Instructions to use anonymouscla/phyjudge-9B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use anonymouscla/phyjudge-9B with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-9B") model = PeftModel.from_pretrained(base_model, "anonymouscla/phyjudge-9B") - Notebooks
- Google Colab
- Kaggle
sync local anonymous/model/ contents
Browse files- subq+human.yaml +106 -0
subq+human.yaml
ADDED
|
@@ -0,0 +1,106 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
scheme: subq_hint
|
| 2 |
+
description: |-
|
| 3 |
+
JSON-only per-task prompts with observable sub-questions/checklists. The
|
| 4 |
+
subq+human setting uses sub-question prompts as input and human scores as
|
| 5 |
+
training targets.
|
| 6 |
+
sub_questions:
|
| 7 |
+
source: static
|
| 8 |
+
answer_format: hint
|
| 9 |
+
system_prompt: You are a strict video evaluation model.
|
| 10 |
+
general_keys:
|
| 11 |
+
- SA
|
| 12 |
+
- PTV
|
| 13 |
+
- persistence
|
| 14 |
+
eval_prompts:
|
| 15 |
+
SA: |-
|
| 16 |
+
Evaluate Prompt Alignment (SA).
|
| 17 |
+
|
| 18 |
+
Caption:
|
| 19 |
+
"{prompt}"
|
| 20 |
+
|
| 21 |
+
The video was generated using a text+image-to-video (ti2v) model, conditioned on the first frame and the text prompt above.
|
| 22 |
+
|
| 23 |
+
Sub-questions to consider in your mind before scoring:
|
| 24 |
+
{questions_block}
|
| 25 |
+
|
| 26 |
+
Score 1-5:
|
| 27 |
+
5=fully aligned
|
| 28 |
+
4=mostly aligned with minor deviations
|
| 29 |
+
3=partially aligned with notable gaps
|
| 30 |
+
2=mostly misaligned
|
| 31 |
+
1=not aligned
|
| 32 |
+
|
| 33 |
+
Then output ONLY a JSON object with exactly one key: SA.
|
| 34 |
+
|
| 35 |
+
Example:
|
| 36 |
+
{{"SA": 3}}
|
| 37 |
+
PTV: |-
|
| 38 |
+
Evaluate Temporal Coherence (PTV).
|
| 39 |
+
|
| 40 |
+
Caption:
|
| 41 |
+
"{prompt}"
|
| 42 |
+
|
| 43 |
+
The video was generated using a text+image-to-video (ti2v) model, conditioned on the first frame and the text prompt above.
|
| 44 |
+
|
| 45 |
+
Sub-questions to consider in your mind before scoring:
|
| 46 |
+
{questions_block}
|
| 47 |
+
|
| 48 |
+
Score 1-5:
|
| 49 |
+
5=fully plausible event order
|
| 50 |
+
4=mostly plausible with minor timing issues
|
| 51 |
+
3=partially plausible
|
| 52 |
+
2=mostly implausible
|
| 53 |
+
1=completely implausible order
|
| 54 |
+
|
| 55 |
+
Then output ONLY a JSON object with exactly one key: PTV.
|
| 56 |
+
|
| 57 |
+
Example:
|
| 58 |
+
{{"PTV": 4}}
|
| 59 |
+
persistence: |-
|
| 60 |
+
Evaluate Object Persistence.
|
| 61 |
+
|
| 62 |
+
Caption, for context only:
|
| 63 |
+
"{prompt}"
|
| 64 |
+
|
| 65 |
+
The video was generated using a text+image-to-video (ti2v) model, conditioned on the first frame and the text prompt above.
|
| 66 |
+
|
| 67 |
+
Sub-questions to consider in your mind before scoring:
|
| 68 |
+
{questions_block}
|
| 69 |
+
|
| 70 |
+
Score 1-5:
|
| 71 |
+
5=fully consistent
|
| 72 |
+
4=mostly consistent with minor flicker
|
| 73 |
+
3=noticeable issues
|
| 74 |
+
2=major inconsistencies
|
| 75 |
+
1=severe disappearance or identity changes
|
| 76 |
+
|
| 77 |
+
Then output ONLY a JSON object with exactly one key: persistence.
|
| 78 |
+
|
| 79 |
+
Example:
|
| 80 |
+
{{"persistence": 4}}
|
| 81 |
+
physical_sub_questions: true
|
| 82 |
+
physical_template: |-
|
| 83 |
+
Evaluate physical realism for one physical law: {law}.
|
| 84 |
+
|
| 85 |
+
Criterion:
|
| 86 |
+
{criteria}
|
| 87 |
+
|
| 88 |
+
Caption, for context only:
|
| 89 |
+
"{prompt}"
|
| 90 |
+
|
| 91 |
+
Sub-questions to consider in your mind before scoring:
|
| 92 |
+
{questions_block}
|
| 93 |
+
|
| 94 |
+
Judge the video itself. Do not penalize prompt mismatch unless it affects whether this physical law can be evaluated.
|
| 95 |
+
|
| 96 |
+
Score 1-5:
|
| 97 |
+
5=clearly correct
|
| 98 |
+
4=mostly correct with minor issues
|
| 99 |
+
3=partially correct or ambiguous
|
| 100 |
+
2=mostly incorrect
|
| 101 |
+
1=severely incorrect
|
| 102 |
+
|
| 103 |
+
Then output ONLY a JSON object with exactly one key: {law}.
|
| 104 |
+
|
| 105 |
+
Example:
|
| 106 |
+
{{"{law}": 3}}
|