Update README.md
Browse files
README.md
CHANGED
@@ -96,7 +96,7 @@ generation is bounded by memory speed, so smaller quants help, but they
|
|
96 |
cause the LLM to hallucinate more. However that doesn't mean they can't
|
97 |
think correctly. A highly degraded quant like `Q2_K` may not make a
|
98 |
great encyclopedia, but it's still capable of logical reasoning and
|
99 |
-
|
100 |
|
101 |
Good quants for reading (evaluation speed) are BF16, F16, Q4\_0, and
|
102 |
Q8\_0 (ordered from fastest to slowest). Prompt evaluation is bounded by
|
|
|
96 |
cause the LLM to hallucinate more. However that doesn't mean they can't
|
97 |
think correctly. A highly degraded quant like `Q2_K` may not make a
|
98 |
great encyclopedia, but it's still capable of logical reasoning and
|
99 |
+
the emergent capabilities LLMs exhibit.
|
100 |
|
101 |
Good quants for reading (evaluation speed) are BF16, F16, Q4\_0, and
|
102 |
Q8\_0 (ordered from fastest to slowest). Prompt evaluation is bounded by
|