alphabell-research commited on
Commit
9df33e3
路
verified 路
1 Parent(s): de87d4a

Make Nanbeige architecture map discoverable in Space

Browse files
Files changed (1) hide show
  1. README.md +8 -2
README.md CHANGED
@@ -7,7 +7,7 @@ sdk: static
7
  app_file: index.html
8
  pinned: true
9
  license: mit
10
- short_description: "Real Flint and Qwen Q8 captures, replayed token by token"
11
  thumbnail: https://huggingface.co/spaces/alphabell/Model-Observatory/resolve/main/cover.png
12
  models:
13
  - RecursiveSI/flint
@@ -22,6 +22,8 @@ tags:
22
  - recurrent-memory
23
  - qwen
24
  - flint
 
 
25
  - educational
26
  ---
27
 
@@ -31,10 +33,14 @@ tags:
31
 
32
  Explore two different architectures with real Q8_0 tensor captures: Flint's sparse expert routing and alphaBell Qwen Open's recurrent memory. Open a block, head or expert, then step through the tokens that used it.
33
 
34
- [**Open the explorer**](https://huggingface.co/spaces/alphabell/Model-Observatory) 路 [Qwen view](https://alphabell-model-observatory.static.hf.space/?model=qwen) 路 [Source](https://huggingface.co/spaces/alphabell/Model-Observatory/tree/main) 路 [alphaBell](https://alphabell.com)
35
 
36
  ![Model Observatory showing Flint attention heads and selected experts for one captured token.](preview.png)
37
 
 
 
 
 
38
  ## Five measured experiments
39
 
40
  [**Open all five experiments**](https://alphabell-model-observatory.static.hf.space/experiments/index.html)
 
7
  app_file: index.html
8
  pinned: true
9
  license: mit
10
+ short_description: "Flint/Qwen traces and Nanbeige/Qwen architecture"
11
  thumbnail: https://huggingface.co/spaces/alphabell/Model-Observatory/resolve/main/cover.png
12
  models:
13
  - RecursiveSI/flint
 
22
  - recurrent-memory
23
  - qwen
24
  - flint
25
+ - nanbeige
26
+ - architecture
27
  - educational
28
  ---
29
 
 
33
 
34
  Explore two different architectures with real Q8_0 tensor captures: Flint's sparse expert routing and alphaBell Qwen Open's recurrent memory. Open a block, head or expert, then step through the tokens that used it.
35
 
36
+ [**Open the explorer**](https://huggingface.co/spaces/alphabell/Model-Observatory) 路 [Nanbeige and Qwen architecture](https://alphabell-model-observatory.static.hf.space/architecture/index.html) 路 [Qwen view](https://alphabell-model-observatory.static.hf.space/?model=qwen) 路 [Source](https://huggingface.co/spaces/alphabell/Model-Observatory/tree/main) 路 [alphaBell](https://alphabell.com)
37
 
38
  ![Model Observatory showing Flint attention heads and selected experts for one captured token.](preview.png)
39
 
40
+ ## Nanbeige and Qwen architecture
41
+
42
+ [**Explore the layer map**](https://alphabell-model-observatory.static.hf.space/architecture/index.html) to compare Nanbeige4.3-33B's 52 full-attention layers with the Qwen3.8-27B text decoder's 48 recurrent linear and 16 full-attention layers. This comparison uses published configurations; the measured captures in this Space are Flint and Qwen.
43
+
44
  ## Five measured experiments
45
 
46
  [**Open all five experiments**](https://alphabell-model-observatory.static.hf.space/experiments/index.html)