Update README.md
Browse files
README.md
CHANGED
@@ -19,30 +19,30 @@ base_model: gordicaleksa/YugoGPT
|
|
19 |
|
20 |
# Quant. preference
|
21 |
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
|
|
19 |
|
20 |
# Quant. preference
|
21 |
|
22 |
+
| Key | Description |
|
23 |
+
|---------------|---------------------------------------------------------------------------------------|
|
24 |
+
| not_quantized | Recommended. Fast conversion. Slow inference, big files. |
|
25 |
+
| fast_quantized| Recommended. Fast conversion. OK inference, OK file size. |
|
26 |
+
| quantized | Recommended. Slow conversion. Fast inference, small files. |
|
27 |
+
| f32 | Not recommended. Retains 100% accuracy, but super slow and memory hungry. |
|
28 |
+
| f16 | Fastest conversion + retains 100% accuracy. Slow and memory hungry. |
|
29 |
+
| q8_0 | Fast conversion. High resource use, but generally acceptable. |
|
30 |
+
| q4_k_m | Recommended. Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q4_K |
|
31 |
+
| q5_k_m | Recommended. Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q5_K |
|
32 |
+
| q2_k | Uses Q4_K for the attention.vw and feed_forward.w2 tensors, Q2_K for the other tensors.|
|
33 |
+
| q3_k_l | Uses Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K |
|
34 |
+
| q3_k_m | Uses Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K |
|
35 |
+
| q3_k_s | Uses Q3_K for all tensors |
|
36 |
+
| q4_0 | Original quant method, 4-bit. |
|
37 |
+
| q4_1 | Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.|
|
38 |
+
| q4_k_s | Uses Q4_K for all tensors |
|
39 |
+
| q4_k | alias for q4_k_m |
|
40 |
+
| q5_k | alias for q5_k_m |
|
41 |
+
| q5_0 | Higher accuracy, higher resource usage and slower inference. |
|
42 |
+
| q5_1 | Even higher accuracy, resource usage and slower inference. |
|
43 |
+
| q5_k_s | Uses Q5_K for all tensors |
|
44 |
+
| q6_k | Uses Q8_K for all tensors |
|
45 |
+
| iq2_xxs | 2.06 bpw quantization |
|
46 |
+
| iq2_xs | 2.31 bpw quantization |
|
47 |
+
| iq3_xxs | 3.06 bpw quantization |
|
48 |
+
| q3_k_xs | 3-bit extra small quantization |
|