Update README.md
Browse files
README.md
CHANGED
|
@@ -17,12 +17,59 @@ tags:
|
|
| 17 |
- instruction-tuning
|
| 18 |
|
| 19 |
model-index:
|
| 20 |
-
- name: sft_full_final
|
| 21 |
-
results: []
|
| 22 |
-
- name: reward_model_final
|
| 23 |
-
results: []
|
| 24 |
- name: ppo_aligned_final
|
| 25 |
-
results:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
---
|
| 27 |
|
| 28 |
# RLHF-Aligned GPT-2 Pipeline Models
|
|
|
|
| 17 |
- instruction-tuning
|
| 18 |
|
| 19 |
model-index:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
- name: ppo_aligned_final
|
| 21 |
+
results:
|
| 22 |
+
- task:
|
| 23 |
+
type: text-generation
|
| 24 |
+
dataset:
|
| 25 |
+
type: Dahoas/synthetic-instruct-gptj-pairwise
|
| 26 |
+
name: Dahoas/synthetic-instruct-gptj-pairwise
|
| 27 |
+
split: evaluation
|
| 28 |
+
metrics:
|
| 29 |
+
- type: average_reward
|
| 30 |
+
value: 2.37
|
| 31 |
+
name: Average Reward Score
|
| 32 |
+
- type: rouge
|
| 33 |
+
value: 0.337
|
| 34 |
+
name: ROUGE-1
|
| 35 |
+
- type: rouge
|
| 36 |
+
value: 0.139
|
| 37 |
+
name: ROUGE-2
|
| 38 |
+
- type: rouge
|
| 39 |
+
value: 0.252
|
| 40 |
+
name: ROUGE-L
|
| 41 |
+
|
| 42 |
+
- name: reward_model_final
|
| 43 |
+
results:
|
| 44 |
+
- task:
|
| 45 |
+
type: text-classification
|
| 46 |
+
dataset:
|
| 47 |
+
type: Dahoas/synthetic-instruct-gptj-pairwise
|
| 48 |
+
name: Dahoas/synthetic-instruct-gptj-pairwise
|
| 49 |
+
split: evaluation
|
| 50 |
+
metrics:
|
| 51 |
+
- type: accuracy
|
| 52 |
+
value: 0.98
|
| 53 |
+
name: Preference Accuracy
|
| 54 |
+
|
| 55 |
+
- name: sft_full_final
|
| 56 |
+
results:
|
| 57 |
+
- task:
|
| 58 |
+
type: text-generation
|
| 59 |
+
dataset:
|
| 60 |
+
type: Dahoas/synthetic-instruct-gptj-pairwise
|
| 61 |
+
name: Dahoas/synthetic-instruct-gptj-pairwise
|
| 62 |
+
split: evaluation
|
| 63 |
+
metrics:
|
| 64 |
+
- type: rouge
|
| 65 |
+
value: 0.353
|
| 66 |
+
name: ROUGE-1
|
| 67 |
+
- type: rouge
|
| 68 |
+
value: 0.149
|
| 69 |
+
name: ROUGE-2
|
| 70 |
+
- type: rouge
|
| 71 |
+
value: 0.262
|
| 72 |
+
name: ROUGE-L
|
| 73 |
---
|
| 74 |
|
| 75 |
# RLHF-Aligned GPT-2 Pipeline Models
|