Views
No views yet
1
2models:
3 - model: tokyotech-llm/Swallow-7b-instruct-hf
4 # no parameters necessary for base model
5 - model: codellama/CodeLlama-7b-Python-hf # follow user intent
6 parameters:
7 density: 1
8 weight:
9 - filter: mlp.down_proj
10 value: [0.3, 0.25, 0.25, 0.15, 0.1]
11 - filter: mlp.gate_proj
12 value: [0.7, 0.25, 0.5, 0.45, 0.4]
13 - filter: mlp.up_proj
14 value: [0.7, 0.25, 0.5, 0.45, 0.4]
15 - filter: self_attn
16 value: [0.7, 0.25, 0.5, 0.45, 0.4]
17 - value: 0 # fallback for rest of tensors.
18merge_method: dare_ties
19base_model: tokyotech-llm/Swallow-7b-instruct-hf
20dtype: bfloat16
21tokenizer_source: union
22