deploy section: pip install first, hf download for embedded targets
Browse files
README.md
CHANGED
|
@@ -176,7 +176,22 @@ Recognized keys are `date`, `locale`, `device`, `battery`, `network`, `location`
|
|
| 176 |
|
| 177 |
## Deploy Needle
|
| 178 |
|
| 179 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 180 |
|
| 181 |
| your device | folder | command-line | library |
|
| 182 |
| --- | --- | --- | --- |
|
|
|
|
| 176 |
|
| 177 |
## Deploy Needle
|
| 178 |
|
| 179 |
+
The recommended install on anything that runs Python is pip; it fetches the right build for your platform automatically on first use and caches it:
|
| 180 |
+
|
| 181 |
+
```sh
|
| 182 |
+
pip install cactus-needle
|
| 183 |
+
```
|
| 184 |
+
|
| 185 |
+
That covers the whole Quickstart above, plus the `needle` CLI: `needle playground` for a local UI, `needle finetune` and `needle build` for training and exporting your own.
|
| 186 |
+
|
| 187 |
+
For the standalone command-line binary or the embedded libraries (C, mobile, WebAssembly), pull your platform folder with the model config using the Hugging Face CLI instead of downloading files by hand:
|
| 188 |
+
|
| 189 |
+
```sh
|
| 190 |
+
pip install -U huggingface_hub
|
| 191 |
+
hf download Cactus-Compute/needle2 --include "macos-arm64/*" "config.json" --local-dir needle2
|
| 192 |
+
```
|
| 193 |
+
|
| 194 |
+
Platform folders:
|
| 195 |
|
| 196 |
| your device | folder | command-line | library |
|
| 197 |
| --- | --- | --- | --- |
|