morikomorizz commited on
Commit
40829db
·
verified ·
1 Parent(s): 54c0665

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -0
README.md CHANGED
@@ -25,6 +25,11 @@ This repository contains the **GGUF** quantized files for **[zai-org/GLM-5.2](ht
25
  **Note:**
26
  The frontmost FFN layers in the first block are all set to Q8_0, while the remaining tensors remain in BF16 and FP32 formats.
27
 
 
 
 
 
 
28
  ---
29
 
30
  **Prompt used:**
 
25
  **Note:**
26
  The frontmost FFN layers in the first block are all set to Q8_0, while the remaining tensors remain in BF16 and FP32 formats.
27
 
28
+ > **⚠️ Field Test Notice: The "Garbage In, Garbage Out" Hypothesis**
29
+ > Please note that the aggressive quantization applied to the very first layers in these custom builds is an intentional experiment. The core hypothesis we are testing is **"Garbage In, Garbage Out"**: if the frontmost layers (which capture and process the initial input context) are heavily compressed, does it inevitably trap the model with degraded inputs and result in poor final generations?
30
+ >
31
+ > **Call for Feedback:** If you are willing to test these models, please share your performance benchmarks, perplexity scores, or qualitative observations in the Community Discussion or Issues tab. Your field reports are crucial to help determine if early-layer degradation is a fatal flaw or a manageable trade-off in MoE architectures.
32
+
33
  ---
34
 
35
  **Prompt used:**