Update README.md
Browse filesMD formatting updated
README.md
CHANGED
@@ -10,18 +10,41 @@ tags:
|
|
10 |
---
|
11 |
# ANEMLL
|
12 |
|
13 |
-
|
14 |
-
|
15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
python prefill.py --meta meta.yaml
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
-
|
19 |
-
M1 Ultra, M2 Ultra and M4 Max
|
20 |
-
It is single chunk for Batch=64/Window=64
|
21 |
-
See https://docs.google.com/spreadsheets/d/1OCxn730D5h8rvS2IHsSi0UBYbsP_lV-W-0uVdVDCvIk
|
22 |
-
FP16 tab for baseline numbers
|
23 |
|
24 |
-
For M3U/M4P see original post:
|
25 |
https://x.com/anemll/status/1919796143787278623
|
26 |
|
27 |
|
|
|
10 |
---
|
11 |
# ANEMLL
|
12 |
|
13 |
+
Here’s your text formatted in Markdown (MD):
|
14 |
+
|
15 |
+
# PREFILL Test for M3 Ultra
|
16 |
+
|
17 |
+
After unzipping :
|
18 |
+
|
19 |
+
```bash
|
20 |
+
find . -type f -name "*.zip" -exec unzip {} \;
|
21 |
+
````
|
22 |
+
|
23 |
+
Run:
|
24 |
+
```bash
|
25 |
python prefill.py --meta meta.yaml
|
26 |
+
````
|
27 |
+
|
28 |
+
|
29 |
+
⸻
|
30 |
+
|
31 |
+
The repository contains an extra file:
|
32 |
+
|
33 |
+
nemotron_prefill_chunk_01of16_64x64.mlpackage
|
34 |
+
|
35 |
+
This will be interesting to profile with Xcode on:
|
36 |
+
• M1 Ultra
|
37 |
+
• M2 Ultra
|
38 |
+
• M4 Max
|
39 |
+
|
40 |
+
It represents a single chunk for Batch=64 / Window=64.
|
41 |
+
|
42 |
+
If you have results, please email them to: [[email protected]](mailto:[email protected])
|
43 |
+
|
44 |
+
See the FP16 tab for baseline numbers.
|
45 |
|
46 |
+
For M3U / M4P reference, see the original post on X.
|
|
|
|
|
|
|
|
|
47 |
|
|
|
48 |
https://x.com/anemll/status/1919796143787278623
|
49 |
|
50 |
|