Remove the evaluation section from the model card
Browse files
README.md
CHANGED
|
@@ -103,16 +103,6 @@ The generation config of Qwen3 samples by default; pass `do_sample=False` for re
|
|
| 103 |
Stage 3 uses the progressive TopK schedule (0.9 to 0.1 during the first 60% of the training steps).
|
| 104 |
Recipe: `scripts/train/video_384` of the repository.
|
| 105 |
|
| 106 |
-
## Evaluation
|
| 107 |
-
|
| 108 |
-
Measured with the evaluation code of the repository (`lmms_eval_integration/`, lmms-eval).
|
| 109 |
-
|
| 110 |
-
Video benchmarks, 64 uniformly sampled frames (greedy decoding):
|
| 111 |
-
|
| 112 |
-
| Video-MME | MLVU | EgoSchema (subset) | LongVideoBench | NExT-QA | TempCompass |
|
| 113 |
-
|---:|---:|---:|---:|---:|---:|
|
| 114 |
-
| 60.0 | 67.4 | 53.8 | 56.7 | 80.2 | 63.9 |
|
| 115 |
-
|
| 116 |
## License
|
| 117 |
|
| 118 |
Apache License 2.0. The model is built on [Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)
|
|
|
|
| 103 |
Stage 3 uses the progressive TopK schedule (0.9 to 0.1 during the first 60% of the training steps).
|
| 104 |
Recipe: `scripts/train/video_384` of the repository.
|
| 105 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
## License
|
| 107 |
|
| 108 |
Apache License 2.0. The model is built on [Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)
|