初始化项目,由ModelHub XC社区提供模型
Model: TRICELL-Inc/Progenitor.Pure-Virus-3.2-1B Source: Original Platform
This commit is contained in:
39
mergekit_config.yml
Normal file
39
mergekit_config.yml
Normal file
@@ -0,0 +1,39 @@
|
||||
|
||||
|
||||
merge_method: dare_ties
|
||||
dtype: float16
|
||||
out_dtype: float16
|
||||
|
||||
base_model: TRICELL-Inc/Progenitor.Pure-Virus-3.2-1B
|
||||
|
||||
models:
|
||||
- model: TRICELL-Inc/Progenitor.Pure-Virus-3.2-1B
|
||||
parameters:
|
||||
weight: 0.45
|
||||
density: 0.32
|
||||
- model: TRICELL-Inc/Progenitor.Pure-Virus-3.2-1B
|
||||
parameters:
|
||||
weight: 0.35
|
||||
density: 0.32
|
||||
|
||||
parameters:
|
||||
t: 0.25 # menos interpolación → más dominancia del base
|
||||
lambda: -0.62 # más negativo para matar cualquier alineamiento residual
|
||||
normalize: false
|
||||
rescale: true
|
||||
rescale_factor: 1.28 # subí un toque para amplificar el trash y degeneración
|
||||
memory_efficient: true
|
||||
low_cpu_mem_usage: true
|
||||
|
||||
layer_range:
|
||||
- value: [5, 22] # protejo más los embeddings y lm_head
|
||||
|
||||
tie_word_embeddings: true
|
||||
tie_output_embeddings: true
|
||||
|
||||
parameters:
|
||||
t: 0.40
|
||||
normalize: false
|
||||
rescale: true
|
||||
memory_efficient: true
|
||||
low_cpu_mem_usage: true
|
||||
Reference in New Issue
Block a user