-v3.4-abliterated
parameters:
  t:
    - filter: self_attn
      value: [0.1, 0.6, 0.3, 0.8, 0.5]
    - filter: mlp
      value: [0.9, 0.4, 0.7, 0.2, 0.5]
    - value: 0.5
dtype: float32
models:
  - model: arcee-ai/Llama-3.1-SuperNova-Lite
    parameters:
      weight: 1.0
  - model: v000000/L3.1-Niitorm-8B-t0.0001
    parameters:
      weight: 0.4
merge_method: task_arithmetic
base_model: arcee-ai/Llama-3.1-SuperNova-Lite
parameters:
    normalize: false
dtype: float16
models:
  - model: arcee-ai/Llama-3.1-SuperNova-Lite
    parameters:
      weight: 0.0
  - model: v000000/L3.1-Niitorm-8B-t0.0001
    parameters:
      weight: 1.25
merge_method: task_arithmetic
base_model: arcee-ai/Llama-3.1-SuperNova-Lite
parameters:
    normalize: false
dtype: float16
models:
  - model: v000000/L3.1-8B-RP-Test-003-Task_Arithmetic
merge_method: slerp
base_model: v000000/L3.1-8B-RP-Test-002-Task_Arithmetic+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
# This model needed some abliteration^
parameters:
  t:
    - value: [0, 0, 0.3, 0.4, 0.5, 0.6, 0.5, 0.4, 0.3, 0, 0]
dtype: float16
base_model: nothingiisreal/L3.1-8B-Celeste-V1.5+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
dtype: bfloat16
merge_method: task_arithmetic
parameters:
  normalize: false
slices:
- sources:
  - layer_range: [0, 32]
    model: nothingiisreal/L3.1-8B-Celeste-V1.5+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
    parameters:
      weight: 0.7
  - layer_range: [0, 32]
    model: v000000/L3.1-Sthenorm-8B
    parameters:
      weight: 0.2
  - layer_range: [0, 32]
    model: nothingiisreal/L3.1-8B-Celeste-V1.5
    parameters:
      weight: 0.2
base_model: crestf411/L3.1-8B-sunfall-stheno-v0.6.1
experts_per_token: 2
local_experts: 2
gate_mode: random
dtype: bfloat16
experts:
  - source_model: v000000/L3.1-Storniitova-8B
  - source_model: x0000001/l3.1-part_aaa