wbit5 commited on
Commit
dedade8
Β·
verified Β·
1 Parent(s): 5847998

publish from LatentPortPort 7bfa698

Browse files
Files changed (1) hide show
  1. README.md +12 -0
README.md CHANGED
@@ -82,6 +82,18 @@ Qwen3-4B reads; Qwen3-8B answers, across all 36 layers.
82
  | 🎯 **Held-out questions** | **7 of 7**, the same as each model alone |
83
  | ⚑ **Time to first token** | **1.12Γ—** sooner (95% CI 1.12–1.12) on two RTX 3090s at ~256 tokens; ceiling 1.61Γ— |
84
 
 
 
 
 
 
 
 
 
 
 
 
 
85
  ### πŸͺœ A 13.6Γ— gap, at the bar β€” [`qwen3-0.6b-to-8b-bridge`](https://huggingface.co/agentic-underground/qwen3-0.6b-to-8b-bridge)
86
 
87
  Qwen3-0.6B reads; Qwen3-8B, 13.6 times larger, answers.
 
82
  | 🎯 **Held-out questions** | **7 of 7**, the same as each model alone |
83
  | ⚑ **Time to first token** | **1.12Γ—** sooner (95% CI 1.12–1.12) on two RTX 3090s at ~256 tokens; ceiling 1.61Γ— |
84
 
85
+ ### πŸš€ Fifty-four times larger β€” [`qwen3-0.6b-to-32b-bridge`](https://huggingface.co/agentic-underground/qwen3-0.6b-to-32b-bridge)
86
+
87
+ Qwen3-0.6B reads; Qwen3-32B, quantized to Q4K, answers. Published with its miss stated, while a
88
+ refit is under way.
89
+
90
+ | | |
91
+ |:--|:--|
92
+ | ⚑ **Time to first token** | **4.21Γ—** sooner (95% CI 4.19–4.23) at ~256 tokens, **7.17Γ—** at ~1K, on two RTX 3090s; ceilings 11.60Γ— and 14.33Γ— |
93
+ | 🎯 **Held-out questions** | **6 of 7** β€” the 32B alone gets 7, so it is **below our answer bar** |
94
+ | πŸ” **Fidelity (NCR)** | **0.866** with the paper's learned correction (0.794 from the maps alone) |
95
+ | πŸ§ͺ **Control (wrong donor)** | 0.197 |
96
+
97
  ### πŸͺœ A 13.6Γ— gap, at the bar β€” [`qwen3-0.6b-to-8b-bridge`](https://huggingface.co/agentic-underground/qwen3-0.6b-to-8b-bridge)
98
 
99
  Qwen3-0.6B reads; Qwen3-8B, 13.6 times larger, answers.