Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -24,14 +24,15 @@ Built on Qwen/Qwen3.5-2B, this checkpoint returns decision distributions from te
|
|
| 24 |
|
| 25 |
## On-device exports (Apple Silicon)
|
| 26 |
|
| 27 |
-
Merged + quantized builds for macOS/iOS (text
|
| 28 |
-
is not ported to MLX):
|
| 29 |
|
| 30 |
| Repo | Size | JevBench | Target |
|
| 31 |
| --- | --- | --- | --- |
|
| 32 |
| [Linnaeus-0.1.0-2B-merged](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-merged) | 4.3 GB | 71.0% (MPS) | Mac dev / conversion source |
|
| 33 |
| [Linnaeus-0.1.0-2B-MLX-8bit](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-MLX-8bit) | 1.9 GB | 70.56% | Mac + iPhone |
|
| 34 |
| [Linnaeus-0.1.0-2B-MLX-4bit](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-MLX-4bit) | 1.0 GB | 67.53% | iPhone size-optimized |
|
|
|
|
|
|
|
| 35 |
|
| 36 |
The merged checkpoint embeds the decision head as an extra lm_head row
|
| 37 |
(score_row_id), so any stock LM runtime produces decision scores at
|
|
|
|
| 24 |
|
| 25 |
## On-device exports (Apple Silicon)
|
| 26 |
|
| 27 |
+
Merged + quantized builds for macOS/iOS (text MLX builds are text-only; the VLM builds keep the vision tower):
|
|
|
|
| 28 |
|
| 29 |
| Repo | Size | JevBench | Target |
|
| 30 |
| --- | --- | --- | --- |
|
| 31 |
| [Linnaeus-0.1.0-2B-merged](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-merged) | 4.3 GB | 71.0% (MPS) | Mac dev / conversion source |
|
| 32 |
| [Linnaeus-0.1.0-2B-MLX-8bit](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-MLX-8bit) | 1.9 GB | 70.56% | Mac + iPhone |
|
| 33 |
| [Linnaeus-0.1.0-2B-MLX-4bit](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-MLX-4bit) | 1.0 GB | 67.53% | iPhone size-optimized |
|
| 34 |
+
| [Linnaeus-0.1.0-2B-MLX-VLM-8bit](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-MLX-VLM-8bit) | 2.5 GB | text 70.56% + **images** | Mac + iPhone multimodal |
|
| 35 |
+
| [Linnaeus-0.1.0-2B-MLX-VLM-4bit](https://huggingface.co/pi-dal/Linnaeus-0.1.0-2B-MLX-VLM-4bit) | 1.6 GB | text ~67% + **images** | iPhone multimodal, size pick |
|
| 36 |
|
| 37 |
The merged checkpoint embeds the decision head as an extra lm_head row
|
| 38 |
(score_row_id), so any stock LM runtime produces decision scores at
|