pi-mtplx 0.1.1 → 0.1.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +22 -22
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -14,54 +14,54 @@ Restart Pi after installation.
|
|
|
14
14
|
|
|
15
15
|
Once installed, Pi automatically manages your MTPLX workflow:
|
|
16
16
|
|
|
17
|
-
- **Model discovery** —
|
|
18
|
-
- **Auto-start** — The MTPLX server starts
|
|
17
|
+
- **Model discovery** — A built-in model (`mtplx-qwen38-27b-optimized-quality`) is pre-registered. More are discoverable via `/mtplx`.
|
|
18
|
+
- **Auto-start** — The MTPLX server starts when you switch to an `mtplx` model and shuts down cleanly when Pi exits.
|
|
19
19
|
- **Token speed** — A `⚡N.N tk/s` indicator appears in the footer showing the generation speed of the last assistant turn.
|
|
20
20
|
|
|
21
21
|
## Commands
|
|
22
22
|
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
|
26
|
-
|
|
|
27
|
-
|
|
|
28
|
-
|
|
|
29
|
-
|
|
|
23
|
+
Run `/mtplx` to open an interactive menu:
|
|
24
|
+
|
|
25
|
+
| Option | What it does |
|
|
26
|
+
| -------- | ------------- |
|
|
27
|
+
| **Toggle (on/off)** | Start or stop the MTPLX server |
|
|
28
|
+
| **Fan Curves** | Set the thermal profile (`default`, `smart`, `max`) |
|
|
29
|
+
| **Models (register)** | Scan installed MTPLX models and register one with Pi |
|
|
30
|
+
| **Remove Model** | Unregister a model from Pi |
|
|
31
|
+
| **Uninstall** | Remove the `mtplx` provider from Pi's config |
|
|
30
32
|
|
|
31
33
|
## Configuration
|
|
32
34
|
|
|
33
35
|
### Model registry
|
|
34
36
|
|
|
35
|
-
Models are registered in `~/.pi/agent/mtplx-models.json`. Each entry maps a Pi model ID to an MTPLX artifact
|
|
37
|
+
Models are registered in `~/.pi/agent/mtplx-models.json`. Each entry maps a Pi model ID to an MTPLX artifact ref:
|
|
36
38
|
|
|
37
39
|
```json
|
|
38
40
|
{
|
|
39
|
-
"mtplx
|
|
41
|
+
"mtplx-qwen38-27b-optimized-quality": {
|
|
42
|
+
"ref": "Youssofal/Qwen3.8-27B-MTPLX-Optimized-Quality"
|
|
43
|
+
}
|
|
40
44
|
}
|
|
41
45
|
```
|
|
42
46
|
|
|
43
|
-
Register models
|
|
47
|
+
Register new models via the `/mtplx` → **Models** menu, or add entries manually to this file (then run `/reload`).
|
|
44
48
|
|
|
45
49
|
### Fan mode
|
|
46
50
|
|
|
47
|
-
Controls the thermal profile of your MTPLX server. Saved to `~/.pi/agent/mtplx-
|
|
51
|
+
Controls the thermal profile of your MTPLX server. Saved to `~/.pi/agent/mtplx-fanmode.json` and persists across restarts.
|
|
48
52
|
|
|
49
53
|
| Mode | Behavior |
|
|
50
54
|
| ------ | ---------- |
|
|
51
|
-
| `
|
|
52
|
-
| `
|
|
53
|
-
| `
|
|
54
|
-
|
|
55
|
-
### Provider name
|
|
56
|
-
|
|
57
|
-
The provider is named `mtplx` by default. Override it with `MTPLX_PROVIDER` in your environment.
|
|
55
|
+
| `default` | System default fan curve |
|
|
56
|
+
| `smart` | Default — adaptive thermal management |
|
|
57
|
+
| `max` | Maximum fan, fastest inference |
|
|
58
58
|
|
|
59
59
|
## Troubleshooting
|
|
60
60
|
|
|
61
61
|
| Problem | Fix |
|
|
62
62
|
| --------- | ----- |
|
|
63
|
-
| **"MTPLX not started"** — Pi warns when you ask an MTPLX model to respond |
|
|
64
|
-
| **"No MTPLX models registered"** | Run `/mtplx
|
|
63
|
+
| **"MTPLX not started"** — Pi warns when you ask an MTPLX model to respond | Run `/mtplx` → **Toggle** to start the server |
|
|
64
|
+
| **"No MTPLX models registered"** | Run `/mtplx` → **Models** to discover and register one |
|
|
65
65
|
| **MTPLX startup timed out after 180s** | Run `mtplx status --deep` for MTPLX-side diagnostics (model validation, memory, thermal) |
|
|
66
66
|
|
|
67
67
|
## License
|