Update README.md
Browse files
README.md
CHANGED
|
@@ -27,6 +27,31 @@ distilgpt2, available in as many GGUF quantization levels as possible as of Marc
|
|
| 27 |
|
| 28 |
Importance matrix was used with context set to 1,024, created with [Kalomaze's "groups_merged.txt"](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384).
|
| 29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
Original model card below.
|
| 31 |
***
|
| 32 |
# DistilGPT2
|
|
|
|
| 27 |
|
| 28 |
Importance matrix was used with context set to 1,024, created with [Kalomaze's "groups_merged.txt"](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384).
|
| 29 |
|
| 30 |
+
|Quantization|HellaSwag|
|
| 31 |
+
|--------|--------|
|
| 32 |
+
|IQ1_S |25.3% |
|
| 33 |
+
|IQ2_XXS |27.3% |
|
| 34 |
+
|IQ2_XS |27.5% |
|
| 35 |
+
|IQ2_S |26.6% |
|
| 36 |
+
|IQ2_M |27.4% |
|
| 37 |
+
|Q2_K_S |24.4% |
|
| 38 |
+
|Q2_K |27.3% |
|
| 39 |
+
|IQ3_XXS |27.1% |
|
| 40 |
+
|IQ3_XS |26.6% |
|
| 41 |
+
|IQ3_S |26.2% |
|
| 42 |
+
|IQ3_M |26.2% |
|
| 43 |
+
|Q3_K_S |27.8% |
|
| 44 |
+
|Q3_K_M |26.2% |
|
| 45 |
+
|IQ4_XS |27.1% |
|
| 46 |
+
|IQ4_NL |27.1% |
|
| 47 |
+
|Q4_0 |26.3% |
|
| 48 |
+
|Q4_K_M |26.8% |
|
| 49 |
+
|Q5_0 |27.2% |
|
| 50 |
+
|Q5_1 |26.7% |
|
| 51 |
+
|Q5_K_M |26.8% |
|
| 52 |
+
|Q6_K |26.7% |
|
| 53 |
+
|Q8_K |26.5% |
|
| 54 |
+
|
| 55 |
Original model card below.
|
| 56 |
***
|
| 57 |
# DistilGPT2
|