Results section first
Browse files
README.md
CHANGED
|
@@ -55,6 +55,13 @@ By **Adrian Ciprian Iancu** 路 [GitHub](https://github.com/ArtyomITA) 路 [Huggin
|
|
| 55 |
> [prism-ml/Ternary-Bonsai-2-9B-gguf](https://huggingface.co/prism-ml/Ternary-Bonsai-2-9B-gguf), loaded at scale 0 and
|
| 56 |
> switched on per request, so normal chat stays exactly the base model.
|
| 57 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
## Pick a file
|
| 59 |
|
| 60 |
| file | for | size | what it is |
|
|
@@ -63,13 +70,6 @@ By **Adrian Ciprian Iancu** 路 [GitHub](https://github.com/ArtyomITA) 路 [Huggin
|
|
| 63 |
| `eval/` | checking my numbers | small | summaries and every answer of the base model and v2 on each benchmark |
|
| 64 |
| `training/` | rebuilding it | small | data builder, trainer (Blackwell dense mode), benchmark runner |
|
| 65 |
|
| 66 |
-
## At a glance
|
| 67 |
-
|
| 68 |
-
Every case is asked to the base model (adapter at scale 0) and to v2 on the same llama.cpp server. Greedy decoding,
|
| 69 |
-
images at 1024 tokens, `Ternary-Bonsai-2-9B-PQ2_0-MTP-Q8_0.gguf` with the Q8_0 vision projector.
|
| 70 |
-
|
| 71 |
-
**Benchmarks are running now** (full ScreenSpot-v2, full ScreenSpot-Pro, GUI-360 test grounding and actions, base model against v2 on the same server). This section is filled in as soon as they finish.
|
| 72 |
-
|
| 73 |
## Results
|
| 74 |
|
| 75 |
**Benchmarks are running now** (full ScreenSpot-v2, full ScreenSpot-Pro, GUI-360 test grounding and actions, base model against v2 on the same server). This section is filled in as soon as they finish.
|
|
|
|
| 55 |
> [prism-ml/Ternary-Bonsai-2-9B-gguf](https://huggingface.co/prism-ml/Ternary-Bonsai-2-9B-gguf), loaded at scale 0 and
|
| 56 |
> switched on per request, so normal chat stays exactly the base model.
|
| 57 |
|
| 58 |
+
## Results at a glance
|
| 59 |
+
|
| 60 |
+
Every case is asked to the base model (adapter at scale 0) and to v2 on the same llama.cpp server. Greedy decoding,
|
| 61 |
+
images at 1024 tokens, `Ternary-Bonsai-2-9B-PQ2_0-MTP-Q8_0.gguf` with the Q8_0 vision projector.
|
| 62 |
+
|
| 63 |
+
**Benchmarks are running now** (full ScreenSpot-v2, full ScreenSpot-Pro, GUI-360 test grounding and actions, base model against v2 on the same server). This section is filled in as soon as they finish.
|
| 64 |
+
|
| 65 |
## Pick a file
|
| 66 |
|
| 67 |
| file | for | size | what it is |
|
|
|
|
| 70 |
| `eval/` | checking my numbers | small | summaries and every answer of the base model and v2 on each benchmark |
|
| 71 |
| `training/` | rebuilding it | small | data builder, trainer (Blackwell dense mode), benchmark runner |
|
| 72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
## Results
|
| 74 |
|
| 75 |
**Benchmarks are running now** (full ScreenSpot-v2, full ScreenSpot-Pro, GUI-360 test grounding and actions, base model against v2 on the same server). This section is filled in as soon as they finish.
|