add bpw with embedding included

Author: pkmandke

models/qwen2/README.md

@@ -79,9 +79,9 @@ Perplexity score on the [`WikiText-2`](https://huggingface.co/datasets/EleutherA
 | --------------------- | ----------------------------------------- | --------------------- | -------- | ---------------- |
 | Qwen2.5 1.5B Instruct | none (`float16`)                          | 16.00                 | macOS    | 12.21            |
 | Qwen2.5 1.5B Instruct | [4-bit quantized][p-4bit]                 | 4.50                  | macOS    | 14.79            |
-| Qwen2.5 1.5B Instruct | none (`float16`)                          | 16.00\*               | iOS      | 12.21            |
-| Qwen2.5 1.5B Instruct | [4-bit palettized (group size 8)][p-4bit] | 4.02\*                | iOS      | 14.64            |
+| Qwen2.5 1.5B Instruct | none (`float16`)                          | 16.00                 | iOS      | 12.21            |
+| Qwen2.5 1.5B Instruct | [4-bit palettized (group size 8)][p-4bit] | 4.63\*                | iOS      | 14.64            |
 
-\* BPW excludes the Embedding which is quantized to INT8 per-tensor.
+\* BPW includes the Embedding which is quantized to INT8 per-tensor.
 
 [p-4bit]: ../README.md#quantization-options

models/qwen3/README.md

@@ -79,16 +79,16 @@ Perplexity score on the [`WikiText-2`](https://huggingface.co/datasets/EleutherA
 
 | Model      | Compression                                          | Bits Per Weight (BPW) | Platform | Perplexity Score |
 | ---------- | ---------------------------------------------------- | --------------------- | -------- | ---------------- |
-| Qwen3 0.6B | none (`float16`)                                     | 16.00\*               | iOS      | 26.16            |
-| Qwen3 0.6B | [Mixed 4-bit/8-bit palettized][mixed-4bit-8bit-yaml] | 4.88\*                | iOS      | 30.90            |
+| Qwen3 0.6B | none (`float16`)                                     | 16.00                 | iOS      | 26.16            |
+| Qwen3 0.6B | [Mixed 4-bit/8-bit palettized][mixed-4bit-8bit-yaml] | 5.71\*                | iOS      | 30.90            |
 | Qwen3 4B   | none (`float16`)                                     | 16.00                 | macOS    | 16.41            |
 | Qwen3 4B   | [4-bit quantized][presets-info]                      | 4.50                  | macOS    | 18.33            |
-| Qwen3 4B   | none (`float16`)                                     | 16.00\*               | iOS      | 16.41            |
-| Qwen3 4B   | [Mixed 4-bit/8-bit palettized][qwen3-4b-mixed-yaml]  | 4.56\*                | iOS      | 18.80            |
+| Qwen3 4B   | none (`float16`)                                     | 16.00                 | iOS      | 16.41            |
+| Qwen3 4B   | [Mixed 4-bit/8-bit palettized][qwen3-4b-mixed-yaml]  | 4.89\*                | iOS      | 18.80            |
 | Qwen3 8B   | none (`float16`)                                     | 16.00                 | macOS    | 12.19            |
 | Qwen3 8B   | [4-bit quantized][presets-info]                      | 4.50                  | macOS    | 12.90            |
 
-\* BPW excludes the Embedding which is quantized to INT8 per-tensor.
+\* BPW includes the Embedding which is quantized to INT8 per-tensor.
 
 [presets-info]: ../README.md#quantization-options
 [mixed-4bit-8bit-yaml]: qwen3_0_6b_mixed_4bit_8bit.yaml