QuixiAI
/

Qwen3-72B-Synthesis

Model card Files Files and versions

ehartford commited on Jun 11

Commit

fd45325

·

verified ·

1 Parent(s): 19b6339

Update README.md

Files changed (1) hide show

README.md +4 -4

README.md CHANGED Viewed

@@ -68,11 +68,11 @@ slices:
     - model: ./Qwen3-32B-Upscaled
       layer_range: [0, 32]
       parameters:
-        weight: 0.3
     - model: ./Qwen2.5-72B-Instruct-Aligned
       layer_range: [0, 32]
       parameters:
-        weight: 0.7
   # Slice 2: The "Knowledge Bridge" - transplant a pure block from the donor
   - merge_method: passthrough
@@ -86,11 +86,11 @@ slices:
     - model: ./Qwen3-32B-Upscaled
       layer_range: [32, 64]
       parameters:
-        weight: 0.3
     - model: ./Qwen2.5-72B-Instruct-Aligned
       layer_range: [48, 80]
       parameters:
-        weight: 0.7
 tokenizer_source: ./Qwen3-32B-Upscaled
 ```

     - model: ./Qwen3-32B-Upscaled
       layer_range: [0, 32]
       parameters:
+        weight: 0.5
     - model: ./Qwen2.5-72B-Instruct-Aligned
       layer_range: [0, 32]
       parameters:
+        weight: 0.5
   # Slice 2: The "Knowledge Bridge" - transplant a pure block from the donor
   - merge_method: passthrough
     - model: ./Qwen3-32B-Upscaled
       layer_range: [32, 64]
       parameters:
+        weight: 0.5
     - model: ./Qwen2.5-72B-Instruct-Aligned
       layer_range: [48, 80]
       parameters:
+        weight: 0.5
 tokenizer_source: ./Qwen3-32B-Upscaled
 ```